skilld-harness 3.6.7 → 3.6.8

package/dist/skills/generate-package-skill/SKILL.md

@@ -144,6 +144,10 @@ Report to the user:
   Give each item expected, observed, a minimal repro with its output, and a source link at the release tag. File it only if asked.
 - The source path or documentation URL behind each version-specific rule.
 - If the Skill sits inside the published package directory, add its directory to `files` in `package.json`. After one build, list packed files with `npm pack --dry-run --ignore-scripts`; `pnpm pack` rejects that flag.
+- For npm packages, use `skills/<name>/SKILL.md` beside the package's `package.json`. Include linked files in the tarball.
+  [pnpm 12.11 and newer](https://pnpm.io/agent-skills) link Skills from approved direct dependencies.
+  Document `pnpm approve` for developers who want these links. Approval covers every Skill and later version of the package.
+  Never approve a package for the user without their request. pnpm owns its links, dependency updates, and removal.
 - Edit the README beside the Skill's `package.json`. If it already links the skilld.dev page, keep that link. Else add the badge below after the others.
   Replace a `skilld add` tip that names this package in place with the tip below. Else add the tip after the install command.
   Replace `OWNER`, `REPOSITORY`, and `PACKAGE`. Count every `SKILL.md` in the Repository, hidden Agent folders included; the skilld.dev indexer skips test and fixture folders.

package/dist/skills/skilld/SKILL.md

@@ -16,6 +16,7 @@ Use `--json` with `search`, `run`, and `update --check`.
 Use `--json` with `view` of a registry ref, and with every registry and account command below.
 Their `data` is the skilld.dev answer. A command whose answer has no body returns `data: null`.
 Use `--json` with `sync` and `sync --check` for declared Skills.
+Use `doctor --json` for read-only Skill discovery and cleanup findings.
 The remaining commands do not support JSON output.
 Use `--plain` when another command needs stable text.
 
@@ -421,6 +422,28 @@ Use `outdated` for stale, unverified, local, and unmanaged Skill reports.
 Read every proposed command before using it.
 Never delete an unmanaged Skill unless the user names it for removal.
 
+## Review Skill files across projects
+
+Run `skilld doctor --json` to scan the user's home directory.
+Pass explicit roots to narrow the scan or inspect files outside home.
+The default scan excludes dependencies, caches, generated files, backups, and Git worktrees.
+Use `--exclude GLOB` to prune more paths.
+Use `--include-excluded` or `--include-worktrees` only when those files matter to the task.
+
+Read owner labels, duplicate paths, source matches, and scan problems separately.
+The `pnpm` owner names Skills linked from npm dependencies. Update these through pnpm, not skilld.
+Doctor never migrates or removes pnpm-owned links. `outdated --all` excludes them from replacement suggestions.
+Exit code `1` with success data means some files could not be checked.
+Use `--check-sources` for up to 20 source candidates through skilld.dev.
+A name match alone never establishes provenance.
+
+The interactive `skilld doctor` separates skills.sh installs with Tab.
+The user can review migration with `m` or removal with `d`.
+Each action shows affected paths and requires confirmation.
+Backups include original files, lock metadata, and recovery instructions.
+Use the existing update workflow for skilld installs.
+Historical commit searches and project-to-global deduplication are outside Doctor's current actions.
+
 ## Manage account authentication
 
 Check account authentication before starting login:

package/dist/skills/skilld-maintained-skills.json

@@ -3,5 +3,6 @@
   "update-package-skill",
   "generate-project-skill",
   "review-skill",
-  "skilld"
+  "skilld",
+  "skilld-tui-design"
 ]

