@uipath/delegate-sdk 1.203.1 → 1.203.3
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +41 -3
- package/dist/index.mjs +2 -2
- package/dist/skills/business-analysis/SKILL.md +729 -0
- package/dist/skills/business-analysis/as-is-capture.md +444 -0
- package/dist/skills/business-analysis/assets/progress.yaml +19 -0
- package/dist/skills/business-analysis/assets/task-list.md +34 -0
- package/dist/skills/business-analysis/automation-hub-publishing.md +47 -0
- package/dist/skills/business-analysis/bpmn-to-be-design.md +274 -0
- package/dist/skills/business-analysis/case-to-be-design.md +223 -0
- package/dist/skills/business-analysis/diagram-authoring.md +224 -0
- package/dist/skills/business-analysis/industry-gaps.md +134 -0
- package/dist/skills/business-analysis/interview-spine.md +297 -0
- package/dist/skills/business-analysis/locales/de.json +54 -0
- package/dist/skills/business-analysis/locales/en.json +54 -0
- package/dist/skills/business-analysis/locales/es-MX.json +54 -0
- package/dist/skills/business-analysis/locales/es.json +54 -0
- package/dist/skills/business-analysis/locales/fr.json +54 -0
- package/dist/skills/business-analysis/locales/ja.json +54 -0
- package/dist/skills/business-analysis/locales/ko.json +54 -0
- package/dist/skills/business-analysis/locales/pt-BR.json +54 -0
- package/dist/skills/business-analysis/locales/pt.json +54 -0
- package/dist/skills/business-analysis/locales/ro.json +54 -0
- package/dist/skills/business-analysis/locales/tr.json +54 -0
- package/dist/skills/business-analysis/locales/zh-CN.json +54 -0
- package/dist/skills/business-analysis/locales/zh-TW.json +54 -0
- package/dist/skills/business-analysis/pdd-generation.md +204 -0
- package/dist/skills/business-analysis/pdd-template.md +562 -0
- package/dist/skills/business-analysis/progress-presentation.json +13 -0
- package/dist/skills/business-analysis/recording-first.md +19 -0
- package/dist/skills/business-analysis/recording-processing.md +152 -0
- package/dist/skills/business-analysis/references/automation-hub-open-api.md +57 -0
- package/dist/skills/business-analysis/references/case-definition-spec.md +90 -0
- package/dist/skills/business-analysis/references/case-design-patterns/loan-origination-spec.md +86 -0
- package/dist/skills/business-analysis/references/component-type-mapping.md +52 -0
- package/dist/skills/business-analysis/references/uipath-products/case-management-schema.md +62 -0
- package/dist/skills/business-analysis/references/uipath-products/maestro-case-framework.md +65 -0
- package/dist/skills/business-analysis/references/uipath-products/maestro-case-management.md +49 -0
- package/dist/skills/business-analysis/research-offer.md +104 -0
- package/dist/skills/business-analysis/research-pass.md +149 -0
- package/dist/skills/business-analysis/scripts/assemble-pdd.ps1 +294 -0
- package/dist/skills/business-analysis/scripts/assemble-pdd.sh +453 -0
- package/dist/skills/business-analysis/scripts/check-phase-ready.ps1 +72 -0
- package/dist/skills/business-analysis/scripts/check-phase-ready.sh +98 -0
- package/dist/skills/business-analysis/scripts/lint-diagram.ps1 +524 -0
- package/dist/skills/business-analysis/scripts/lint-diagram.sh +519 -0
- package/dist/skills/business-analysis/scripts/stamp-pages.ps1 +363 -0
- package/dist/skills/business-analysis/scripts/stamp-pages.sh +360 -0
- package/dist/skills/business-analysis/to-be-design.md +199 -0
- package/dist/skills/confluence/SKILL.md +386 -0
- package/dist/skills/doc-coauthoring/SKILL.md +295 -0
- package/dist/skills/excel-architect/SKILL.macos.md +161 -0
- package/dist/skills/excel-architect/SKILL.md +180 -0
- package/dist/skills/excel-architect/SKILL.windows.md +61 -0
- package/dist/skills/excel-architect/archetypes/01-financial-model-advanced.md +332 -0
- package/dist/skills/excel-architect/archetypes/01-financial-model.md +248 -0
- package/dist/skills/excel-architect/archetypes/02-data-analysis.md +171 -0
- package/dist/skills/excel-architect/archetypes/03-tracking-dashboard.md +160 -0
- package/dist/skills/excel-architect/archetypes/04-project-planning.md +193 -0
- package/dist/skills/excel-architect/archetypes/05-simple-table.md +111 -0
- package/dist/skills/excel-architect/building.md +279 -0
- package/dist/skills/excel-architect/create.md +64 -0
- package/dist/skills/excel-architect/data-and-formulas.md +186 -0
- package/dist/skills/excel-architect/generator.md +272 -0
- package/dist/skills/excel-architect/macos/scripts/add-data-validation.applescript +242 -0
- package/dist/skills/excel-architect/macos/scripts/append-rows.applescript +310 -0
- package/dist/skills/excel-architect/macos/scripts/apply-conditional-format.applescript +565 -0
- package/dist/skills/excel-architect/macos/scripts/autofit-columns.applescript +79 -0
- package/dist/skills/excel-architect/macos/scripts/benchmark.sh +135 -0
- package/dist/skills/excel-architect/macos/scripts/check-compile.sh +30 -0
- package/dist/skills/excel-architect/macos/scripts/clear-range.applescript +152 -0
- package/dist/skills/excel-architect/macos/scripts/copy-paste.applescript +179 -0
- package/dist/skills/excel-architect/macos/scripts/create-chart.applescript +531 -0
- package/dist/skills/excel-architect/macos/scripts/create-pivot.applescript +443 -0
- package/dist/skills/excel-architect/macos/scripts/create-sheets.applescript +264 -0
- package/dist/skills/excel-architect/macos/scripts/create-workbook.applescript +92 -0
- package/dist/skills/excel-architect/macos/scripts/delete-chart.applescript +105 -0
- package/dist/skills/excel-architect/macos/scripts/delete-sheet.applescript +127 -0
- package/dist/skills/excel-architect/macos/scripts/filter-data.applescript +213 -0
- package/dist/skills/excel-architect/macos/scripts/find-replace.applescript +238 -0
- package/dist/skills/excel-architect/macos/scripts/format-range.applescript +531 -0
- package/dist/skills/excel-architect/macos/scripts/freeze-panes.applescript +133 -0
- package/dist/skills/excel-architect/macos/scripts/get-current-file.applescript +24 -0
- package/dist/skills/excel-architect/macos/scripts/get-sheet-dimensions.applescript +115 -0
- package/dist/skills/excel-architect/macos/scripts/health-check.applescript +704 -0
- package/dist/skills/excel-architect/macos/scripts/list-open-workbooks.applescript +80 -0
- package/dist/skills/excel-architect/macos/scripts/maximize-window.applescript +47 -0
- package/dist/skills/excel-architect/macos/scripts/merge-cells.applescript +227 -0
- package/dist/skills/excel-architect/macos/scripts/open-workbook.applescript +139 -0
- package/dist/skills/excel-architect/macos/scripts/read-range.applescript +267 -0
- package/dist/skills/excel-architect/macos/scripts/rename-sheet.applescript +231 -0
- package/dist/skills/excel-architect/macos/scripts/select-sheet.applescript +133 -0
- package/dist/skills/excel-architect/macos/scripts/set-tab-color.applescript +109 -0
- package/dist/skills/excel-architect/macos/scripts/sort-data.applescript +136 -0
- package/dist/skills/excel-architect/macos/scripts/undo.applescript +59 -0
- package/dist/skills/excel-architect/macos/scripts/workbook-lifecycle.applescript +149 -0
- package/dist/skills/excel-architect/macos/scripts/write-formula.applescript +225 -0
- package/dist/skills/excel-architect/macos/scripts/write-range.applescript +351 -0
- package/dist/skills/excel-architect/reference/api/advanced.md +381 -0
- package/dist/skills/excel-architect/reference/api/basic.md +747 -0
- package/dist/skills/excel-architect/reference/guides.md +348 -0
- package/dist/skills/excel-architect/reference/repair.md +174 -0
- package/dist/skills/excel-architect/reference/troubleshooting.md +46 -0
- package/dist/skills/excel-architect/scripting.md +266 -0
- package/dist/skills/excel-architect/styling-and-features.md +433 -0
- package/dist/skills/excel-architect/windows/api/composites.md +337 -0
- package/dist/skills/excel-architect/windows/api/scripting.md +139 -0
- package/dist/skills/excel-architect/windows/examples/level-1-workflows.md +485 -0
- package/dist/skills/excel-architect/windows/examples/level-2-advanced.md +690 -0
- package/dist/skills/excel-architect/windows/examples/level-3-scripting.md +337 -0
- package/dist/skills/github/SKILL.md +446 -0
- package/dist/skills/github/recipes/add-comment.md +85 -0
- package/dist/skills/github/recipes/check-pr-status.md +122 -0
- package/dist/skills/github/recipes/compare-branches.md +79 -0
- package/dist/skills/github/recipes/create-issue.md +103 -0
- package/dist/skills/github/recipes/create-pr.md +97 -0
- package/dist/skills/github/recipes/get-file-contents.md +76 -0
- package/dist/skills/github/recipes/get-last-commit.md +128 -0
- package/dist/skills/github/recipes/list-my-open-prs.md +105 -0
- package/dist/skills/github/recipes/list-recent-commits.md +84 -0
- package/dist/skills/github/recipes/search-issues.md +115 -0
- package/dist/skills/google-workspace/SKILL.md +217 -0
- package/dist/skills/google-workspace/services/calendar.md +311 -0
- package/dist/skills/google-workspace/services/docs.md +426 -0
- package/dist/skills/google-workspace/services/drive.md +504 -0
- package/dist/skills/google-workspace/services/gmail.md +466 -0
- package/dist/skills/google-workspace/services/sheets.md +370 -0
- package/dist/skills/internal-comms/SKILL.md +35 -0
- package/dist/skills/internal-comms/examples/3p-updates.md +47 -0
- package/dist/skills/internal-comms/examples/company-newsletter.md +65 -0
- package/dist/skills/internal-comms/examples/faq-answers.md +30 -0
- package/dist/skills/internal-comms/examples/general-comms.md +15 -0
- package/dist/skills/jira/SKILL.md +397 -0
- package/dist/skills/jira/recipes/add-comment.md +123 -0
- package/dist/skills/jira/recipes/assign-issue.md +157 -0
- package/dist/skills/jira/recipes/create-issue.md +274 -0
- package/dist/skills/jira/recipes/get-issue-details.md +154 -0
- package/dist/skills/jira/recipes/list-projects.md +151 -0
- package/dist/skills/jira/recipes/my-issues.md +180 -0
- package/dist/skills/jira/recipes/search-issues.md +215 -0
- package/dist/skills/jira/recipes/transition-issue.md +154 -0
- package/dist/skills/jira/recipes/troubleshooting.md +387 -0
- package/dist/skills/knowledge-base/SKILL.md +289 -0
- package/dist/skills/media-tools/SKILL.md +204 -0
- package/dist/skills/media-tools/references/encode-runbook.md +24 -0
- package/dist/skills/media-tools/references/resolve-binaries.md +88 -0
- package/dist/skills/media-tools/scripts/macos/ff-launch.sh +22 -0
- package/dist/skills/media-tools/scripts/macos/ff-poll.sh +10 -0
- package/dist/skills/media-tools/scripts/windows/ff-launch.ps1 +34 -0
- package/dist/skills/media-tools/scripts/windows/ff-poll.ps1 +20 -0
- package/dist/skills/office-download/SKILL.md +161 -0
- package/dist/skills/outlook-365/SKILL.macos.md +331 -0
- package/dist/skills/outlook-365/SKILL.md +171 -0
- package/dist/skills/outlook-365/SKILL.windows.md +124 -0
- package/dist/skills/outlook-365/api/bulk-email-operations.md +178 -0
- package/dist/skills/outlook-365/api/check-availability.md +87 -0
- package/dist/skills/outlook-365/api/create-inbox-rule.md +182 -0
- package/dist/skills/outlook-365/api/download-attachments.md +149 -0
- package/dist/skills/outlook-365/api/find-and-email-contact.md +185 -0
- package/dist/skills/outlook-365/api/manage-calendar-events.md +189 -0
- package/dist/skills/outlook-365/api/organize-with-folders.md +178 -0
- package/dist/skills/outlook-365/api/reply-to-conversation.md +68 -0
- package/dist/skills/outlook-365/api/schedule-meeting.md +73 -0
- package/dist/skills/outlook-365/api/search-and-read-email.md +93 -0
- package/dist/skills/outlook-365/api/send-email-with-attachments.md +94 -0
- package/dist/skills/outlook-365/api/send-email.md +396 -0
- package/dist/skills/outlook-365/api/set-out-of-office.md +196 -0
- package/dist/skills/outlook-365/macos/scripts/bulk-operations.applescript +429 -0
- package/dist/skills/outlook-365/macos/scripts/calendar.applescript +642 -0
- package/dist/skills/outlook-365/macos/scripts/contacts.applescript +635 -0
- package/dist/skills/outlook-365/macos/scripts/helpers.applescript +227 -0
- package/dist/skills/outlook-365/macos/scripts/read-emails.applescript +299 -0
- package/dist/skills/outlook-365/macos/scripts/search-emails.applescript +339 -0
- package/dist/skills/outlook-365/macos/scripts/send-email.applescript +245 -0
- package/dist/skills/outlook-365/uia/BASIC_OPERATIONS.md +278 -0
- package/dist/skills/outlook-365/uia/CALENDAR_OPERATIONS.md +386 -0
- package/dist/skills/outlook-365/uia/COMPOSE_EMAIL.md +426 -0
- package/dist/skills/outlook-365/uia/README.md +75 -0
- package/dist/skills/outlook-365/windows/com/bulk-operations.ps1 +241 -0
- package/dist/skills/outlook-365/windows/com/calendar.ps1 +227 -0
- package/dist/skills/outlook-365/windows/com/contacts.ps1 +270 -0
- package/dist/skills/outlook-365/windows/com/helpers.ps1 +158 -0
- package/dist/skills/outlook-365/windows/com/read-emails.ps1 +115 -0
- package/dist/skills/outlook-365/windows/com/search-emails.ps1 +141 -0
- package/dist/skills/outlook-365/windows/com/send-email.ps1 +121 -0
- package/dist/skills/pdf-architect/SKILL.md +640 -0
- package/dist/skills/powerpoint-architect/SKILL.macos.md +271 -0
- package/dist/skills/powerpoint-architect/SKILL.md +853 -0
- package/dist/skills/powerpoint-architect/SKILL.windows.md +458 -0
- package/dist/skills/powerpoint-architect/custom_themes.md +141 -0
- package/dist/skills/powerpoint-architect/editing-guide.md +424 -0
- package/dist/skills/powerpoint-architect/generator-reference.md +668 -0
- package/dist/skills/powerpoint-architect/images-guide.md +207 -0
- package/dist/skills/powerpoint-architect/issue-fix-mapping.md +768 -0
- package/dist/skills/powerpoint-architect/macos/scripts/add-html-to-slide.applescript +389 -0
- package/dist/skills/powerpoint-architect/macos/scripts/add-slide.applescript +255 -0
- package/dist/skills/powerpoint-architect/macos/scripts/add-speaker-notes.applescript +132 -0
- package/dist/skills/powerpoint-architect/macos/scripts/add-text.applescript +245 -0
- package/dist/skills/powerpoint-architect/macos/scripts/apply-slide-layout.applescript +80 -0
- package/dist/skills/powerpoint-architect/macos/scripts/apply-theme.applescript +122 -0
- package/dist/skills/powerpoint-architect/macos/scripts/autofit-placeholders.applescript +62 -0
- package/dist/skills/powerpoint-architect/macos/scripts/close-presentation.applescript +24 -0
- package/dist/skills/powerpoint-architect/macos/scripts/copy-slides.applescript +303 -0
- package/dist/skills/powerpoint-architect/macos/scripts/create-chart-with-data.applescript +305 -0
- package/dist/skills/powerpoint-architect/macos/scripts/create-with-theme.applescript +56 -0
- package/dist/skills/powerpoint-architect/macos/scripts/delete-slides.applescript +137 -0
- package/dist/skills/powerpoint-architect/macos/scripts/export-presentation.applescript +100 -0
- package/dist/skills/powerpoint-architect/macos/scripts/extract-slide-data.applescript +138 -0
- package/dist/skills/powerpoint-architect/macos/scripts/format-shape.applescript +192 -0
- package/dist/skills/powerpoint-architect/macos/scripts/format-slide-content.applescript +336 -0
- package/dist/skills/powerpoint-architect/macos/scripts/get-current-slide-number.applescript +38 -0
- package/dist/skills/powerpoint-architect/macos/scripts/get-file-path.applescript +94 -0
- package/dist/skills/powerpoint-architect/macos/scripts/get-layout-elements.applescript +141 -0
- package/dist/skills/powerpoint-architect/macos/scripts/get-layouts.applescript +58 -0
- package/dist/skills/powerpoint-architect/macos/scripts/get-opened-files.applescript +50 -0
- package/dist/skills/powerpoint-architect/macos/scripts/get-presentation-info.applescript +150 -0
- package/dist/skills/powerpoint-architect/macos/scripts/get-recent-files.applescript +68 -0
- package/dist/skills/powerpoint-architect/macos/scripts/get-slide-count.applescript +37 -0
- package/dist/skills/powerpoint-architect/macos/scripts/get-slide-info.applescript +105 -0
- package/dist/skills/powerpoint-architect/macos/scripts/health-check.applescript +690 -0
- package/dist/skills/powerpoint-architect/macos/scripts/insert-shape.applescript +188 -0
- package/dist/skills/powerpoint-architect/macos/scripts/insert-table.applescript +112 -0
- package/dist/skills/powerpoint-architect/macos/scripts/insert-text-box.applescript +225 -0
- package/dist/skills/powerpoint-architect/macos/scripts/manage-image.applescript +486 -0
- package/dist/skills/powerpoint-architect/macos/scripts/maximize-window.applescript +48 -0
- package/dist/skills/powerpoint-architect/macos/scripts/navigate-to-slide.applescript +84 -0
- package/dist/skills/powerpoint-architect/macos/scripts/open-presentation.applescript +38 -0
- package/dist/skills/powerpoint-architect/macos/scripts/reorder-slides.applescript +106 -0
- package/dist/skills/powerpoint-architect/macos/scripts/replace-text.applescript +193 -0
- package/dist/skills/powerpoint-architect/macos/scripts/save-presentation.applescript +54 -0
- package/dist/skills/powerpoint-architect/macos/scripts/set-animation.applescript +203 -0
- package/dist/skills/powerpoint-architect/macos/scripts/set-transition.applescript +130 -0
- package/dist/skills/powerpoint-architect/visual-styling-guide.md +206 -0
- package/dist/skills/powerpoint-architect/windows/api/advanced.md +317 -0
- package/dist/skills/powerpoint-architect/windows/api/basic.md +392 -0
- package/dist/skills/powerpoint-architect/windows/examples/10-slide-presentation.md +140 -0
- package/dist/skills/powerpoint-architect/windows/examples/custom-theme-workflow.md +219 -0
- package/dist/skills/powerpoint-architect/windows/examples/level-1-workflows.md +531 -0
- package/dist/skills/powerpoint-architect/windows/examples/level-2-scripting.md +239 -0
- package/dist/skills/powerpoint-architect/windows/scripts/create-with-theme.ps1 +81 -0
- package/dist/skills/project-task-list/SKILL.md +174 -0
- package/dist/skills/slack-operation/SKILL.md +97 -0
- package/dist/skills/slack-operation/api/advanced-api/GUIDE.md +277 -0
- package/dist/skills/slack-operation/api/search-and-read/EXAMPLE.md +75 -0
- package/dist/skills/slack-operation/api/search-and-read/GUIDE.md +108 -0
- package/dist/skills/slack-operation/api/send-message/EXAMPLE.md +112 -0
- package/dist/skills/slack-operation/api/send-message/GUIDE.md +63 -0
- package/dist/skills/slack-operation/api/share-files/EXAMPLE.md +94 -0
- package/dist/skills/slack-operation/api/share-files/GUIDE.md +63 -0
- package/dist/skills/slack-operation/ui/desktop-automation/GUIDE.md +55 -0
- package/dist/skills/system-diagnostics/SKILL.md +263 -0
- package/dist/skills/teams-operation/SKILL.md +79 -0
- package/dist/skills/teams-operation/api/advanced-api/GUIDE.md +234 -0
- package/dist/skills/teams-operation/api/meetings/EXAMPLE.md +63 -0
- package/dist/skills/teams-operation/api/meetings/GUIDE.md +52 -0
- package/dist/skills/teams-operation/api/search-and-read/EXAMPLE.md +72 -0
- package/dist/skills/teams-operation/api/search-and-read/GUIDE.md +54 -0
- package/dist/skills/teams-operation/api/send-message/EXAMPLE.md +92 -0
- package/dist/skills/teams-operation/api/send-message/GUIDE.md +60 -0
- package/dist/skills/teams-operation/ui/desktop-automation/GUIDE.md +66 -0
- package/dist/skills/transcribe-media/SKILL.md +142 -0
- package/dist/skills/uipath-integrations/SKILL.md +370 -0
- package/dist/skills/uipath-integrations/reference/connectors.md +331 -0
- package/dist/skills/uipath-integrations/reference/endpoints.md +220 -0
- package/dist/skills/uipath-integrations/reference/errors.md +211 -0
- package/dist/skills/uipath-integrations/reference/metadata.md +317 -0
- package/dist/skills/uipath-integrations/reference/operations.md +297 -0
- package/dist/skills/uipath-orchestrator/SKILL.md +312 -0
- package/dist/skills/uipath-orchestrator/recipes/assets.md +282 -0
- package/dist/skills/uipath-orchestrator/recipes/audit-logs.md +215 -0
- package/dist/skills/uipath-orchestrator/recipes/buckets.md +205 -0
- package/dist/skills/uipath-orchestrator/recipes/jobs.md +286 -0
- package/dist/skills/uipath-orchestrator/recipes/list-folders.md +207 -0
- package/dist/skills/uipath-orchestrator/recipes/queue-items.md +292 -0
- package/dist/skills/uipath-orchestrator/recipes/queues.md +232 -0
- package/dist/skills/uipath-orchestrator/recipes/releases.md +212 -0
- package/dist/skills/uipath-orchestrator/recipes/robots-sessions.md +274 -0
- package/dist/skills/uipath-orchestrator/recipes/schedules.md +280 -0
- package/dist/skills/uipath-orchestrator/reference/endpoints.md +85 -0
- package/dist/skills/uipath-orchestrator/reference/schemas.md +153 -0
- package/dist/skills/uipath-taskautomation/SKILL.md +310 -0
- package/dist/skills/uipath-taskautomation/assets/CodedWorkflow.template.cs +98 -0
- package/dist/skills/uipath-taskautomation/assets/ExampleWorkflow.cs +146 -0
- package/dist/skills/uipath-taskautomation/references/action-activity-map.json +32 -0
- package/dist/skills/uipath-taskautomation/references/action-activity-map.md +44 -0
- package/dist/skills/uipath-taskautomation/references/app-profiles.md +63 -0
- package/dist/skills/uipath-taskautomation/references/automation-spec.md +88 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/assets/before-after-hooks-template.md +115 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/assets/codedworkflow-template.md +186 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/assets/helper-utility-template.md +22 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/assets/json-template.md +91 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/assets/testcase-template.md +92 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/cli-reference.md +249 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/coded/codedworkflow-reference.md +275 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/coded/coding-guidelines.md +229 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/coded/inspect-package-guide.md +81 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/coded/integration-service-guide.md +507 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/coded/operations-guide.md +410 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/coded/third-party-packages-guide.md +67 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/data-manipulation-guide.md +110 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/environment-setup.md +194 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/error-handling-guide.md +195 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/powershell-interop-guide.md +57 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/project-structure-guide.md +114 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/project-structure.md +126 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/uia-elements-interaction-guide.md +109 -0
- package/dist/skills/uipath-taskautomation/references/coded-reference/validation-guide.md +147 -0
- package/dist/skills/uipath-taskautomation/references/coded-workflow-authoring.md +219 -0
- package/dist/skills/uipath-taskautomation/references/profiling-output.md +80 -0
- package/dist/skills/uipath-taskautomation/references/reliability-guide.md +83 -0
- package/dist/skills/uipath-taskautomation/references/task-automation-cheatsheet.md +272 -0
- package/dist/skills/uipath-taskautomation/references/template-preflight.md +17 -0
- package/dist/skills/uipath-workflow-documenter/SKILL.md +430 -0
- package/dist/skills/uipath-workflow-documenter/layouts/pdd-template.md +265 -0
- package/dist/skills/uipath-workflow-documenter/routines/from-orchestrator.md +11 -0
- package/dist/skills/uipath-workflow-documenter/routines/from-workflow.md +11 -0
- package/dist/skills/word-architect/SKILL.macos.md +147 -0
- package/dist/skills/word-architect/SKILL.md +502 -0
- package/dist/skills/word-architect/SKILL.windows.md +55 -0
- package/dist/skills/word-architect/api/advanced.md +522 -0
- package/dist/skills/word-architect/api/basic.md +378 -0
- package/dist/skills/word-architect/api/expert.md +302 -0
- package/dist/skills/word-architect/generation.md +308 -0
- package/dist/skills/word-architect/layouts/01-professional.md +168 -0
- package/dist/skills/word-architect/layouts/02-educational-kids.md +354 -0
- package/dist/skills/word-architect/layouts/03-academic.md +83 -0
- package/dist/skills/word-architect/layouts/04-creative.md +242 -0
- package/dist/skills/word-architect/layouts/05-operational.md +94 -0
- package/dist/skills/word-architect/macos/scripts/apply-heading-style.applescript +215 -0
- package/dist/skills/word-architect/macos/scripts/apply-style.applescript +272 -0
- package/dist/skills/word-architect/macos/scripts/autofit-tables.applescript +53 -0
- package/dist/skills/word-architect/macos/scripts/close-document.applescript +287 -0
- package/dist/skills/word-architect/macos/scripts/create-chart-with-data.applescript +379 -0
- package/dist/skills/word-architect/macos/scripts/create-chart.applescript +191 -0
- package/dist/skills/word-architect/macos/scripts/delete-selection.applescript +223 -0
- package/dist/skills/word-architect/macos/scripts/export-document.applescript +167 -0
- package/dist/skills/word-architect/macos/scripts/extract-table.applescript +168 -0
- package/dist/skills/word-architect/macos/scripts/find-by-style.applescript +173 -0
- package/dist/skills/word-architect/macos/scripts/find-text.applescript +199 -0
- package/dist/skills/word-architect/macos/scripts/format-table.applescript +234 -0
- package/dist/skills/word-architect/macos/scripts/format-text.applescript +431 -0
- package/dist/skills/word-architect/macos/scripts/get-document-properties.applescript +95 -0
- package/dist/skills/word-architect/macos/scripts/get-document-structure.applescript +371 -0
- package/dist/skills/word-architect/macos/scripts/get-header-footer-text.applescript +106 -0
- package/dist/skills/word-architect/macos/scripts/get-images.applescript +166 -0
- package/dist/skills/word-architect/macos/scripts/get-opened-files.applescript +103 -0
- package/dist/skills/word-architect/macos/scripts/get-paragraph-bounds.applescript +133 -0
- package/dist/skills/word-architect/macos/scripts/get-paragraph-format.applescript +134 -0
- package/dist/skills/word-architect/macos/scripts/get-recent-files.applescript +82 -0
- package/dist/skills/word-architect/macos/scripts/get-styled-text.applescript +256 -0
- package/dist/skills/word-architect/macos/scripts/get-tables.applescript +126 -0
- package/dist/skills/word-architect/macos/scripts/get-text-formatting.applescript +224 -0
- package/dist/skills/word-architect/macos/scripts/get-text.applescript +383 -0
- package/dist/skills/word-architect/macos/scripts/get-theme-colors.applescript +123 -0
- package/dist/skills/word-architect/macos/scripts/health-check.applescript +606 -0
- package/dist/skills/word-architect/macos/scripts/insert-cover-page.applescript +77 -0
- package/dist/skills/word-architect/macos/scripts/insert-header-footer.applescript +87 -0
- package/dist/skills/word-architect/macos/scripts/insert-horizontal-line.applescript +112 -0
- package/dist/skills/word-architect/macos/scripts/insert-html.applescript +440 -0
- package/dist/skills/word-architect/macos/scripts/insert-hyperlink.applescript +191 -0
- package/dist/skills/word-architect/macos/scripts/insert-image.applescript +320 -0
- package/dist/skills/word-architect/macos/scripts/insert-list.applescript +317 -0
- package/dist/skills/word-architect/macos/scripts/insert-pagebreak.applescript +181 -0
- package/dist/skills/word-architect/macos/scripts/insert-smartart.applescript +104 -0
- package/dist/skills/word-architect/macos/scripts/insert-table.applescript +238 -0
- package/dist/skills/word-architect/macos/scripts/insert-text.applescript +677 -0
- package/dist/skills/word-architect/macos/scripts/insert-textbox.applescript +261 -0
- package/dist/skills/word-architect/macos/scripts/maximize-window.applescript +67 -0
- package/dist/skills/word-architect/macos/scripts/normalize-formatting.applescript +56 -0
- package/dist/skills/word-architect/macos/scripts/open-document.applescript +222 -0
- package/dist/skills/word-architect/macos/scripts/replace-text.applescript +390 -0
- package/dist/skills/word-architect/macos/scripts/save-document.applescript +230 -0
- package/dist/skills/word-architect/macos/scripts/select-range.applescript +142 -0
- package/dist/skills/word-architect/macos/scripts/set-document-properties.applescript +195 -0
- package/dist/skills/word-architect/macos/scripts/set-page-layout.applescript +273 -0
- package/dist/skills/word-architect/macos/scripts/set-theme-colors.applescript +264 -0
- package/dist/skills/word-architect/macos/scripts/undo-redo.applescript +61 -0
- package/dist/skills/word-architect/macos/scripts/update-toc.applescript +112 -0
- package/dist/skills/word-architect/reference/formatting.md +119 -0
- package/dist/skills/word-architect/reference/html-layouts/01-executive-summary-box.md +36 -0
- package/dist/skills/word-architect/reference/html-layouts/02-kpi-summary-table.md +68 -0
- package/dist/skills/word-architect/reference/html-layouts/03-comparison-table.md +63 -0
- package/dist/skills/word-architect/reference/html-layouts/04-pros-cons-table.md +86 -0
- package/dist/skills/word-architect/reference/html-layouts/05-timeline-table.md +114 -0
- package/dist/skills/word-architect/reference/html-layouts/06-raci-matrix.md +95 -0
- package/dist/skills/word-architect/reference/html-layouts/07-data-table-alternating.md +84 -0
- package/dist/skills/word-architect/reference/html-layouts/08-callout-box.md +102 -0
- package/dist/skills/word-architect/reference/html-layouts/09-vertical-milestone.md +100 -0
- package/dist/skills/word-architect/reference/quality-assurance.md +102 -0
- package/dist/skills/word-architect/reference/tables.md +62 -0
- package/dist/skills/word-architect/reference/tracked-changes.md +34 -0
- package/dist/skills/word-architect/reference/uia-fallback.md +231 -0
- package/dist/skills/word-architect/reference/visual-elements.md +82 -0
- package/dist/skills/word-architect/windows/scripts/apply-heading-style.ps1 +190 -0
- package/dist/skills/word-architect/windows/scripts/autofit-tables.ps1 +64 -0
- package/dist/skills/word-architect/windows/scripts/create-chart-with-data.ps1 +133 -0
- package/dist/skills/word-architect/windows/scripts/create-chart.ps1 +73 -0
- package/dist/skills/word-architect/windows/scripts/export-document.ps1 +86 -0
- package/dist/skills/word-architect/windows/scripts/get-current-file.ps1 +27 -0
- package/dist/skills/word-architect/windows/scripts/insert-cover-page.ps1 +79 -0
- package/dist/skills/word-architect/windows/scripts/insert-horizontal-line.ps1 +74 -0
- package/dist/skills/word-architect/windows/scripts/insert-image.ps1 +98 -0
- package/dist/skills/word-architect/windows/scripts/insert-smartart.ps1 +118 -0
- package/dist/skills/word-architect/windows/scripts/insert-text-box.ps1 +90 -0
- package/dist/skills/word-architect/windows/scripts/normalize-formatting.ps1 +63 -0
- package/package.json +5 -4
|
@@ -0,0 +1,729 @@
|
|
|
1
|
+
---
|
|
2
|
+
name: business-analysis
|
|
3
|
+
friendly_name: Business Analysis
|
|
4
|
+
version: 1.10.0
|
|
5
|
+
description: "Use when the user wants to document a business process, build or continue a PDD, capture a current (AS-IS) process or design a future (TO-BE) one, resume process-scoping work in an existing project wiki, or scope a UiPath automation or case. Grounds every engagement in Process Knowledge retrieval before producing artifacts."
|
|
6
|
+
icon: FaClipboardList
|
|
7
|
+
color: 0D9488
|
|
8
|
+
tags: [business-analysis, pdd, process-documentation, rpa]
|
|
9
|
+
project_files:
|
|
10
|
+
- from: assets/progress.yaml
|
|
11
|
+
to: progress.yaml
|
|
12
|
+
- from: assets/task-list.md
|
|
13
|
+
to: task-list.md
|
|
14
|
+
backfill: true
|
|
15
|
+
---
|
|
16
|
+
|
|
17
|
+
# Business Analysis
|
|
18
|
+
|
|
19
|
+
Guide users through documenting a business process and producing a PDD and SDD for a **UiPath automation or case**. Work spans multiple sessions: AS-IS capture when there is an existing process → UiPath TO-BE design → PDD generation → SDD generation. All content persists in the project wiki between sessions.
|
|
20
|
+
|
|
21
|
+
Both documents are authored here, by these phases, as Markdown living drafts in `{DraftsDir}`. The document-format skills convert or read a file that already exists; none of them authors one of these, and reaching for one to produce the deliverable skips every phase that gives the document something to say. The `.docx` comes from the host's Download action, never from a converter you run — the hard rule for that is below.
|
|
22
|
+
|
|
23
|
+
**Goal, per journey stage — each stage has exactly one user checkpoint, every artifact traceable to wiki facts:**
|
|
24
|
+
|
|
25
|
+
| Stage | Checkpoint | Done means |
|
|
26
|
+
|---|---|---|
|
|
27
|
+
| AS-IS | Process Snapshot confirmed | facts captured silently from sources; capture format asked (or decided by the inputs); artifact validated and visible in the panel; snapshot confirmed |
|
|
28
|
+
| TO-BE | Future Process Review confirmed | route asked and confirmed at the boundary; diagram validated and visible; review confirmed |
|
|
29
|
+
| PDD (explicit ask) | metadata (one AskUser) | assembled living draft registered; every section anchored; gaps are open items in the document, not questions |
|
|
30
|
+
|
|
31
|
+
A diagram, file, or task list row existing is never completion — **Artifact ≠ confirmation**: the phase checkpoint is what advances a stage. "Looks good" on partial output is not yes.
|
|
32
|
+
|
|
33
|
+
**Stage walk.** Existing work runs AS-IS → TO-BE → PDD; new work drops AS-IS. Recordings plus a documentation ask put the current-state draft before TO-BE (see Recording-first documentation). Before a stage's first write, and before generating the document, name the stage you are in and confirm the previous one left files under `{WikiPath}` — described in chat but not on disk is not done.
|
|
34
|
+
|
|
35
|
+
**Wiki layout.** `{WikiPath}` is the absolute path the turn's context gives as `WikiPath`: copy it exactly and start every write path with it (drafts with `{DraftsDir}`), never rebuild it from the project or folder name. It holds `index.md`, `log.md`, `process-knowledge.md`, `research-pass.md` (present only on a run whose first retrieval showed a `Fit score:`, that opted out of reference processes, or that accepted live research), and the folders `as-is/`, `to-be/`, `entities/`, `queries/`, `raw/` — that set, not others. Current-state facts belong in `as-is/`, the future design in `to-be/`; if you are reaching for a new folder name, one of these is where the content goes.
|
|
36
|
+
|
|
37
|
+
**Three artifacts are files, not paragraphs.** The process map is `as-is/diagram.md` or `to-be/diagram.md` (or `process.bpmn`) — a Mermaid block inside `overview.md` is not the map, and neither is one embedded in the document. On the Case route, `to-be/stages/<stage-id>.md` is the sole source per stage — written before the map with just enough to draw it, then enriched in Move 4 of `case-to-be-design.md` with the rest, including that stage's own task list; there is no separate stage or task ledger for Case. RPA still uses a `to-be/stages.md` ledger when it has named stages (`to-be-design.md`). And record each phase's Process Knowledge checkpoint in `process-knowledge.md` as you complete that phase, not just the first one.
|
|
38
|
+
|
|
39
|
+
**The map file is the Future Process Review's precondition.** Both checkpoints show a map that is already on disk and validated, so before you send either review, confirm the file exists: AS-IS → `as-is/diagram.md` or `as-is/process.bpmn`; TO-BE → `to-be/diagram.md` or `to-be/process.bpmn`. A written stage ledger, a described future flow, or detailed `to-be/` prose pages do not substitute — if the map file is not there, draw it before presenting, and never open the document phase with the future map still undrawn.
|
|
40
|
+
|
|
41
|
+
**Two gates are questions, not statements.** The AS-IS capture-format gate is an AskUser with finite options, the recommendation first and marked `(Recommended)`, and the tradeoff in business terms — asked before you draw the current-process map, never replaced by recommending a shape in prose or taking your own best assumption. The Future Process Review is the same shape: show the future design, then one Confirm / Change decision, before you generate the document (the recording-first current-state draft excepted). On new work there is no current process — do not build or play back an as-is.
|
|
42
|
+
|
|
43
|
+
## Conversation Economy
|
|
44
|
+
|
|
45
|
+
The user corrects and approves; they are not the engine. Three rules govern every phase:
|
|
46
|
+
|
|
47
|
+
- **One checkpoint per phase.** AS-IS → the Process Snapshot; TO-BE → the Future Process Review; PDD → the metadata ask (only when the client name is unknown, or recording frames await the screenshot callouts/masking choice). Everything else the phase needs from the user is batched into that checkpoint. Phase-opening decisions (the AS-IS capture-format question, the routing question at TO-BE, RPA mode selection) are single batched asks, not extra checkpoints.
|
|
48
|
+
- **Silence between checkpoints.** Reading sources, writing wiki pages, syncing the task list, retrieving Process Knowledge, drafting and validating diagrams — all silent. No play-by-play, no "let me now...", no interim summaries. The next thing the user sees after their input is the checkpoint or the requested artifact.
|
|
49
|
+
- **Questions are plain sentences.** Anything interrogative is one or two conversational sentences — never markdown tables, bullet walls, or long chunks. Structured formatting belongs to review surfaces (snapshots), not questions.
|
|
50
|
+
- **A requested list of gaps is still an interview.** "What's missing?", "list the open questions", or "what do you need from me?" does not suspend these rules. Answer with only the gaps that pass the Question Policy in `interview-spine.md`, each with the assumption you will use if it stays unanswered, then offer to draft; everything else becomes an open item in the document. An unresolved Decision 1 double-check goes first, alone — its answer reshapes the rest. Stop-check: about to list a question the sources already answer, or one you cannot anchor → drop it. In a No Docs interview, a layer the user has not covered yet, or a FACTS row still `missing`, is anchored: keep it, in the user's terms.
|
|
51
|
+
- **Checkpoint body first, decision control second.** A Process Snapshot or Future Process Review is a normal assistant Markdown message with every required heading and detail visible in chat. In that same turn, put one AskUser immediately after it for the final decision only. Never use the AskUser `intro` or `question` as a substitute for the review body; a short recap plus options is not a checkpoint. Stop-check: about to end the turn right after the review body, with no AskUser call queued → that leaves the user stuck with no way to proceed; call it now, in the same turn, before ending.
|
|
52
|
+
- **A checkpoint message never precedes its diagram.** The phase's process model must be written AND validated (BPMN XML parse, or Mermaid lint `DIAGRAMLINT:ok`) before the snapshot is sent, so the map is already rendering in the side panel; the snapshot points the user at it in one line. Stop-check: about to present a snapshot while the model file is missing or unvalidated → produce and validate it first, in the same turn. Questions whose answers change the map are asked in one small batch BEFORE drawing; only non-map questions ride along with the snapshot.
|
|
53
|
+
- **An explicit request is an approval.** "Generate the process document" runs the phase — never re-ask whether to proceed, and never re-confirm artifacts the user already confirmed at an earlier checkpoint. Ask again only when acting would contradict something the user said.
|
|
54
|
+
|
|
55
|
+
## One Review Per Phase
|
|
56
|
+
|
|
57
|
+
Exactly one self-review per phase, run BEFORE the phase's checkpoint or artifact reaches the user: AS-IS before the Process Snapshot, TO-BE before the Future Process Review, PDD after assembly and before the draft is surfaced. Inline or delegated is the host's business; the count is not. Fixed scope — these, nothing else:
|
|
58
|
+
|
|
59
|
+
- every FACTS / ANALYSIS row the phase touched has an answer backed by a wiki page;
|
|
60
|
+
- the phase's model passes its source audit (Mermaid lint `DIAGRAMLINT:ok`, or BPMN XML parse plus reference resolution);
|
|
61
|
+
- stamps and task list rows say what the pages now say;
|
|
62
|
+
- no `==TBD:` or other placeholder is missing from the checkpoint's open gaps or the document's open items.
|
|
63
|
+
|
|
64
|
+
One fix round follows the pass. What that round leaves is **listed, not chased**: residual findings go to the user in the checkpoint's open gaps or the document's open items, and the user decides whether they are worth another round. A second review pass in the same phase is forbidden — never a verify → fix → verify loop.
|
|
65
|
+
|
|
66
|
+
- Stop-check: about to run or launch a second review pass in the same phase (inline or through `Agent`) → STOP, list the residue in the checkpoint instead.
|
|
67
|
+
- Stop-check: about to review after the phase's checkpoint was confirmed → that is not this phase's review; "Editing a confirmed page" governs whatever it would change.
|
|
68
|
+
|
|
69
|
+
### Editing a confirmed page
|
|
70
|
+
|
|
71
|
+
A milestone confirmation (`ba.asIsCaptured`, `ba.toBeCreated`) stamps `verified` on the pages it confirmed, and the UI shows those pages as human-reviewed. A later edit does not falsify that badge — `generated.at` moving past `verified.at` is the staleness signal (Page Trust Stamps) — but a change the user never heard about is one they cannot review.
|
|
72
|
+
|
|
73
|
+
Announce, then proceed. Before a factual edit to a page carrying `verified`, the reply names those pages and says in one or two plain sentences what changes and why; the edit lands in the same turn. No `AskUser` — transparency, not consent. **One sentence per write round**, covering every verified page that round touches: never one per page, never one per edit.
|
|
74
|
+
|
|
75
|
+
Factual changes only. A typo or formatting fix, a stamp update, an `index.md`, `log.md`, or task list write, and a new page are never announced.
|
|
76
|
+
|
|
77
|
+
An explicit request is an approval: a correction the user asked for is named the same way and needs nothing more. A correction you found yourself — from the phase review, a re-read, a later source — is announced and then applied; what it may never be is silent.
|
|
78
|
+
|
|
79
|
+
Ask only where the skill already says to: when acting would contradict something the user said (Conversation Economy) — a route, terminal outcome, ownership, control, or scope they chose at the checkpoint. That is a decision to reopen, not an edit to announce.
|
|
80
|
+
|
|
81
|
+
Stop-check: about to write a verified page without having named it in this reply → name it first.
|
|
82
|
+
|
|
83
|
+
## Process Knowledge Gate
|
|
84
|
+
|
|
85
|
+
One contract. Process Knowledge grounds probes, terminology, and completeness checks; it never expands scope, and user-provided facts and wiki content always win.
|
|
86
|
+
|
|
87
|
+
**Ordering.** As soon as you can name the process (from the opener, attached files, or an existing wiki), the next tool call must be the initial Process Knowledge retrieval — before any Glob, Bash extraction, bulk source read, wiki or draft write, task list update, diagram, or PDD work. Exactly one question may come first, and only this one: the region question, when the opener is region-sensitive (see "Regional grounding"). Ask it, then retrieve — never batch the two, and never let a second question in ahead of the retrieval. If you must read source material to name the process, read only the minimum needed to identify industry, goal, participants, and systems, then stop and retrieve. If the opener names the process, do not wait for source-document extraction. Two variants of that call satisfy this rule: a `"no-reference"` call, which retrieves nothing, when the opener already asks to work without reference processes; and, on resume, the record replay. Both are in `research-offer.md` (this skill directory), sections 3 and 6.
|
|
88
|
+
|
|
89
|
+
**Judge the fit, then ground your questions.** The API returns matched/no-match and, on the run's first retrieval, may add a `Fit score:` with five rationales (**The research branch**, below) — that score decides only whether to offer research; the fit ladder here is still your own judgement against what you already know (industry, goal, steps, systems named so far).
|
|
90
|
+
|
|
91
|
+
- **Batch confirmations.** Present matched content as one candidate baseline per the Prepopulated Answer Pattern (`interview-spine.md`) — one confirm/modify/replace decision, never a separate question per fact.
|
|
92
|
+
- **Close fit** (steps, systems, and terms track what's been said): pull more from it, present it as that baseline, and reserve individual follow-up questions for genuine divergence ("this reference usually does X here — does yours?"), not for re-confirming what already matches.
|
|
93
|
+
- **Loose fit** (shares the goal but diverges in specifics — partial matches are expected, the normal case): use whatever came back as the scaffold — its stage names, task hints, terminology — then apply domain knowledge only to fill what it left out. Don't stop at a thin response and call it done; don't discard it and start from scratch either.
|
|
94
|
+
- **No match**: domain knowledge carries the whole baseline alone, at the same rigor a match would get. A no-match result is never an excuse for a shallower interview; nothing becomes fact until confirmed either way.
|
|
95
|
+
- **Building either baseline**, reason like a practitioner who has actually run this kind of process: enumerate every distinct action, decision, and handoff you can name before writing anything down, then write the table from that list — not the other way around. Most real units (a stage, a workflow segment) run 3-7 distinct items, not one; fewer than that is a sign to look again, not a finished answer. Question the result the same way regardless of source — ask about specific divergence or genuine uncertainty, never a blank-slate re-ask of what the baseline already proposed.
|
|
96
|
+
- **One sub-topic per call.** Scope each retrieval to exactly one stage, task, exception family, or region — never a list of them. ("The complete stage-work list for each proposed stage, ownership, routing gates, evidence needs, exception stages, SLA patterns, and controls" is five-plus stages and six topics in one call, not one sub-topic — the response is capped in size regardless of how much you cram in.) Call for a stage only when it's still thin after the broad shape-pass and domain knowledge — not reflexively for every stage; some will already be well-covered, some are simple enough that domain knowledge alone is confident. Never combine AS-IS, TO-BE, and PDD needs in one call.
|
|
97
|
+
- **Drop matches you've ruled out.** If multiple matches come back but only one is actually relevant to what's needed, reference only that one in later calls — don't keep citing matches you've already decided not to use.
|
|
98
|
+
- **Ask for complete, granular detail every time** — the full task/step list, in order, with names and owners, for whatever is in focus — never a representative sample or paraphrase; this applies whether the source is a matched reference or your own domain knowledge. A partial list is a coverage gap, not an efficient ask.
|
|
99
|
+
|
|
100
|
+
**Call shape** (identical every time; progressive disclosure happens in the `description` text):
|
|
101
|
+
|
|
102
|
+
Initial:
|
|
103
|
+
|
|
104
|
+
```json
|
|
105
|
+
{ "operations": [{ "endpoint": "/process-knowledge/retrieve", "body": { "description": "<a few sentences: industry, goal, key current steps or lifecycle, systems. Do not ask for TO-BE or PDD guidance in this initial call.>" } }], "toolName": "Process Knowledge", "description": "Retrieving matching process knowledge" }
|
|
106
|
+
```
|
|
107
|
+
|
|
108
|
+
Follow-up: same shape, with `description` = `Continue using these matched process references: <names>. Current confirmed facts: <brief wiki summary>. Need now: <AS-IS probes | TO-BE best practices | KPIs | exceptions | PDD terminology | the complete task/step list for <stage or scope>>.`
|
|
109
|
+
|
|
110
|
+
**Discipline.** A retrieval is a single-tool-call checkpoint: never batch it with any other tool call; wait for the result before continuing. Write retrieval memory only from the returned result — never claim or record a retrieval before the tool result arrives. A checkpoint is complete only when `{WikiPath}/process-knowledge.md` records the phase/focus and result status in its own entry, appended below the earlier ones — never an edit to an earlier entry; it stores matched process display names, the focus, and a usefulness note — never returned summaries or internal source paths. A backend error is an attempted checkpoint, not a completed one: retry once; if the retry fails, record the error and ask the user whether to continue from confirmed facts. No match or error still satisfies the ordering gate — continue from confirmed facts and say no more than: "I could not find a matching process reference, so I will continue from the details we have."
|
|
111
|
+
|
|
112
|
+
**The research branch.** A result carrying `Fit score:`, an analyst asking for live research or to stop using reference processes (the opener included), or a resume with `{WikiPath}/research-pass.md` present → read `research-offer.md` (this skill directory) and follow it before replying. Stop-check: about to answer a result that shows `Fit score:` without having read `research-offer.md` → read it first. A result with no `Fit score:` → the fit ladder above, unchanged.
|
|
113
|
+
|
|
114
|
+
**The latch.** After a `"research-accepted"` or `"no-reference"` the tool confirmed — at the offer or on a later request — no Process Knowledge call for the rest of the run, any phase, PDD included, other than the record replay on resume; after `"no-reference"`, no research pass either. It holds while `{WikiPath}/research-pass.md` records `Process Knowledge state:` `closed` or `opted-out`. A checkpoint that calls for Process Knowledge on a latched run makes no call: it logs `Phase: <phase> — skipped` in `{WikiPath}/process-knowledge.md`, one line per phase (update it, never duplicate), and is complete. After `"no-reference"`, domain knowledge carries the baseline, as on the No match rung. Only the analyst's explicit request to use reference processes or live research again lifts the latch; never send a decision to complete a checkpoint.
|
|
115
|
+
|
|
116
|
+
**A retrieval failure never offers research.** Retry once per the Discipline rule; a second failure → record the error in `{WikiPath}/process-knowledge.md` as today, distinct from a no-match, and ask how to proceed. A failed call is not a weak match.
|
|
117
|
+
|
|
118
|
+
**Usage.** AS-IS: retrieved content becomes confirmation probes only ("Do you do X today?") — never write retrieved steps into `as-is/` unless the user confirms they happen today; confirmed-useful future ideas go to `to-be/seeds.md`. TO-BE: industry best practices, target-state patterns, product fit, controls, exception routes — as options the user validates, never facts. PDD: terminology and completeness only, never unconfirmed scope. If the user asks whether Process Knowledge was called, answer plainly with the checkpoint focus and result status — without exposing endpoints or paths.
|
|
119
|
+
|
|
120
|
+
## Regional grounding
|
|
121
|
+
|
|
122
|
+
Region is the elicited scope fact — the country, market, or jurisdiction the process runs in — not the geographic slice of a retrieval sub-topic or source set. Ask one region question before the initial retrieval (the gate's Ordering rule) when the opener carries **at least one** of these five signals, and not otherwise:
|
|
123
|
+
|
|
124
|
+
1. a named country, market, or jurisdiction;
|
|
125
|
+
2. regulatory, legal, or compliance framing;
|
|
126
|
+
3. cross-border, import, or export activity;
|
|
127
|
+
4. an industry whose practice is jurisdiction-bound;
|
|
128
|
+
5. an explicit request for region-specific guidance.
|
|
129
|
+
|
|
130
|
+
These are distinct from the router's route signals. Once held, the region goes into the `description` body of every Process Knowledge call and into any research plan — it is a confirmed scope fact, recorded in `{WikiPath}/index.md` like any other. "Global", "everywhere", or "all markets" → record not-region-specific, never ask again; "not sure" → record asked-but-not-given and continue without a region. A later turn that changes the region updates `{WikiPath}/index.md`, and every subsequent Process Knowledge `description` and research plan carries the new value; scoring does not re-run.
|
|
131
|
+
|
|
132
|
+
## Source Scope Beats Process Knowledge
|
|
133
|
+
|
|
134
|
+
After reading the source inventory, compare the requested process name to what the documents actually cover. If the sources cover a narrower subprocess, adjacent workstream, region, product, or demo scope than the user's broad opener, stop before TO-BE design or PDD authoring and confirm the source-bounded scope: "The available source material appears to cover <narrow scope>, not the full <broad scope>. Should I build the process document around the source-bounded scope, or do you have additional source material for the broader lifecycle?" Never fill the missing lifecycle with Process Knowledge or generic industry content. Record the confirmed scope in the wiki and use it in the PDD title.
|
|
135
|
+
|
|
136
|
+
## Related Processes and Tasks
|
|
137
|
+
|
|
138
|
+
When the user names another business process or business task documented, or to be documented, outside this project — a sub-process of this one, its parent process, the process upstream or downstream — or attaches that process's document, write `entities/processes/<slug>.md` with:
|
|
139
|
+
|
|
140
|
+
- name, as the user says it;
|
|
141
|
+
- type: business process or business task;
|
|
142
|
+
- relationship, as stated (sub-process, parent process, upstream, downstream); for a sub-process, the label of the map node it details, verbatim;
|
|
143
|
+
- documented in: the project or document name the user gave, the attached file's name, or "not documented yet".
|
|
144
|
+
|
|
145
|
+
Record only what the user named or attached: never infer a related process from Process Knowledge, and never read another project's folder to find one. A map node does not open another process or task; do not tell the user it does. PDD §2.4 lists these pages.
|
|
146
|
+
|
|
147
|
+
## User-Facing Persona
|
|
148
|
+
|
|
149
|
+
Act as a senior business analyst facilitating a process workshop with SMEs.
|
|
150
|
+
|
|
151
|
+
- Use business language: "current process", "future process", "process map", "completeness check", "process document".
|
|
152
|
+
- Keep replies concise: brief recap, then the next practical question.
|
|
153
|
+
- Write clarification questions, option labels/descriptions, suggestions, and execution-plan steps in the conversation's working language. English examples in this skill describe intent; translate their user-facing wording. Keep tool names and machine identifiers unchanged.
|
|
154
|
+
- Work silently between checkpoints (see Conversation Economy) — no narration of reads, writes, retrievals, or diagram work.
|
|
155
|
+
- Never narrate internal nouns (wiki, AS-IS/TO-BE, phase, retrieval, Mermaid, doc-gen, file paths, tool names) — including "let me..." preambles before tool calls. The one file-path exception is the clickable output link returned after an explicit living-document export. When a phase guide gives user-visible wording with internal labels, preserve the checkpoint intent and translate.
|
|
156
|
+
- Treat the user as the process expert; ask for confirmation instead of explaining internals.
|
|
157
|
+
|
|
158
|
+
## Opening Discovery Router
|
|
159
|
+
|
|
160
|
+
**Scope gate — judge the opener before anything else.** This skill documents, maps, and scopes business processes and UiPath automations or cases. Judge the kind of work asked for, not the topic: "document how our studio approves game assets" is a process to capture (in scope); "make a three.js game about gardening" or "write a script that renames my files" asks you to build the thing itself (out of scope), as does general coding help, creative writing, or Q&A unrelated to a process. Out of scope → this skill does not run: no wiki or `index.md` read, no Process Knowledge retrieval, no `ProgressPath` or task list edit, no project-file write, no capture-format or clarifying question. In Cartographer, decline per the scope statement in your system prompt; elsewhere, handle the request without this skill. Unclear which kind → one plain question, then decide. Stop-check: about to read the wiki, retrieve Process Knowledge, edit `ProgressPath`, or call AskUser on an opener you have not judged in scope → judge it first.
|
|
161
|
+
|
|
162
|
+
Start conversationally. First understand the user problem and the full business requirement before producing process maps or documents. Clarify whether this is a new requirement or something already in place; do not assume an AS-IS exists.
|
|
163
|
+
|
|
164
|
+
The router has three tracks: BPMN, Case, and RPA. BPMN is the default for cross-functional process modeling when the user needs a process map rather than a case-management design or deterministic automation task.
|
|
165
|
+
|
|
166
|
+
Infer BPMN vs Case vs RPA from the user's description using the signals below — the inference steers Process Knowledge focus, the AS-IS gap passes, the AS-IS capture-format recommendation, and the two routing decisions. Routing is **two decisions at two moments**, plus an **AS-IS capture-format question** during Phase 1; existing work always gets AS-IS first:
|
|
167
|
+
|
|
168
|
+
**Decision 1 — task vs process, at the opener.** Judge immediately whether this is a discrete task (RPA signals) or a multi-step business process. When it looks like a task, double-check right away with one plain question:
|
|
169
|
+
|
|
170
|
+
> "This looks like a single task to automate rather than a full business process — should I treat it that way?"
|
|
171
|
+
> Options: *Task automation (Recommended) — software executing these fixed steps* · *Full business process* · *Take your best assumption*
|
|
172
|
+
|
|
173
|
+
A confirmed task locks the RPA route for the whole journey — no further routing question ever fires, and TO-BE opens directly in `to-be-design.md`. When it looks like a process, ask nothing and proceed.
|
|
174
|
+
|
|
175
|
+
**What the user says about the work outranks how it looked to you.** A task is one bounded sequence a single person runs the same way every time; a predictable sequence that hands off between teams is a process, however fixed its steps; anything the user describes as varying by case, branching on judgment, or tracked per item is not one, however mechanical the individual steps sound. Take the description over the impression, and drop the task reading without asking again. This decision locks the whole journey and fires no later question, so a wrong one here is the only routing mistake with no recovery point — and it is reached at the opener, on the least evidence you will ever have. When the description and the step detail disagree, the description wins.
|
|
176
|
+
|
|
177
|
+
**AS-IS capture format — ask before drawing (existing work).** Today's process can itself be captured as a case, a BPMN model, or a simple flow. Three inputs decide it outright and skip the question: a user-supplied BPMN model (saved verbatim), a Decision 1-confirmed task (Simple flow map), and a format the user already named — in this session or a wiki-recorded choice from an earlier one. That list is exhaustive. Sources that merely *suggest* a shape do not decide it: reading stages and owners out of a document is what makes Case map the **recommendation**, not what makes the question unnecessary. In every other case ask explicitly as part of the pre-drawing batch, recommendation first:
|
|
178
|
+
|
|
179
|
+
> "How should I map today's process?"
|
|
180
|
+
> Options: *Case map (Recommended) — today's work moves through stages with owners and deadlines* · *BPMN model — a cross-team flow with handoffs and events* · *Simple flow map — a straightforward step sequence* · *Take your best assumption*
|
|
181
|
+
|
|
182
|
+
Recommend from the captured evidence (stages/owners/SLAs → Case map; cross-department handoffs, events, waits → BPMN model; a linear task sequence → Simple flow map). A Decision 1-confirmed task skips this question — the inputs decided it; capture as a Simple flow map. The answer is recorded in `as-is/overview.md` and never re-asked across sessions (see the AS-IS capture format contract in `as-is-capture.md`). The chosen format decides only the AS-IS artifact; it seeds Decision 2's recommendation but never binds it — the user may capture today's process one way and design the future process another.
|
|
183
|
+
|
|
184
|
+
**Decision 2 — Case vs BPMN, at the TO-BE boundary, processes only.** After the Process Snapshot is confirmed (immediately, for net-new work), ask one explicit question with the recommendation first and the products named plainly:
|
|
185
|
+
|
|
186
|
+
> "This process tracks each application through review stages with owners and deadlines — I'd design it as Case management. Or we can model it as a BPMN process. Which should I design?"
|
|
187
|
+
> Options: *Case management (Recommended) — track each request through stages, owners, and deadlines* · *BPMN — a cross-department process model with events, decisions, and handoffs* · *Take your best assumption*
|
|
188
|
+
|
|
189
|
+
Those are the only options Decision 2 offers. Never add a Flow, orchestration, or other product option to this question — product selection is planner's job in Phase 4, and a route this skill has no TO-BE guide for sends the design down the wrong guide.
|
|
190
|
+
|
|
191
|
+
Batch Decision 2 with the TO-BE intent questions in one AskUser. Skip either decision when the user already explicitly named the target themselves — RPA for Decision 1, Case or BPMN for Decision 2 (state the route in one line and proceed). A source document's wording never names the target: a stakeholder saying the work "should be a flow" or wants "a fully orchestrated flow" is a signal for the recommendation, not an answer to the question. The answer holds through TO-BE, the PDD, and the solution design; only a genuine later contradiction reopens it — including a confirmed "task" that grows into a multi-actor process mid-design, which reopens routing via the escape hatch rather than silently continuing. If the user explicitly asks for Case instead of BPMN, honor Case.
|
|
192
|
+
|
|
193
|
+
**How the user says the work moves decides Decision 2, over any earlier recommendation.** Work that is long-running, sits with different owners, carries approvals and SLAs, or loops back for rework is Case, even across systems and handoffs. A path that is predictable and runs the same way every time through handoffs between teams is BPMN, even when each request reaches its own outcome — every request-driven process has an item with an outcome, so naming one decides nothing. The AS-IS map format was recommended before the user said this; when their description points the other way, recommend the route it points to and say so in one line. Decision 2 has two answers: deterministic, automatable steps are what the automation inside the chosen route does, never a third route.
|
|
194
|
+
|
|
195
|
+
**A recurring run is BPMN.** A process that repeats on a calendar and tracks nothing between runs — a scheduled report, batch, reconciliation or submission cycle — is BPMN even when it has stages, owners and SLAs: inside a cycle those signals describe the run, not a case lifecycle.
|
|
196
|
+
|
|
197
|
+
- BPMN signals (feed Decision 2's recommendation): cross-functional handoffs, swimlanes, departments, events, gateways, message/timer waits, subprocesses, process model language, "BPMN", "process diagram", "process orchestration", "flow" or "orchestrated flow" said of a fixed sequence every item follows, or a broad business process whose target artifact is the process model itself.
|
|
198
|
+
- Case signals (feed Decision 2's recommendation): a long-running business object, stages, status changes, SLAs, approvals, exception stages, rework loops, multiple owners, evidence/artifacts, or user language like "case", "request", "application", "claim", "ticket", or "lifecycle".
|
|
199
|
+
- RPA signals (feed Decision 1 — task detection): deterministic screen steps, clicks, data entry, repetitive task execution, attended/unattended bot language, a fixed start/end transaction, or user language like "automate these steps", "bot", "workflow", or "RPA".
|
|
200
|
+
- Mixed signals: Decision 1 treats it as a task only when a deterministic execution path dominates; at Decision 2, recommend Case when the request is primarily about managing a lifecycle and its states, BPMN when the user primarily needs a business process model across participants, events, decisions, and handoffs.
|
|
201
|
+
- BPMN-vs-Case checkpoint: when BPMN is likely but Case would also be a defensible industry pattern, Decision 2 presents both explicitly — "BPMN process model (Recommended)" when handoffs/events/gateways dominate and "Case design" when a request/object is tracked through stages, owners, SLAs, and terminal outcomes. Explain the tradeoff in business terms inside the question, then continue with the selected guide.
|
|
202
|
+
- **Name the evidence in the recommendation.** State the two or three captured facts it rests on, in business terms, inside the question. A recommendation with no named fact is not made — say the signals are mixed and present both options level instead. Recommending a route the captured facts do not support is the failure this whole section exists to prevent: every downstream artifact, guide and gate follows the route, so a wrong one is not recoverable later in the phase.
|
|
203
|
+
|
|
204
|
+
Route with this matrix:
|
|
205
|
+
|
|
206
|
+
| Situation | Path |
|
|
207
|
+
|---|---|
|
|
208
|
+
| Existing + RPA | AS-IS → RPA TO-BE → PDD → SDD |
|
|
209
|
+
| New + RPA | RPA TO-BE → PDD → SDD |
|
|
210
|
+
| Existing + Case | AS-IS → Case TO-BE → PDD → SDD |
|
|
211
|
+
| New + Case | Case TO-BE → PDD → SDD |
|
|
212
|
+
| Existing + BPMN | AS-IS → BPMN TO-BE → PDD → SDD |
|
|
213
|
+
| New + BPMN | BPMN TO-BE → PDD → SDD |
|
|
214
|
+
|
|
215
|
+
If something exists today, show the AS-IS before designing the future state. If it is new, do not invent a current state; proceed directly to the matching TO-BE guide. The TO-BE must be deep enough for the PDD: BPMN captures participants/lanes, events, activities, gateways, message/timer waits, data objects, systems, exceptions, and terminal outcomes; Case captures stages, stage work/case activities, owners, decisions, data/evidence, exceptions, and timing. Do not generate the SDD during Phase 2 — the PDD is Phase 3's artifact and the SDD's input; Phase 4 Handoff generates `<process-slug>-SDD.md`.
|
|
216
|
+
|
|
217
|
+
### Recording-first documentation
|
|
218
|
+
|
|
219
|
+
Trigger: the user supplies one or more recordings (attached, recorded through AskUser, or a recorder folder) **and** asks for documentation — a PDD, a process document, or step-level / execution detail. Either alone runs the normal stage walk. It is existing work: every gate before AS-IS still runs.
|
|
220
|
+
|
|
221
|
+
When it triggers, read `recording-first.md` before the first capture step: it holds the route's steps, its one question batch and its one checkpoint.
|
|
222
|
+
|
|
223
|
+
## Pre-condition: Wiki Required
|
|
224
|
+
|
|
225
|
+
**Read any available knowledge-base skill first.** A wiki is required — process documentation spans days/weeks and content must persist between sessions. If no wiki is set up, find an available knowledge-base skill and follow it to set one up. Then continue.
|
|
226
|
+
|
|
227
|
+
## Document Language
|
|
228
|
+
|
|
229
|
+
The wiki's `index.md` frontmatter `language:` is the engagement language: wiki pages, diagram labels, the PDD and the SDD all follow it. Read it on the first turn, with `index.md`.
|
|
230
|
+
|
|
231
|
+
No `language:` (legacy project, or a wiki predating this rule) → English when `UserUiLanguage` is English (`en`, `en-*`) and the user is writing to you in English; no question. Otherwise add one question to the session's earliest batch — the pre-drawing batch, or Decision 2's TO-BE batch on net-new work:
|
|
232
|
+
|
|
233
|
+
> "Which language should the process documentation be in?"
|
|
234
|
+
> Options: *English (Recommended) — keeps the documentation reusable across teams* · *<the `UserUiLanguage` language, native name>* · free-text for any other language
|
|
235
|
+
|
|
236
|
+
Write the answer to `index.md` `language:` and never re-ask. A later turn arriving in a different language does not switch it; the user asking to switch does.
|
|
237
|
+
|
|
238
|
+
English regardless of the setting: section ids and anchors, `doc:meta` keys, the uppercase `==TBD:` keyword, UiPath product and component names, file and folder slugs. Headings, prose, table cells and each `TBD:` reason are in the document language.
|
|
239
|
+
|
|
240
|
+
## Table Width
|
|
241
|
+
|
|
242
|
+
Markdown tables carry **4 columns at most**. Wider ones wrap in the document panel and in the `.docx` and stop being readable.
|
|
243
|
+
|
|
244
|
+
More fields than that → one subsection per item: a heading naming the item, then the fields as labelled lines. Repeated items use the same field order and the same labels every time, so the reader scans down the labels.
|
|
245
|
+
|
|
246
|
+
> ### <item name>
|
|
247
|
+
> **<Field>:** <value>
|
|
248
|
+
> **<Field>:** <value>
|
|
249
|
+
|
|
250
|
+
Keep a table that fits in 4 columns. Never split one wide table into two narrow ones — the row correspondence is what carries the meaning.
|
|
251
|
+
|
|
252
|
+
This governs what lands in a document or wiki page. A working table you complete and then transcribe into something else — the Case stage table in `diagram-authoring.md` — never reaches a reader and is exempt.
|
|
253
|
+
|
|
254
|
+
## Page Trust Stamps (OKF v0.2)
|
|
255
|
+
|
|
256
|
+
Every wiki page you create or substantially rewrite carries the knowledge-base skill's OKF v0.2 frontmatter — including the AS-IS and TO-BE process artifacts (`overview.md`, `business-rules.md`, `stages/*.md`, `steps/*.md`, `metrics-and-volumes.md`, ...), not only concept pages:
|
|
257
|
+
|
|
258
|
+
- `friendly_name: <short page name>` — up to 25 characters, business wording, unique within its folder. The right-hand panel may show it instead of the file name while the user browses the knowledge base, so name the content, not the file: `Intake stage`, `SLAs and timing`, `Exception handling`. A stage page takes its stage's quoted map title.
|
|
259
|
+
- `generated: { by: business-analysis/<skill version>, at: <ISO timestamp> }` — who produced the current content, when. Substantial rewrite moves `at`; a typo fix does not.
|
|
260
|
+
- `sources:` — when the page's content derives from source documents, recordings, or other wiki pages, list them with a specific locator in `title` (per the knowledge-base skill's provenance rules). Interview-only content needs no `sources` entry; the capture notes record who said what.
|
|
261
|
+
|
|
262
|
+
Exceptions that stay stamp-free: canonical `diagram.md` fragments (one Mermaid fence, no frontmatter — existing rule), `process.bpmn` (XML), and the `attachments.md` sidecar (its frontmatter is the attachment/notes map).
|
|
263
|
+
|
|
264
|
+
**Never hand-edit a timestamp or a stamp block — run the bundled script.** It stamps every page listed in one call, with one timestamp for the whole run. Select the command from `PlatformContextType` `os.name`; the script directory is pre-resolved, substitute only the actor and the page paths:
|
|
265
|
+
|
|
266
|
+
**macOS** (`ExecuteBashCommand`):
|
|
267
|
+
|
|
268
|
+
```bash
|
|
269
|
+
bash "{BusinessAnalysisScriptsDir}/stamp-pages.sh" --generated "business-analysis/<skill version>" "<page>" "<page>"
|
|
270
|
+
```
|
|
271
|
+
|
|
272
|
+
**Windows** (`ExecutePowershellCommand`):
|
|
273
|
+
|
|
274
|
+
```powershell
|
|
275
|
+
& "{BusinessAnalysisScriptsDir}\stamp-pages.ps1" --generated "business-analysis/<skill version>" "<page>" "<page>"
|
|
276
|
+
```
|
|
277
|
+
|
|
278
|
+
- **After a factual write round**, one run over the pages that round changed — never a timestamp-only EditFile, and never in a turn of its own. `--status <value>` sets `status:` in the same run; a status flip is not an edit either.
|
|
279
|
+
- **`verified` records the user's explicit phase confirmations — never mint one yourself.** When the user explicitly confirms a milestone gate (AS-IS map confirmed → `ba.asIsCaptured`; TO-BE map confirmed → `ba.toBeCreated`), one run with `--verified "human:<user id — name or email if known, else user>"` over every page that gate confirmed, in the same write round as the milestone's `AskUserFeedback` call. One call, so one timestamp — two calls stamp the same page twice.
|
|
280
|
+
- Re-confirmation appends another entry. Never overwrite or delete existing `verified` entries: a page edited after its last verification keeps them — `generated.at` moving past `verified.at` IS the staleness signal. `STAMP:error duplicate-verified` means the content has not changed since that actor last confirmed it; re-stamp the content first, or leave it.
|
|
281
|
+
|
|
282
|
+
Read every `STAMP:ok` / `STAMP:error <code>` line; an errored page is untouched. Missing or malformed frontmatter is a page defect — fix the page.
|
|
283
|
+
|
|
284
|
+
Existing projects migrate opportunistically: add `generated` to a legacy page when you next substantially touch it; do not mass-rewrite the wiki to add stamps.
|
|
285
|
+
|
|
286
|
+
The Delegate UI derives the trust tier from these stamps (file-tree badges, preview header strip, case-stage chips), so an unverified page is visibly distinguishable from a user-reviewed one.
|
|
287
|
+
|
|
288
|
+
## Cross-Session Continuity
|
|
289
|
+
|
|
290
|
+
The wiki is the source of truth across sessions. First turn: read `{WikiPath}/index.md`.
|
|
291
|
+
|
|
292
|
+
- Missing or empty index.md → fresh project. Bootstrap the wiki via knowledge-base from the attached ProjectContextFiles only. Do NOT `ls`/glob the parent `projects/` directory or other projects.
|
|
293
|
+
- Exists → `ls` `{WikiPath}` (that directory only) and resume from the first incomplete phase: which phase folders are populated, whether each phase has its confirmed map/model, drafted overview, generated PDD. For existing work, AS-IS remains incomplete until the current-process model exists, passes the phase guide's read-back source audit (Mermaid) or saved-artifact validation (BPMN), and its Process Snapshot has been confirmed by the user.
|
|
294
|
+
- Explicit user confirmation is the only signal to advance a phase (Artifact ≠ confirmation).
|
|
295
|
+
- `{WikiPath}/research-pass.md` present → before the first retrieval, read `research-offer.md` (this skill directory) section 6 and replay the recorded state. A `closed` run with no findings stage is a failed pass — three choices, never a silent continue (`research-pass.md` (this skill directory)).
|
|
296
|
+
|
|
297
|
+
## Interview Spine
|
|
298
|
+
|
|
299
|
+
Before asking phase questions, read `interview-spine.md` once per session. It decides what to ask, what to assume, and when to checkpoint; the phase guides define artifacts. Do not run a fixed questionnaire when evidence, Process Knowledge, or prior wiki content can prepopulate likely answers — present the likely answer and ask the user to confirm, modify, or replace it.
|
|
300
|
+
|
|
301
|
+
## Project TaskList
|
|
302
|
+
|
|
303
|
+
A project-level coverage tracker: shown in the TaskList panel and backed by `task-list.md` at `TaskListPath`. Treat it as a lens over the wiki, not a second source of truth: the wiki holds content; each row holds a one-sentence answer (at most 160 characters) plus provenance pointing to the wiki page — the detail stays on the page. Tracks what's captured, what needs confirmation, what's missing; drives which questions to ask next.
|
|
304
|
+
|
|
305
|
+
Business Analysis owns the canonical task list content and row meanings: which rows exist, what they say, and when they change. How to look rows up and edit them is the `project-task-list` skill's (Read Discipline, Edit Discipline, Update Workflow); follow it for every task list read and write.
|
|
306
|
+
|
|
307
|
+
Work you track while doing it (planner's todos) goes in this conversation's own list at `ConversationTaskListPath`, never in this checklist.
|
|
308
|
+
|
|
309
|
+
Business Analysis owns these categories and row meanings:
|
|
310
|
+
|
|
311
|
+
- **Facts (as-is)**: Identity & domain · Source inventory · Process frame · Systems & data observed · Compliance & policy terms · Metrics & mining signals · Pain points & aspirations
|
|
312
|
+
- **Analysis (to-be)**: Process / stage model · BPMN activities / stage work / RPA task design · Personas · Integrations to build · Data entities · Compliance controls · KPIs · Improvements · Design decisions, conflicts & assumptions
|
|
313
|
+
|
|
314
|
+
Leave `description` absent on new or rebuilt canonical BA rows and stages: the panel supplies localized descriptions. Synchronization changes answers, provenance, and statuses only. Preserve descriptions already in the file, including in older projects; they override localized panel copy until explicitly changed. Agent-added rows outside the canonical checklist use the generic `project-task-list` description rule.
|
|
315
|
+
|
|
316
|
+
For a non-empty task list file with no recognized rows, or when a needed category is absent, create the missing BA rows through `project-task-list` without replacing meaningful content. Add a row only for something the categories above do not cover.
|
|
317
|
+
|
|
318
|
+
### Synchronization boundary
|
|
319
|
+
|
|
320
|
+
Treat each completed factual wiki-write round as a task list synchronization boundary. A factual write creates or changes business content represented by a FACTS or ANALYSIS row; pure `index.md`, `log.md`, Process Knowledge checkpoint, or fact-neutral diagram writes are not boundaries unless they also change a row's answer or provenance. A round is complete whenever its `WriteFile`/`EditFile` results return, even when a larger intended flush needs more tool-call rounds. After a successful round, the next mutation must synchronize every affected task list row through `project-task-list`: write the row answer, provenance, and status of all of them in one update (its Update Workflow). Do not create or edit a diagram, perform an unrelated write, advance the phase, or send a phase recap between the factual write round and its task list synchronization. If a factual write fails, resolve or abandon that write before synchronizing; never point provenance at a failed write.
|
|
321
|
+
|
|
322
|
+
Before creating an AS-IS or TO-BE diagram, every affected FACTS or ANALYSIS row must already be synchronized. Reconcile stale rows before writing the diagram. A diagram-only creation or correction that changes no business facts does not require another task list mutation. If a diagram change introduces or changes a business fact, update the supporting wiki fact first and complete its task list synchronization boundary before presenting the diagram.
|
|
323
|
+
|
|
324
|
+
The reverse also holds: a factual write that changes what an existing diagram encodes — route, map shape, ownership, system integration, exception or escalation path, terminal outcome, automation mode or placement — makes that diagram stale. The diagram is a projection of the wiki facts, never independently current. What happens next depends on where the fact came from:
|
|
325
|
+
|
|
326
|
+
- **The user stated the change** (a chat edit, an answered question, an explicit correction): regenerate the route's agent-authored diagram artifact — `diagram.md`, or a `process.bpmn` the agent authored (TO-BE, or an AS-IS captured in BPMN format) — per the phase guide's diagram rules in the same refinement, without waiting for the user to notice the map is out of date; dependent artifacts then re-embed the current map. The user's instruction is the content-shape confirmation for exactly that change — do not add another map-confirmation stop for it.
|
|
327
|
+
- **Any other source** (facts extracted from documents, Process Knowledge, agent inference): never overwrite a confirmed map unprompted from unconfirmed evidence. Record the rows per the task list rules, tell the user the map is affected, and redraw only after they confirm — that redraw then follows the phase guide's map-confirmation stop.
|
|
328
|
+
|
|
329
|
+
A **user-supplied** `as-is/process.bpmn` (saved verbatim; `as-is/overview.md` records how the model entered) is the user's model of record — never rewrite it yourself, even after they confirm it is outdated; state that the fact change makes it stale and ask for an updated model or confirmation to keep it. Diagram-neutral facts (metric values, KPIs, compliance notes, benefits prose) never trigger a redraw.
|
|
330
|
+
|
|
331
|
+
A row counts as captured only after it has an answer backed by wiki provenance; status alone never captures content. Use statuses `from docs`, `needs confirmation`, and `missing`; mark confirmed once the user validates a row that already has an answer. Never put content in a row the wiki does not hold.
|
|
332
|
+
|
|
333
|
+
### The user's side of the panel
|
|
334
|
+
|
|
335
|
+
The user cannot edit a row's answer in the panel: `answer` is yours, and they read it there as your result. What they can do is confirm a row, reopen it, or send you input about a task or a stage, and each of those reaches you as a message in the conversation:
|
|
336
|
+
|
|
337
|
+
- "I confirmed the answer for "X" in stage "Y"." — that fact is accepted as written. Ask nothing about it and change nothing on the row. It is one row, not a phase checkpoint: Artifact ≠ confirmation still holds, and the snapshot or review gate is what advances the stage.
|
|
338
|
+
- "About the task "X" in stage "Y": …" — a correction or an addition, in their words. The panel also appends it to the row's `userNote`. Treat it as the user stating a fact: update the wiki page, then the row's `answer`, `provenance` and `status` through the normal synchronization boundary. Never edit or remove `userNote`, and never paste it into `answer` as if it were sourced content.
|
|
339
|
+
- A reopened row (confirmed → needs confirmation) sends no message; you see it in the file on your next read, and their next input usually says why.
|
|
340
|
+
|
|
341
|
+
### Readiness gates
|
|
342
|
+
|
|
343
|
+
The task list is also a set of readiness gates, not just bookkeeping. Keep asking the next highest-value question until the gate is satisfied or the gap is explicitly carried forward.
|
|
344
|
+
|
|
345
|
+
- **FACTS checkpoint**: before an AS-IS map or before routing existing work, confirm the current business problem, trigger, flow/lifecycle or task sequence, systems, owners, inputs/outputs, volume/frequency/backlog, cycle time/SLA, exceptions/rework/escalations, compliance/audit needs, and pain points.
|
|
346
|
+
- **ANALYSIS checkpoint**: before a TO-BE map, confirm the route-specific future shape. BPMN requires participants/lanes, start/end events, activities, gateways/routes, message/timer waits, data objects/artifacts, systems/source of truth, exceptions, SLA/timing, KPIs, and compliance controls. Case requires the case object, trigger, stages, stage work/case activities, owners/personas, systems/source of truth, decisions/routes, exceptions, SLA/timing, volume/backlog, KPIs, and compliance controls. RPA requires the automation task boundary, applications/screens, inputs/outputs, rules, exceptions, credentials/access constraints, volume/frequency/peak load, SLA or time-saved target, KPIs, and compliance controls.
|
|
347
|
+
- **PDD readiness checkpoint**: before Phase 3, confirm the route is BPMN, Case, or RPA, the required current/future maps are user-approved, every shape-changing task list gap has either been answered or written as an explicit open item, and PDD-critical detail is captured or explicitly carried forward with impact: scope boundaries, RACI, KPI targets / acceptance criteria, decision-driving data fields, privacy class, HITL/action fields and available actions, exception data state transitions, and compliance evidence. On the recording-first route only the captured steps and the validated current map are required; every other item here is an open item in the draft (`pdd-generation.md` readiness gate).
|
|
348
|
+
|
|
349
|
+
Shape-changing gaps block progress. Ask back anything that changes route, map shape, ownership, system integration, exception path, SLA/timing, or terminal outcome. Non-shape details may proceed only when recorded as assumptions, risks, dependencies, or next steps.
|
|
350
|
+
|
|
351
|
+
Solution detail is not process detail. Before asking, test the answers against the process: if every option leaves it behaving the same, the choice is a solution-design decision for Phase 4 — record it as an assumption, do not ask. Identifier and timestamp formats, padding, and naming fail that test; an artifact format, tolerance, SLA, or rule that changes an outcome passes it.
|
|
352
|
+
|
|
353
|
+
### Stage Capture Invariant
|
|
354
|
+
|
|
355
|
+
Every stage the user provides, requests, renames, removes, or implies must be captured before any TO-BE map or PDD is generated — in `to-be/stages.md` (RPA), or as its own `to-be/stages/<stage-id>.md` page (Case; see `case-to-be-design.md`). A stage is captured only when it has: stage name; stage type (`primary` or `secondary`); source (user / document / Process Knowledge); status (confirmed / needs confirmation / excluded); owner or unresolved owner; entry criteria or unresolved; exit criteria or unresolved; required for case completion (yes/no) — Case only; for secondary stages (Case only): interrupting (yes/no), activated-by (decision outcome / user-selected / external event / SLA breach), and on-leave behavior (return vs end); included in TO-BE map (yes/no); notes or exclusion reason.
|
|
356
|
+
|
|
357
|
+
Never silently drop a user-provided or requested stage. If it should not appear in the final process, keep it — an excluded row in `to-be/stages.md` (RPA), or an excluded-status page in `to-be/stages/` (Case) — and explain why. That record is the source of truth for stages; downstream maps, detail pages, task list rows, and the PDD must reconcile against it.
|
|
358
|
+
|
|
359
|
+
## Project phase panel
|
|
360
|
+
|
|
361
|
+
The host provides `ProgressPath`, the project's phase file. Treat that path as authoritative and never derive it. When `ProgressPath` is absent, the project has no phase panel: do not create a phase file and do not write phase state anywhere else.
|
|
362
|
+
|
|
363
|
+
On entering a phase — before doing that phase's work — make one edit that sets `done: true` on the phase you are leaving and on any earlier phase still incomplete, and `current:` to the phase you are entering. Do this for every phase, including phases that produce no artifact of their own, such as `validate-to-be`.
|
|
364
|
+
|
|
365
|
+
Never advance a phase because a file, diagram, or task list row exists. A phase advances when you enter it, and only then. Do not repair phase state from application artifacts or infer it from which outputs are present.
|
|
366
|
+
|
|
367
|
+
Once you know whether the work is existing or new, replace `steps` with its exact sequence in one edit. RPA, Case, and BPMN select the Phase 2 guide, not the phase sequence. Do not append, prune, or compute a delta. For every surviving phase, keep its existing `label-key`, keep the `done` flag of every phase that remains, and keep `current` unchanged when that phase remains. Repeating the same replacement must leave the file byte-identical.
|
|
368
|
+
|
|
369
|
+
Use these replacements:
|
|
370
|
+
|
|
371
|
+
| Situation | Exact `steps` id order |
|
|
372
|
+
|---|---|
|
|
373
|
+
| Existing | `discovery, as-is, to-be, validate-to-be, architect-process, architect-solution, handoff` |
|
|
374
|
+
| New | `discovery, to-be, validate-to-be, architect-process, architect-solution, handoff` |
|
|
375
|
+
|
|
376
|
+
A project scaffolded before the Architect split carries a six-step file whose fifth step is `architect`, and the application never rewrites a project file that already has content, so these files persist for as long as those projects do. If the file you read still names `architect`, apply the matching replacement above before any other phase edit. Treat `architect` as the old name of `architect-process` when you apply the replacement: carry its `done` flag onto `architect-process` and give that step the new `label-key`, because this one step is renamed rather than surviving unchanged. If `current` was `architect`, it becomes `architect-process`. If `current` was `handoff`, read that step's `done` flag before deciding. An incomplete legacy `handoff` becomes `architect-solution`: it was entered as soon as the PDD was delivered, so it records that delivery rather than a hand-off the user reached, and solution design is the phase that now follows it. A legacy `handoff` already marked done is a finished project — keep `current: handoff` and mark `architect-solution` done as well, because that terminal completion attested the SDD had been surfaced and the user had confirmed the hand-off. Do not inspect the project for an SDD when deciding either case — the rule against repairing phase state from artifacts holds here too.
|
|
377
|
+
|
|
378
|
+
Enter `architect-process` when Phase 3 begins and keep it current through PDD generation. Surfacing the promoted or edited PDD dispatches `AskUserFeedback` with `checkpoint: 'ba.pddGenerated'` and nothing more: that checkpoint reports an artifact, it never advances a phase. A PDD close the user never follows up leaves `architect-process` current.
|
|
379
|
+
|
|
380
|
+
Enter `architect-solution` when Phase 4 opens on the user's ask, in the same edit that marks `architect-process` done. The work itself is "Solution design (Phase 4)" below — entering the phase moves the panel, never the work.
|
|
381
|
+
|
|
382
|
+
Enter `handoff` once the generated SDD has been surfaced, on the same turn Phase 4 below fires `ba.sddGenerated` — that dispatch belongs to Phase 4, never repeated here. Use the normal single transition edit to mark `architect-solution` done and set `current: handoff`.
|
|
383
|
+
|
|
384
|
+
`handoff` has no successor. Mark it done only after the user has explicitly confirmed the handoff. This terminal completion is the only standalone `done` write; every other `done` update belongs to the single edit that enters its successor.
|
|
385
|
+
|
|
386
|
+
You own `steps`, `current`, and `done`. Do not edit anything else in the file. After scaffolding, write phase state without waiting to be asked; the application does not advance or repair it.
|
|
387
|
+
|
|
388
|
+
## Phase Router
|
|
389
|
+
|
|
390
|
+
Read the phase MD that matches the current phase:
|
|
391
|
+
|
|
392
|
+
| Current phase | Read | What this phase produces |
|
|
393
|
+
|---|---|---|
|
|
394
|
+
| Any interview turn | `interview-spine.md` once per session | adaptive question behavior and shape checkpoints |
|
|
395
|
+
| Any diagram authoring | `diagram-authoring.md` once per session | shared Mermaid subset, semantic classes, source audit |
|
|
396
|
+
| Research pass (the tool confirmed `"research-accepted"`) | `research-pass.md` (this skill directory) | `{WikiPath}/research-pass.md` — the run's gate state and vetted findings, entering the interview as confirmation probes |
|
|
397
|
+
| Phase 1 (AS-IS) | `as-is-capture.md` (+ `industry-gaps.md` when it's gap-pass time) | populated `entities/`, `as-is/` |
|
|
398
|
+
| Phase 2 (RPA TO-BE) | `to-be-design.md` | populated `to-be/` |
|
|
399
|
+
| Phase 2 (Case TO-BE) | `case-to-be-design.md` | populated `to-be/` |
|
|
400
|
+
| Phase 2 (BPMN TO-BE) | `bpmn-to-be-design.md` | populated `to-be/` |
|
|
401
|
+
| Phase 3 (PDD) | `pdd-generation.md` (+ `pdd-template.md`) | `<process>-PDD.md` living draft, composed as part files and assembled — never written whole (host Download produces the `.docx`) |
|
|
402
|
+
| Phase 4 (Solution design) | **Solution design (Phase 4)** below — no separate phase file | `<process>-SDD.md` living draft authored by `uipath-planner` (host Download produces the `.docx`) |
|
|
403
|
+
|
|
404
|
+
## Solution design (Phase 4)
|
|
405
|
+
|
|
406
|
+
Business Analysis does **not** author Solution Design Documents. `uipath-planner` owns SDD authoring; BA hands it the confirmed design, runs it in this session, and registers the result so the solution-design panel can render it. There is no BA SDD template and no BA assembly script — never reconstruct them.
|
|
407
|
+
|
|
408
|
+
From this phase onward, UiPath product names and design checkpoints are user-visible. Internal mechanics (file paths, tool names, wiki structure) stay hidden.
|
|
409
|
+
|
|
410
|
+
### Entry
|
|
411
|
+
|
|
412
|
+
Phase 4 starts only when the user asks for it ("design the solution", "build the SDD", "generate the Solution Design Document"). At PDD close it is offered as the first ```sf chip, never as an extra question.
|
|
413
|
+
|
|
414
|
+
Run these checks in order and stop at the first failure. Each failure is one plain line; write nothing and register nothing.
|
|
415
|
+
|
|
416
|
+
1. **The process document exists.** Prefer `{DraftsDir}/<process-slug>-PDD.md`; otherwise `LsDirectory` `{DraftsDir}` and take the `*-PDD.md` whose marker or H1 matches this process, asking the user which one if several match.
|
|
417
|
+
> "A solution design needs the process document first. Let's finish that, then I can turn it into the solution design."
|
|
418
|
+
2. **`uipath-planner` is available** in this session's skills. A packaged Delegate bundles the UiPath skills plugin, so the usual cause is that its skills were switched off rather than that anything needs installing. Tell the two apart before naming a remedy: take the skills-folder path of any other `uipath-*` skill in this session and `LsDirectory` its parent.
|
|
419
|
+
- A `uipath-planner` folder is there but the skill is not in this session — switched off:
|
|
420
|
+
> "The UiPath skills plugin is installed, but its solution-design skill is switched off. Turn `uipath-planner` back on in Add-ons & Integrations → Skills, then ask me again."
|
|
421
|
+
- No such folder — this build does not carry the plugin (a development build, or a packaging step that did not run):
|
|
422
|
+
> "This build doesn't include the UiPath skills plugin, so I can't design the solution here. Install it from the Store, or use a packaged Delegate build."
|
|
423
|
+
|
|
424
|
+
Never improvise SDD architecture in planner's place, and never hand-write a solution design as a fallback.
|
|
425
|
+
|
|
426
|
+
### Route
|
|
427
|
+
|
|
428
|
+
Planner needs the design's shape. Take the first that resolves:
|
|
429
|
+
|
|
430
|
+
1. an explicit `Route:` line in `{WikiPath}/index.md`;
|
|
431
|
+
2. `{WikiPath}/to-be/process.bpmn` exists → BPMN;
|
|
432
|
+
3. `{WikiPath}/to-be/diagram.md` fence carries `diagramFamily=case-design` → Case, `diagramFamily=generic` → RPA.
|
|
433
|
+
|
|
434
|
+
Derive it from the artifacts, never from memory of the conversation: a later session has no recollection of the earlier route choice, and a wrong hint sends planner into the wrong design lane.
|
|
435
|
+
|
|
436
|
+
|
|
437
|
+
### Run planner in this session
|
|
438
|
+
|
|
439
|
+
Run planner **inline in this session** — never through `Agent`. Delegation exists, and the host's own contract excludes exactly this work: an agent run must not do user interaction, document authoring, or planning, and a planner design run is all three. Its design review is a mandatory stop for the user (see the constraint below), and a delegated run cannot hold one. Read `uipath-planner`'s `SKILL.md`, then the guide files it names for the selected lane, and follow them inline here, translating tools:
|
|
440
|
+
|
|
441
|
+
| Planner instruction | Delegate equivalent |
|
|
442
|
+
|---|---|
|
|
443
|
+
| `AskUserQuestion` (numbered options) | `AskUser` with `options:` |
|
|
444
|
+
| `EnterPlanMode` / `ExitPlanMode` | present the checkpoint in chat; wait for an explicit yes. Delegate's own `EnterPlan` / `ExitPlan` stay unused here — this skill does not enter plan mode |
|
|
445
|
+
| `Agent` (spawn a subagent) | run that lane inline yourself — delegating it is forbidden here, not unavailable |
|
|
446
|
+
| `Bash` / shell | `ExecuteBashCommand` (macOS) / `ExecutePowershellCommand` (Windows), per `PlatformContextType` `os.name` |
|
|
447
|
+
| `Read` / `Write` / `Glob` / `Grep` | `ReadFile` / `WriteFile` / `LsDirectory` + targeted reads |
|
|
448
|
+
| `TaskCreate` / `TodoWrite` | rows in the conversation task list at `ConversationTaskListPath`, per the `project-task-list` skill — never the project checklist |
|
|
449
|
+
|
|
450
|
+
Phase 4 is the longest tool-activity stretch in this skill, so it is where context compaction lands. If planner's rules — or this section's — are no longer verbatim in context, re-read them before continuing: the same-session don't-re-read rule exists to avoid redundant reads, never to make you continue a phase from a summary's paraphrase of its mechanics.
|
|
451
|
+
|
|
452
|
+
Pin planner to the settled design. It grounds, selects products, and renders; it never redesigns and never re-interviews:
|
|
453
|
+
|
|
454
|
+
- **Input document:** the PDD found at entry. Planner's own entry guard reads a PDD, so it enters its design phase.
|
|
455
|
+
- **Confirmed facts:** `{WikiPath}/index.md`, `{WikiPath}/to-be/`, `{WikiPath}/entities/`, and `{WikiPath}/process-knowledge.md` when present. Wiki facts win over planner reference content.
|
|
456
|
+
- **Route hint:** the route resolved above.
|
|
457
|
+
- **Output path:** `{DraftsDir}/<process-slug>-SDD.source.md`. That file is planner's own and stays byte-identical: planner's template-conformance gate passed against it, so nothing may edit it afterwards. The document the user sees is a separate projection Delegate writes at `{DraftsDir}/<process-slug>-SDD.md`. When the user names an output path, that path is the projection and planner's source sits beside it with `.source.md` in place of `.md`.
|
|
458
|
+
- **Skip every planner question the wiki already answers** (stages, owners, systems, task types, SLAs, exceptions). Ask only what it genuinely cannot know — delivery model, execution autonomy, tenant resource picks — through one `AskUser` with finite options.
|
|
459
|
+
- **The user is present — this is the constraint that rules out delegation.** Never run planner through `Agent`, and never auto-approve its gates. Present its design review with its own structure intact — snapshot, data contract, stages and tasks, other paths, resources, decisions, ⚠ flags — and take the confirmation through one `AskUser` (proceed / change something). A correction to a confirmed process fact goes wiki-first and then re-enters this step; a design-level correction is a targeted planner edit.
|
|
460
|
+
- **Tenant grounding is real work, where the host can do it.** When the design references tenant-bound resources and this host has a real shell, run planner's own resolution chain through the shell tools and never record an identity you did not read back. No CLI, no session, or a host that emulates the shell instead of running one is planner's degraded path. Never block the solution design on tenant access.
|
|
461
|
+
|
|
462
|
+
Planner writes the file. Confirm it exists afterwards and carry the path it actually wrote into `readFrom` below. Leave its `## Planner Handoff` header and `<!-- planner-handoff:v1 -->` marker in place.
|
|
463
|
+
|
|
464
|
+
### Register it
|
|
465
|
+
|
|
466
|
+
Call `GenerateLivingDocument` on planner's file. `decorateFrom` tells Delegate what it is registering; Delegate reads planner's file, writes the document panel's copy itself, and leaves planner's file untouched — never hand-write Delegate markers, and never edit planner content to make room for them. Send this profile verbatim, filling only the three paths:
|
|
467
|
+
|
|
468
|
+
Pass paths unchanged. `{DraftsDir}` is absolute; a user-chosen project folder lives outside the user-data root, so rewriting to a user-data-relative form yields `../../../..` and names the wrong file.
|
|
469
|
+
|
|
470
|
+
```json
|
|
471
|
+
{
|
|
472
|
+
"draftPath": "<{DraftsDir}/<process-slug>-SDD.md, exactly as {DraftsDir} is given>",
|
|
473
|
+
"replace": true,
|
|
474
|
+
"decorateFrom": {
|
|
475
|
+
"sourceName": "uipath-planner",
|
|
476
|
+
"readFrom": "<the path planner actually wrote, exactly as it was given>",
|
|
477
|
+
"requireAny": ["<!-- planner-handoff:v1 -->", "## Planner Handoff"],
|
|
478
|
+
"meta": {
|
|
479
|
+
"type": "sdd",
|
|
480
|
+
"label": "Solution Design Document",
|
|
481
|
+
"icon": "FaDiagramProject",
|
|
482
|
+
"metadataFields": ["client", "author", "date", "version"]
|
|
483
|
+
},
|
|
484
|
+
"skipHeadings": [
|
|
485
|
+
"Document History",
|
|
486
|
+
"Planner Handoff",
|
|
487
|
+
"Decisions Made",
|
|
488
|
+
"Recommended Scope",
|
|
489
|
+
"Action Required — SME Review Items"
|
|
490
|
+
],
|
|
491
|
+
"dropSections": ["Table of Contents"],
|
|
492
|
+
"blockingRow": { "label": "Template validation", "blockedValues": ["pending"] },
|
|
493
|
+
"inheritCoverFrom": "<the PDD path found at entry, exactly as it was given>"
|
|
494
|
+
}
|
|
495
|
+
}
|
|
496
|
+
```
|
|
497
|
+
|
|
498
|
+
What each part is for, so a template change upstream lands in the right field:
|
|
499
|
+
|
|
500
|
+
- `readFrom` is what keeps planner's file byte-identical while the panel's copy is free to differ. Without it Delegate would decorate planner's own file in place.
|
|
501
|
+
- `skipHeadings` is planner's front matter — the headings its templates place before the first numbered section. They stay in the document and stay out of the sidebar, so the sidebar matches the count planner's own table of contents states. Add a heading here when a template gains one; the list is this skill's to keep current.
|
|
502
|
+
- `dropSections` omits the authored table of contents from the panel's copy only. The host synthesizes the table of contents for every downloaded document, so keeping planner's as well would put two in the `.docx`. Planner's file still has it.
|
|
503
|
+
- `blockingRow` is the finished/unfinished gate, and it reads `Template validation`, not `Status`. Planner writes both as `draft` / `pending` first and sets `Template validation: passed` once its completeness check passes, so an interrupted run leaves `pending` on disk and Delegate refuses to announce a half-finished design as the user's solution. `Status` answers a different question — may tasks be derived — and stays `draft` on a complete design while a blocking review item is open. That design registers: the user can read and review it, and its open items are planner's `Action Required — SME Review Items` table. A document with no such row registers — an older SDD predates the row.
|
|
504
|
+
- `inheritCoverFrom` carries client, author, date, and version across from the process document, so the user is not asked twice.
|
|
505
|
+
- `replace: true` is what makes regeneration work when a solution design is already registered. It re-registers **the same** `draftPath`, so keep the projection path stable across regenerations. Replacing also clears the previous review state: the content the user marked reviewed is gone, so the badge goes with it.
|
|
506
|
+
|
|
507
|
+
A refusal saying the draft carries none of the authoring signals, or that its `Template validation` row still reads `pending`, means planner did not finish: say so and stop — do not repair the file by hand.
|
|
508
|
+
|
|
509
|
+
**What the solution design offers in the panel.** `refine-section`, `mark-reviewed` and `download` — the same three the process document has. **The profile deliberately sends no `actions` field**, exactly as the process document sends none: an absent list means the whole catalog, so the panel gets all three and `open-diagram` drops itself because a solution design declares no diagram kinds. Never add an `actions` field here to "be explicit". A hand-written list is a literal that has to be re-copied every time the set changes, and naming a subset is how per-section refine gets switched off by accident: the host reads that one list both for whether a `[Refine section]` link is emitted and for whether it is clickable, so a stale two-entry list silently produces a document with no refine affordance at all and no error anywhere.
|
|
510
|
+
|
|
511
|
+
Its anchors still carry no `sources=`, so no section ever shows a drift badge — provenance is what that signal is computed from, and a delegated document has none. Open review items read as planner's own `Action Required — SME Review Items` table rather than per-section gap chips.
|
|
512
|
+
|
|
513
|
+
On success, call `AskUserFeedback` with `checkpoint: 'ba.sddGenerated'` (one non-blocking call; do not announce it or wait for a rating; skip silently if unavailable), then tell the user the solution design is ready in the document panel and name any ⚠ flags or open review items it carries. When planner's handoff still reads `Status: draft`, say the design is ready to review and name its blocking items — answering them is what lets planner flip it to `ready`.
|
|
514
|
+
|
|
515
|
+
This turn first delivers the SDD, so its second ```sf follow-up suggestion is `Publish this SDD to a business process in Automation Hub` — the first stays the flow's next step. Silence conditions: "Automation Hub publishing" below; a publish request → read `automation-hub-publishing.md` in full.
|
|
516
|
+
|
|
517
|
+
That first ```sf follow-up suggestion is `Refine this SDD`. Phase 4 ends with the delivered SDD: it never offers or starts a build.
|
|
518
|
+
|
|
519
|
+
### Client SDD in the customer's Word template
|
|
520
|
+
|
|
521
|
+
When `DocTemplates.sdd` in your context is `client-word`, the user also wants the client-facing SDD/ASDD, in their own Word template. Produce it right after a successful registration, before saying the design is ready, by following planner's ASDD crosswalk guide (`references/asdd-crosswalk-guide.md` in its skill folder) inline:
|
|
522
|
+
|
|
523
|
+
- **Template path:** the `SDD template:` line in `{WikiPath}/index.md`. When it is absent, ask once through `AskUser` for the `.docx` path, then write it there so a regeneration reuses it.
|
|
524
|
+
- **Input and output:** planner's file (`readFrom`), never the panel's copy. Assemble into `{DraftsDir}/<process-slug>-SDD-client.md`, then convert it beside it to `<process-slug>-SDD-client.docx` with planner's `sdd-to-docx` script (`.ps1` on Windows), passing the template as `--reference-doc`.
|
|
525
|
+
- **Conversion fails** (exit 2 means pandoc is missing): keep the assembled `.md`, tell the user in one line what the conversion needs, and carry on. Never hand-build a `.docx`.
|
|
526
|
+
- Name the client document in the ready message. It is a by-product: the registered solution design stays planner's projection.
|
|
527
|
+
|
|
528
|
+
`standard` or no value: skip this; planner produces the client document only when the user asks for it.
|
|
529
|
+
|
|
530
|
+
### Hard rules
|
|
531
|
+
|
|
532
|
+
- **Planner owns SDD authoring.** BA never writes SDD sections, architecture decisions, task tables, planner handoff headers, or case blueprint content.
|
|
533
|
+
- **Delegate markers belong to the host**, added by `GenerateLivingDocument` from the `decorateFrom` profile above — never by hand, never by a script. The profile is this skill's to send and to keep current; the marker mechanics are the host's.
|
|
534
|
+
- **Planner's file is planner's to author.** Nothing in this phase edits it on its own initiative — not to add markers, not to tidy it, not to fix a heading, not to make room for anything. The one write it accepts is **content the user asked for in a named section**, per "SDD refinement": their edit has to reach this file or it would keep a design they rejected, and the host's inline pencil writes it there for the same reason. Everything else Delegate writes goes to the projection.
|
|
535
|
+
- **The `.docx` comes from host Download, here as in Phase 3.** Never run a converter yourself: not Delegate's (`/word/*`, `/doc-gen`, word-architect) and not planner's `scripts/sdd-to-docx.{sh,ps1}`, which is the converter actually resident in context during this phase. The body deliberately omits cover, Document Control, and the table of contents because the host synthesizes them, so a converted file hands the client a document missing all three, with live `delegate:refine-section` links left in it. Stop-check: about to produce a `.docx` → STOP, the Download action does it.
|
|
536
|
+
- **The PDD is the handoff artifact.** The solution design always goes through planner, from the PDD — never from raw TO-BE notes.
|
|
537
|
+
- **Design settled, no re-interview.** Corrections to confirmed process facts go wiki-first, then re-enter the planner step.
|
|
538
|
+
- **Explicit request only.** Never enter Phase 4 automatically at PDD close — the chip offers it, the user chooses it.
|
|
539
|
+
|
|
540
|
+
## Automation Hub publishing
|
|
541
|
+
|
|
542
|
+
Automation Hub is the system of record for an approved process. Publishing is opt-in: the user asks for it, every time.
|
|
543
|
+
|
|
544
|
+
### When to raise it
|
|
545
|
+
|
|
546
|
+
Conversation Economy governs this: the offer rides a ```sf follow-up chip, so it costs no checkpoint, no `AskUser`, and no prose.
|
|
547
|
+
|
|
548
|
+
On the turn a process document is first delivered (PDD or SDD), emit two ```sf suggestions. The first stays the flow's next step. The second is the offer, in the user's own voice, naming the type just delivered — `Publish this PDD to a business process in Automation Hub` after a PDD, `Publish this SDD to a business process in Automation Hub` after an SDD. That second chip is the whole offer: a click arrives as a user request, and that request is itself the approval; a later request to publish carries the same approval, so act on it immediately.
|
|
549
|
+
|
|
550
|
+
Raise it on that turn only. Stay silent while:
|
|
551
|
+
|
|
552
|
+
- Generation is in flight (a `.<process-slug>-PDD.parts/` dir exists in `{DraftsDir}`).
|
|
553
|
+
- The turn is an edit, re-sync, or open-item resolution — a changed document is not a new reason to raise it.
|
|
554
|
+
- A spoken decline stands from this session. An ignored chip is not a decline; a spoken one holds until the user raises publishing themselves.
|
|
555
|
+
|
|
556
|
+
`mark-reviewed` is host-local and dispatches no agent turn, so the moment the user accepts a document never reaches you — the delivery turn is the only one you get.
|
|
557
|
+
|
|
558
|
+
### Handing off
|
|
559
|
+
|
|
560
|
+
A publish request — a chip click or the user's own words — is the approval. Read `automation-hub-publishing.md` in full and follow it; it owns the skill activation, the file pick, the pass list, and the report.
|
|
561
|
+
|
|
562
|
+
## Living draft editing
|
|
563
|
+
|
|
564
|
+
These rules apply on **any turn where a living draft already exists** — a PDD (`{DraftsDir}/<process-slug>-PDD.md`), a PDD still staged as parts (`{DraftsDir}/.<process-slug>-PDD.parts/`, see "Generation in flight"), an SDD (`{DraftsDir}/<process-slug>-SDD.md`), or any combination. They govern every refine, edit, re-sync, and open-item turn for either document. `{DraftsDir}` is the absolute drafts dir from context; use that value verbatim (do not derive it from the wiki path). The target draft is the `*-PDD.md` or `*-SDD.md` for the CURRENT process; a `refine-section` request names a section id — match it to the document whose anchors declare that id. If the slug is not already known, `LsDirectory` `{DraftsDir}` and pick the file whose `doc:meta process=` or legacy `pdd:doc process=` matches this process. If several drafts of the same type exist and the right one is ambiguous, ask the user instead of guessing (the host otherwise just edits the lexicographically-first file). Apply these rules without reading `pdd-generation.md` — that file is loaded only at first-gen. The rules here are authoritative; `pdd-generation.md` cross-references them.
|
|
565
|
+
|
|
566
|
+
**A PDD draft with no `.<process-slug>-PDD.assembled` receipt beside it was written whole, not assembled.** The assembly script writes that file after it promotes the draft, from the parts it promoted; an author never produces it. Refine, re-sync and open-item turns splice anchored sections into the draft in place and re-lift their content from the wiki, so they do not need the parts -- but a draft written whole may lack the anchors, the order or the section set they splice against. Compose the parts and run the assembly first, then edit.
|
|
567
|
+
|
|
568
|
+
**A refinement writes exactly two path roots**: wiki pages under `{WikiPath}` and living drafts under `{DraftsDir}`. There is one further permitted target, and it is a single file rather than a root: **the solution-design source planner wrote**, at the exact path carried in `readFrom`, and only for a user-asked content change to a named section, per "SDD refinement". Not its directory, not its siblings, not anything else planner produced. Every other path is out of bounds — above all this skill's own instruction files (`pdd-template.md`, the phase guides, `scripts/`, `references/`, `assets/`), which are `ReadFile` only: never `WriteFile`, `EditFile`, or `DeleteFile` one of them. Stop-check before every write on an edit turn: resolve the absolute target path; not under `{WikiPath}` or `{DraftsDir}`, and not that one recorded `readFrom` file → STOP and re-target. "Refine section X" never means "edit the template that describes section X"; a template's example anchors and `[Refine section]` lines are illustrations, not splice targets.
|
|
569
|
+
|
|
570
|
+
Namespaces: PDD drafts use `doc:*` anchors (`doc:meta`, `doc:section`). A pre-existing draft may carry legacy `pdd:*` anchors (`pdd:doc`, `pdd:section`); the parser reads both. At the start of a living-draft operation, read the draft and detect the draft namespace once. Use that namespace for the rest of the operation; never mix namespaces within one document.
|
|
571
|
+
|
|
572
|
+
Re-sync and edit turns re-project the draft `.md` — re-emit the matching `doc:section` or `pdd:section` anchor with an advanced `reproj=` token. The `.docx` is a disposable export the host regenerates at Download; editing it never clears `behind` and is overwritten on the next Download. So the `.md` is the only re-sync/edit target: never open or modify the `.docx` (or any binary doc) — no Word/Office tools, no `/word/*` (open, insert, delete, replace), no `/doc-gen`, no word-architect. Stop-check: about to touch a `.docx` for a re-sync/edit → STOP, `EditFile` the `.md` section block and advance its `reproj=` token instead.
|
|
573
|
+
|
|
574
|
+
After surfacing an edit or re-sync result that changed the **PDD** draft, call `AskUserFeedback` with `checkpoint: 'ba.pddGenerated'` (one non-blocking call; do not announce it or wait for a rating; skip silently if the tool is unavailable). An edited draft is a new revision; the prompt cooldown absorbs rapid edit rounds.
|
|
575
|
+
|
|
576
|
+
### Refinement cascade
|
|
577
|
+
|
|
578
|
+
Any edit, refine, or open-item turn runs every leg in one turn. No extra confirmation gate — the user's edit is the approval. Never ask whether to also update the map or the documents.
|
|
579
|
+
|
|
580
|
+
1. **Decompose** — map the request to its wiki fact(s). Not decomposable → reject per "Decompose or reject".
|
|
581
|
+
2. **Facts and ledgers** — one write round: every affected wiki source page, plus the confirmed stage/task ledger when the edit adds, renames, removes, splits, merges, or reorders a stage, task, or step. A case-capture stage change re-keys `as-is/attachments.md` in the SAME change. A ledger the map contradicts is the defect, not the map.
|
|
582
|
+
3. **TaskList** — synchronize every affected FACTS/ANALYSIS row as the next mutation (see Synchronization boundary). No diagram and no splice until its edits succeed.
|
|
583
|
+
4. **Model** — redraw the affected state's canonical model, then validate it as the next tool call (see Diagram leg).
|
|
584
|
+
5. **Documents** — re-project every affected section: PDD first, then SDD.
|
|
585
|
+
6. **Close** — `AskUserFeedback` per the rule above, then one line naming the pages changed, the map redrawn, and the sections that moved.
|
|
586
|
+
|
|
587
|
+
**The affected set comes from the draft, never from memory.** After the wiki writes, read each draft: a section is affected iff a page you wrote this turn appears in its anchor's `sources=` list. Re-project exactly those; every other section stays byte-identical. This bounds the cascade — a refinement never regenerates a document. Two additions: a section embedding a model you redrew (`current-state` for AS-IS, `target-state` for TO-BE); a self-computed section whose inputs moved (PDD `appendix-a`; SDD `appendix-a-readiness` and `appendix-b-lineage`).
|
|
588
|
+
|
|
589
|
+
**Partial cascade is recoverable, not corrupt.** Wiki-before-document ordering leaves the wiki ahead of the documents on any failure — the one direction the host's `behind` signal recovers from. Map validated but a splice failed → that section stays `behind` and is picked up next turn. Do NOT add rollback; report what moved and what did not.
|
|
590
|
+
|
|
591
|
+
**Idempotency.** One write round per page, one redraw-and-validate per state, exactly one `reproj=` advance per section per turn. A section affected for several reasons at once (`behind` + edited + its model redrawn) is re-projected once, last, from the final wiki state.
|
|
592
|
+
|
|
593
|
+
### Diagram leg
|
|
594
|
+
|
|
595
|
+
Redraw iff the edit changes a **map-visible fact**: a step, activity, or task and its order; a decision or one of its labelled branches; an exception or rework route; a terminal outcome; the stage set or a stage's name or grouping; a participant or lane; an event or a message/timer wait; a node's owner, system, mode, or `case-task-*` class.
|
|
596
|
+
|
|
597
|
+
No redraw when the edit changes only what the map never shows: volumes, cycle time, SLA values, rule bodies, KPI targets, compliance text, personas or RACI detail, glossary, assumptions, risks, next steps, screenshots, framing prose. Mirror of the existing carve-out: a diagram-only correction that changes no business fact needs no task list mutation; a fact edit that changes no map-visible fact needs no redraw.
|
|
598
|
+
|
|
599
|
+
Redraw the edited state only — never both states for a one-state edit. Canonical file per route: the Diagram artifacts hard rule already names it. Exception: a user-supplied `as-is/process.bpmn` is the AS-IS model of record and is never altered (`as-is/overview.md` records it) — write the fact, record the divergence in `queries/open-questions.md`, and say so in the close line.
|
|
600
|
+
|
|
601
|
+
`diagram-authoring.md` not read this session → `ReadFile` it first (BPMN: `bpmn-to-be-design.md`'s profile and validation section). **The next tool call after the model write is its validator**: Mermaid lint until `DIAGRAMLINT:ok`, or XML parse plus reference resolution for BPMN. Loop repair-and-revalidate — a successful write is not evidence the map renders. No document splice and no user-facing message while the model is unvalidated.
|
|
602
|
+
|
|
603
|
+
Then re-project the embedding section and re-check parity: the spliced Mermaid block's `diagramFamily` and body match the canonical fragment exactly, or the fenced `bpmn` block's XML bytes equal the canonical `.bpmn`. Keep the figure caption and diagram link. Never redraw a model from memory or paste model source into chat.
|
|
604
|
+
|
|
605
|
+
### SDD refinement
|
|
606
|
+
|
|
607
|
+
The solution design refines section-by-section, but by a different mechanism than the PDD, and the cascade above does not apply to it. `uipath-planner` authors it and Delegate registers a projection; its anchors carry no `sources=`, so there is no per-section wiki source to re-derive a section from. What replaces the cascade is a single question about the change the user asked for:
|
|
608
|
+
|
|
609
|
+
- **A process fact moved** (a stage, task, actor, system, rule, SLA, volume — anything the PDD also states): the fact is upstream, so fix it there. Write the wiki source, run the Refinement cascade for the PDD as usual, then **regenerate the solution design**: re-run "Solution design (Phase 4)" above, which re-registers the same path with `replace: true`. Say that the design was regenerated and that the previous review no longer stands — `replace: true` clears it, so the user accepts the new design rather than inheriting the old acceptance.
|
|
610
|
+
- **Nothing factual moved** and the change is confined to the named section (wording, ordering, an expanded abbreviation, a clarification the design already implies): apply it to that section of planner's file and re-register. **Every other section stays byte-identical**, and no wiki write happens — there is no fact to write. This is the same act the user performs with the pencil; you are doing it because they asked in words instead of typing.
|
|
611
|
+
|
|
612
|
+
Decide from what the user asked, not from which is cheaper. When both are true — a fact moved *and* they also want the section reworded — the fact leads: regenerate, then check whether the wording is still wanted against the regenerated text.
|
|
613
|
+
|
|
614
|
+
`replace: true` never survives a user's own edits to the sections it rewrites. The host records every section the user edited with the pencil in an **edit ledger** beside the draft — `{DraftsDir}/<process-slug>-SDD.edits.json`, shape `{"version": 1, "editedAnchors": ["<anchor-id>", …]}`, written by the host and `ReadFile`-only for you. Before regenerating, read it; if it names sections, name those sections and say the regeneration will discard the user's edits to them **before** doing it, so they can decline. An absent or empty ledger means there is nothing to warn about. Never write, tidy, or delete this file.
|
|
615
|
+
|
|
616
|
+
Appendices, section numbering, template conformance, and heading structure remain planner's. Never edit the SDD to fix those, and never add or move a Delegate marker in either file.
|
|
617
|
+
|
|
618
|
+
On a turn that regenerates the SDD, fire `AskUserFeedback` with `checkpoint: 'ba.sddGenerated'`. Use `ba.pddGenerated` only when the PDD itself moved this turn; the two are separate registered checkpoints and neither substitutes for the other.
|
|
619
|
+
|
|
620
|
+
### Generation in flight (a `.<process-slug>-PDD.parts/` dir exists in `{DraftsDir}`)
|
|
621
|
+
|
|
622
|
+
A parts dir holds sections composed earlier: by this run before a compaction, or by a session that ended before the set was complete. Always `ReadFile` `pdd-generation.md` and `pdd-template.md` to restore the full assembly contract and the canonical anchored section-id list. Then resume from disk, not from this run's history: `LsDirectory` the parts dir to see which `NN-<id>` files exist, keep every valid part, and write only the missing or malformed ones (`pdd-generation.md` items 1 and 5). Clear the dir only when it is foreign. A stale part is no reason to clear: the script refuses any part whose wiki source changed after the part was written (exit `3`). Never overwrite a part blind. The concat-and-promote script (in `pdd-generation.md`) does the byte-concat and promote. The `.<process-slug>-PDD.md.tmp` file exists only transiently inside that script and is never a model write target.
|
|
623
|
+
|
|
624
|
+
**A wiki write in this state obliges a re-projection.** The staged parts ARE the PDD, so the cascade's document leg applies to them: write the wiki source first, then re-compose in the SAME turn every part whose anchor `sources=` names a page you wrote — one whole `WriteFile` per part, `reproj=` advanced, every other part byte-identical. Assembling a part that is older than a source it projects ships a document contradicting the wiki (session ca6518d8: a mid-generation correction landed in the wiki, 16 stale parts assembled anyway). The script refuses it — exit `3` with a `STALEPART: <part> <- <source>` line per stale part, nothing touched — so re-compose the named parts and re-run rather than working around it. Mechanics: `pdd-generation.md` "Stale parts".
|
|
625
|
+
|
|
626
|
+
### Chat-only content editing
|
|
627
|
+
|
|
628
|
+
A user content edit routes through chat. Use the section's provenance — the matching `doc:section sources=` or `pdd:section sources=` anchor in the draft `.md` plus the read-only source map in this skill's own `pdd-template.md` — to locate the correct wiki page(s). The SDD has no such map and no `sources=`: route a content change there through "SDD refinement" above, which decides between fixing an upstream fact and regenerating, or editing the named section of planner's file in place. Write the change to the wiki source(s), then run the Refinement cascade above **in the same turn**. Unsupported-marker cleanup follows "Resolving open items" below. Surface only the result.
|
|
629
|
+
|
|
630
|
+
### Resolving open items
|
|
631
|
+
|
|
632
|
+
- Conclusive answer → write the fact to the wiki, sync the task list, then run the Refinement cascade (the map redraws when the fact is map-visible). Write the value as ordinary content; remove its `==TBD: reason==` markers.
|
|
633
|
+
- Status markers do not exist. Never emit `==Resolved: ...==`, `==Confirmed: ...==`, or `==Done: ...==`.
|
|
634
|
+
- `Resolved`/`Confirmed`/`Done` marker whose value matches the wiki → re-project only its section. Skip wiki and task list writes. Leave unrelated sections unchanged.
|
|
635
|
+
- Partial answer → write the known value; keep one focused standalone `==TBD: reason==` marker for the remaining gap.
|
|
636
|
+
- Answer conflicts with a confirmed fact → ask before changing the wiki or draft.
|
|
637
|
+
|
|
638
|
+
### Same-turn, no pending state
|
|
639
|
+
|
|
640
|
+
Source-changing edit → write the wiki first, then re-project from it in the same turn. Unsupported-marker cleanup above → re-project from the current wiki without writing it. Never leave the wiki behind the document. If the document status was `completed`, the edit reverts it to `draft` (the host reconcile clears `completed` on the next turn-end; the skill does not set status directly). Re-projecting = **`EditFile`-splice** that one section's block in place with an **advanced `reproj=` token** (the orchestrator computes it: `g<N>` → `g<N+1>`; one advance per section per turn), leaving every other section byte-identical. The persisted token is the host's proof (it compares the in-file token against its sidecar); a splice not written to disk clears nothing. **Splice contract (apply without loading `pdd-generation.md`):** `old_content` = the section's current block (its matching `doc:section id=` or `pdd:section id=` anchor line through its `[Refine section]` link — the unique id makes `old_content` unique); `new_content` = that block re-projected with the advanced token, no leading/trailing blank lines so the `\n\n` separators between sections stay intact. On an `old_content` mismatch, re-read and retry once. Large block (`target-state`, `current-state`, any section a whole-block `old_content` will not match byte-for-byte) or the retry mismatched again → split the splice: `EditFile` the unique anchor line alone (advanced token, extended `sources=`), then after a fresh read `EditFile` only the changed subsection with a narrowly bounded `old_content`; the embedded map and every unrelated section stay untouched, and the section still gets one token advance. Re-read the edited section and next anchor. Literal `\n` or an inline next anchor → repair before continuing. Fuller mechanics: `pdd-generation.md` "Re-sync mechanics".
|
|
641
|
+
|
|
642
|
+
### Mark-reviewed (host-local, no agent turn)
|
|
643
|
+
|
|
644
|
+
`mark-reviewed` is handled entirely by the host: it accepts the current draft as the reviewed baseline, re-baselines `behind` sections to their current wiki sources in memory, and writes the sidecar `completed`. It does NOT dispatch an agent turn or ask you to re-project. You re-project only on the user-triggered re-sync and edit turns above (see "Chat-only content editing", "Same-turn, no pending state", "Edit a `behind` section"). A later wiki edit re-flags drift via reconcile, so the staleness signal survives.
|
|
645
|
+
|
|
646
|
+
### Decompose or reject
|
|
647
|
+
|
|
648
|
+
Decompose a narrative edit spanning multiple wiki sources to each source individually. An edit that cannot be decomposed to any wiki source is **rejected** with the explanation:
|
|
649
|
+
|
|
650
|
+
> "Edit the fact, the document follows. Please update the relevant wiki page directly."
|
|
651
|
+
|
|
652
|
+
Reject without any wiki or document change. Structure and formatting are template-owned and are not user-editable — reject the request; never satisfy it by editing a template file.
|
|
653
|
+
|
|
654
|
+
### Edit a `behind` section
|
|
655
|
+
|
|
656
|
+
If the target section's status is `behind` (its wiki source changed via a non-edit path): write BOTH the catch-up (current wiki state) AND the user's edit to the wiki source(s) first, then re-project the section **once** — a single `EditFile`-splice with a single token advance reflecting the final wiki state. Never re-project the same section twice in one turn, and never overwrite newer wiki state with a stale projection. A section that is both `behind` and edited is still one splice with one token advance, run last in the cascade.
|
|
657
|
+
|
|
658
|
+
### Success confirmation
|
|
659
|
+
|
|
660
|
+
On success, name the wiki page(s) changed so the user can verify. Name the map when it was redrawn and the document sections that moved, so the user can verify the whole cascade, not only the wiki.
|
|
661
|
+
|
|
662
|
+
### Cover metadata boundary
|
|
663
|
+
|
|
664
|
+
Editing `client`, `author`, `date`, or `version` is a **host-local viewer affordance** (sidecar-only). If a user asks in chat to change cover metadata, do NOT route the change to the wiki. Respond: "Edit cover fields directly on the document cover — hover a field and click the pencil. They're stored in the document sidecar, not the wiki." The only exception is the **process name** (wiki Domain rename), which IS a wiki edit and follows the chat-only content-editing routing above.
|
|
665
|
+
|
|
666
|
+
## Hard Rules
|
|
667
|
+
|
|
668
|
+
- **If the `process-discovery` skill is available, corpus requests follow it.** Requests about the discovery corpus (task-mining data, a process from a discovery scan) are not business-analysis work; business analysis picks up from its findings. A recording the user records or attaches follows `recording-processing.md`, not discovery.
|
|
669
|
+
- **The skill bundle is read-only.** Never `WriteFile`, `EditFile`, or `DeleteFile` one of this skill's own files — templates, phase guides, scripts, references, and assets are instructions, not work products. A refinement writes only wiki pages under `{WikiPath}` and living drafts under `{DraftsDir}`.
|
|
670
|
+
- **User-facing wording is business-only.** Every reply, including "let me..." preambles before tool calls; internal mechanics hidden always. The one file-path exception is the clickable output link returned after an explicit living-document export. Full guidance: User-Facing Persona.
|
|
671
|
+
- **The initial Process Knowledge retrieval is mandatory** before any artifact write once the process can be named (an opener that opts out of reference processes sends `"no-reference"` instead); further use is fit-judgment, not mandatory per phase, and never while the latch holds. See Process Knowledge Gate.
|
|
672
|
+
- **Source scope beats Process Knowledge.** Confirm source-bounded scope before TO-BE or PDD; never invent an end-to-end lifecycle the source bundle does not document.
|
|
673
|
+
- **Scope gate first.** The opener must ask to document, map, scope, or design a business process or UiPath automation/case before any read, write, retrieval, phase edit, or question. Building the thing itself (software, games, scripts, content) is out of scope: this skill does not run — Cartographer declines per the scope statement in its system prompt, elsewhere the request is handled without this skill. See Opening Discovery Router.
|
|
674
|
+
- **Opening Discovery Router is mandatory** — understand the user problem, determine existing vs new, infer BPMN vs Case vs RPA, and route with the matrix before entering AS-IS or TO-BE. Ask only when genuinely ambiguous, or when BPMN is likely but Case is also a defensible industry pattern. For existing work, the AS-IS capture format (case map vs BPMN model vs simple flow) is put to the user before the current-process map is drawn — the router inference supplies the recommendation, not the final answer (see `as-is-capture.md`). Do not silently turn a BPMN-shaped request into Case or a Case-shaped request into BPMN.
|
|
675
|
+
- **Existing work gets AS-IS; new work does not.** Follow the routing matrix exactly; never invent a current state for net-new work.
|
|
676
|
+
- **Case work produces both a PDD and an SDD.** The PDD remains the Phase 3 business-design artifact and the input `uipath-planner` works from when it authors the solution design in Phase 4.
|
|
677
|
+
- **BA never authors the solution design.** Phase 4 delegates SDD authoring to `uipath-planner` and registers what it wrote; BA writes no SDD content of its own.
|
|
678
|
+
- **Artifact ≠ confirmation.** Do not advance because an artifact exists. Advance only after explicit business-shape confirmation. A canonical map is presentable only after it passes the phase guide's read-back source audit — A successful write alone is not validation.
|
|
679
|
+
- **Project task list: BA owns the canonical task list content and FACTS / ANALYSIS row semantics. Use `project-task-list` for every task list read and write. After every completed factual wiki-write round, make task list synchronization the next mutation. Never write a diagram or advance the phase with stale affected rows. Treat FACTS, ANALYSIS, and PDD readiness as checkpoints.** See Project Task List.
|
|
680
|
+
- **A shape-changing fact write stales the route's diagram. If the user stated the change, regenerate the agent-authored artifact in the same refinement, unprompted — their instruction is the confirmation — and re-project the map-bearing section with it. From any other source (documents, Process Knowledge, inference), flag the impact and confirm before redrawing. A user-supplied `as-is/process.bpmn` is never agent-rewritten: request an updated model or confirmation to keep it.** See Project Task List.
|
|
681
|
+
- **Milestone feedback: call `AskUserFeedback` once per confirmed milestone**: AS-IS Process Snapshot confirmed → `checkpoint: 'ba.asIsCaptured'`; TO-BE Future Process Review confirmed → `'ba.toBeCreated'`; PDD promote succeeded or an edit/re-sync round surfaced → `'ba.pddGenerated'`. Non-blocking; do not announce it in chat or wait for a rating. Tool unavailable → skip silently. The AS-IS and TO-BE confirmations also stamp `verified` on the confirmed pages (see Page Trust Stamps).
|
|
682
|
+
- **A refinement cascades.** Any edit, refine, or open-item turn runs the full cascade in one turn: wiki facts and ledgers → task list → canonical model plus validation → every affected document section. Never update a document section without its wiki source, and never leave the process map showing a shape the wiki no longer says. See Living draft editing.
|
|
683
|
+
- **Use AskUser when the question has options, OR you're asking 2+ questions at once.
|
|
684
|
+
- **IMPORTANT: don't use AskUser for a single free-text question — just ask in chat.**
|
|
685
|
+
- **Every options question includes a final option: "Take your best assumption"** (except the screenshot and future-state questions, whose options are fixed). Selecting it means: decide per the best-assumption playbook, record the decision in the checkpoint's assumption ledger, and move forward — never re-ask. This is the user's fast-forward; it applies to the routing question, mode selection, and every other finite ask. **Only the user selects it.** Choosing it on their behalf is not a fast-forward, it is a skipped gate: you may never resolve an options question by taking your own best assumption instead of asking it. Recording a decision as `(best assumption)` is honest only for an answered question the user answered that way — if no question was put to them, the record is a false confirmation and the gate did not happen.
|
|
686
|
+
- **Interview/clarifying questions always carry the free-text branch.** Nothing appends one — add `{"kind":"userinput"}` to `options:`. The user's real answer is often outside your candidate list.
|
|
687
|
+
- **Max 2 questions per turn** by default. Single multi-question AskUser calls are exempt — the constraint is against interrogation, not batched questions. Summarize, don't interrogate: reflect understanding back; don't list 7 questions.
|
|
688
|
+
- **A phase closes on its own output, not on its argument.** Each phase guide opens with the set of files it produces and closes with the sequence that writes, validates and verifies them — run that sequence before the checkpoint. A review presented from the conversation with the pages still unwritten has closed nothing, and the readiness gate one phase later is too late to repair it.
|
|
689
|
+
- **Write each confirmed fact to its page in the turn it is confirmed:** one page, one `WriteFile`, composed from that page's own facts. Never produce page or part bodies from a shell loop, a heredoc, a `printf`, or a script — `ExecuteBashCommand` / `ExecutePowershellCommand` run the shipped scripts and `mkdir`, never Markdown. There is no background execution; a turn that ends without a write has produced nothing.
|
|
690
|
+
- **One conversation = one phase.** A phase closes when its checkpoint is confirmed; state readiness for the next phase in one line (the first ```sf chip carries it) and stop — do not ask a "start now?" question on top, and do not enter the next phase until the user asks for it.
|
|
691
|
+
- **A ```sf chip is the user's next message to you — never your request to them.** Every chip is something the user would say: an instruction (`Generate the process document`), a decision, or their own question back to you (`Show me what you assumed about the exceptions`). Anything you are asking *of* the user — a document, an attachment, a recording, a clarification — is never a chip. Evidence requests ride the ask itself and its Record + Attach widget (`interview-spine.md`), because a chip click sends a message and a message cannot attach a file. `Attach the current SOP` and `Show me the current process steps and exception paths` are your own asks wearing a chip: clicking either hands your question back to you and advances nothing. On a turn that already carries a question or an `AskUser`, chips carry a *different* next step — never a restatement of the ask, never a duplicate of one of its options.
|
|
692
|
+
- **The first ```sf follow-up suggestion is always the flow's next step**, derived from current state: capture/confirm gaps during a phase → the next confirmation; AS-IS confirmed → start future-process design; TO-BE confirmed + readiness task list complete, no draft yet → `Generate the process document`; draft exists → refine or download it. Business wording, user-perspective phrasing.
|
|
693
|
+
- **Once the process document is complete, `Design the solution` is the first ```sf chip** — ahead of refine and download, because Phase 4 is the flow's next step from there. It stays the first chip until a solution design is registered; after that the chips return to refining or downloading. Offer it as a chip only; never add a "shall I start?" question on top.
|
|
694
|
+
- **On a turn that first delivers a PDD or SDD, the second ```sf follow-up suggestion is the Automation Hub publish offer.** Wording and silence conditions: Automation Hub publishing. A publish request → `automation-hub-publishing.md`.
|
|
695
|
+
- **One review per phase, before its checkpoint; confirmed pages never change silently.** Residue from that single review is listed for the user, never chased through a second pass. A factual edit to a `verified` page is named in chat in the same reply that makes it — one sentence per write round, no `AskUser`. See One Review Per Phase.
|
|
696
|
+
- **MANDATORY STOP checkpoints** — exactly the per-phase checkpoints from Conversation Economy: the AS-IS Process Snapshot (the draft review on the recording-first route), the TO-BE Future Process Review, and the PDD metadata ask. No additional confirmation stops.
|
|
697
|
+
- **The no-additional-stops rule scopes to Phases 1–3.** Phase 4 runs a specialist inline, and the specialist's own gate — planner's design review confirmation — is a required stop, not an extra BA checkpoint. Never skip a specialist gate to satisfy the rule above; translating a gate's surface never removes the stop.
|
|
698
|
+
- **Diagram artifacts follow `diagram-authoring.md`.** The AS-IS artifact follows the capture format: `as-is/process.bpmn` (BPMN choice, or user-supplied verbatim) or `as-is/diagram.md` (`case-design` for a case capture, `generic` for a simple flow). TO-BE follows the confirmed route: `to-be/process.bpmn` for BPMN, or a Mermaid map at `to-be/diagram.md`. `generic` is `diagramFamily=generic`, `flowchart TB`, exactly one `kind-*` and one `mode-*` per node, optimized for the narrow, tall side panel. `case-design` is `diagramFamily=case-design`, `flowchart LR`, stage subgraphs, exactly one `case-task-*` per task, semantic terminal markers, and no authored `classDef`. Validate before presenting or advancing: XML parse for BPMN; read-back source audit for Mermaid.
|
|
699
|
+
- **Don't re-read phase MDs already read this session.**
|
|
700
|
+
- **Recordings come back as human messages with `<agent_hint>`.** Treat as normal input. When analyzing one with `/recorded-task/analyze-recording`, never pass `local_skills_path`: no routine.
|
|
701
|
+
- **Don't enter plan mode.** Just run the funnel, ask questions, batch wiki writes once you've got a coherent picture of the process or context is nearing compaction. Always flush at the end of AS-IS and at the end of TO-BE.
|
|
702
|
+
- **Update `log.md` once per phase, at phase close.** Not per action.
|
|
703
|
+
|
|
704
|
+
## Using Delegate
|
|
705
|
+
|
|
706
|
+
### Exporting a living document
|
|
707
|
+
|
|
708
|
+
When the user explicitly asks for a registered PDD or SDD as a Word document, use the host export endpoint. Do not rebuild the document with word-architect or `/doc-gen`.
|
|
709
|
+
|
|
710
|
+
| Endpoint | Body | Output |
|
|
711
|
+
|---|---|---|
|
|
712
|
+
| `/livingdoc/export` | `type` (registered document type) | Word document |
|
|
713
|
+
|
|
714
|
+
Call it through `ExecuteSkillApi` with one operation. Delegate does not produce PDFs. When the user asks for a PDF, say so, offer the Word document instead, and wait for their agreement before calling the endpoint — they can save it as a PDF from Word. On success, the interface attaches the output file as a clickable link; do not repeat or rewrite its path. State any stale sections or omitted diagrams in plain business language.
|
|
715
|
+
|
|
716
|
+
**AskUser:** each call takes a `questions` array. Types: `text` (with `options:` or free-text) and `media` (`recordable: true` offers screen recording, `attachable: true` offers the file drop zone, both flags allowed, no options). Uploaded audio files are transcribed into independently editable recording cards. Media questions are required by default; set `optional: true` only when the user may skip them. Example:
|
|
717
|
+
|
|
718
|
+
```json
|
|
719
|
+
{ "questions": [ { "type": "media", "payload": { "recordable": true, "attachable": true, "question": "How do you download the report from Application X?" } } ] }
|
|
720
|
+
```
|
|
721
|
+
|
|
722
|
+
**File operations:** Read/Write tool for all wiki file reads and writes.
|
|
723
|
+
|
|
724
|
+
**Diagram artifacts:** the Delegate interface renders diagram files for the user; do NOT paste diagram source into chat when asking for feedback — ask the user to confirm. Authoring rules, families, classes, and audits: `diagram-authoring.md`.
|
|
725
|
+
|
|
726
|
+
**PDD generation:** Read `pdd-generation.md` and `pdd-template.md` together when entering Phase 3. Phase 3 produces a living-draft `.md`; the host Download action or `/livingdoc/export` produces the deliverable. Never rebuild the `.docx` yourself — no word-architect, no `/doc-gen`.
|
|
727
|
+
|
|
728
|
+
For living drafts, emit section anchors immediately before headings. The host
|
|
729
|
+
derives viewer section navigation; never author `[TOC]`, manual entries, or UI.
|