package/dist/skills/skilld-tui-design/SKILL.md

@@ -0,0 +1,100 @@
+---
+name: skilld-tui-design
+description: Designs and verifies skilld CLI output, terminal pickers, and full-screen flows. Use for terminal colours, loading feedback, keyboard controls, layout, errors, or end-to-end CLI UX reviews.
+license: MIT
+---
+
+# Design skilld CLI interactions
+
+Build the user's whole command journey, including failure, cancellation, and the return to their shell.
+Read [source decisions](references/source-review.md) for the two reviewed Skills and their exact revisions.
+
+## Establish the contract
+
+Read the project's `GLOSSARY.md`, `COPY.md`, and `VISION.md` when present.
+Keep canonical nouns and strings. Use short sentences with one idea each.
+Inspect the command parser, output renderer, and terminal lifecycle before changing behavior.
+Classify each affected surface as static output, a picker, a prompt, or a full-screen view.
+Include command help, parser errors, startup notices, and progress indicators in the inventory.
+List its starting state, successful outcome, expected failures, and cancellation behavior.
+Keep the existing JSON envelope, streams, and exit codes unless the task changes that contract.
+
+Use the existing shared renderer and semantic palette.
+Avoid adding a second framework or per-screen theme.
+Start with `crates/skilld-ui`, `crates/skilld-command/src/output.rs`, and `crates/skilld-native/src`.
+Inspect command help and routing in `crates/skilld-command/src/lib.rs`.
+
+## Put the decision first
+
+Static output starts with the requested result or the failed action.
+Group repeated records consistently. Place supporting metadata below each record.
+Use spacing and headings before adding boxes.
+Keep commands copyable. Preserve whitespace inside quoted arguments.
+Wrap prose and paths by terminal cells. Preserve the original value in machine output.
+Give empty results a useful next action based on available commands.
+
+Pickers show the question, query, matching count, selected item, and valid keys.
+Full-screen views keep scope and current state visible.
+For large inventories, start with recommendations and counts. Open projects or folders before listing individual Skills.
+Use words for findings. Label identical copies as duplicates and show their paths; separate symlinks from directory copies.
+State when a filter changes only the view. An action review must still list every affected target.
+Spend rows on the user's decision. Do not reserve space for decorative panels.
+If a pane cannot show complete content, provide reachable scrolling or a detail view.
+At smaller sizes, simplify the layout before hiding required information.
+If the screen is unusable, block hidden actions and show the required size.
+
+## Use colour with independent cues
+
+Inherit the terminal's foreground and background for body text.
+Share roles for brand, action, secondary information, success, warning, error, focus, and selection.
+Use one action accent across screens. Reserve status colours for actual status.
+Show ownership as a label, without implying safety or success.
+Pair colour with words, markers, bold text, or inverse selection.
+Make pane focus visible when colour is disabled.
+Honor `NO_COLOR`, `TERM=dumb`, and existing output capability detection.
+Never claim contrast across unknown terminal themes from one screenshot.
+
+## Make state and input agree
+
+Use explicit states for loading, browsing, editing, review, applying, completion, and failure.
+Prefer tagged states carrying their required data over overlapping pending flags.
+Keep filesystem and network work outside the input loop.
+If operations can overlap, identify their results so stale work cannot replace the current view.
+Derive available actions and footer hints from the same state rules.
+Keep query editing separate from global shortcuts. Typing `q` in a query must remain possible.
+Make Escape predictable: cancel editing or review before leaving the whole screen.
+Preserve selection by stable identity when refreshing data.
+Bound scrolling and expose position when the remaining content is otherwise unclear.
+Provide discoverable help when the current footer cannot explain all controls.
+
+## Explain waiting and outcomes
+
+Show the operation, scope, elapsed time, and available cancellation action.
+When the total is unknown, show activity and observed counts without a percentage.
+When the total is known, show completed and total work.
+Keep previous results readable during refresh when they remain relevant. Label them as previous results if refresh fails.
+Keep errors visible until the user changes state or dismisses them.
+Do not replace an error with a filter label or a generic loading message.
+Describe partial completion explicitly. Preserve recovery and backup paths.
+
+Before a destructive change, show the exact action, item, paths, and content replacement.
+Require deliberate confirmation of that specific plan.
+Discard stale activation keys before entering consent states. Preserve cancellation requests.
+If writes cannot stop safely, defer exit and say so. Never display a cancellation key that does nothing.
+Restore the terminal on success, failure, and cancellation.
+After leaving the alternate screen, print a durable summary of completed changes and recovery paths.
+Provenance describes source evidence. It never proves that instructions are safe.
+
+## Verify the whole journey
+
+Run focused behavior tests for changed transitions and regressions.
+Use the real renderer for layout tests. Do not test source text or private implementation shape.
+Inspect actual terminal captures, including colour and monochrome.
+Cover normal and narrow sizes, short screens, Unicode names, long paths, and resize during confirmation.
+Exercise loading, empty results, errors, filtering, review, cancellation, completion, and return to the shell.
+Use disposable fixtures for migration, removal, and restoration.
+Check plain and JSON output separately for escape sequences, stream changes, and exit codes.
+Run the repository's required checks after integration.
+
+Report what changed and what was exercised. Identify any live path that remains untested.
+Do not treat passing snapshots as proof of keyboard behavior or terminal restoration.

package/dist/skills/skilld-tui-design/references/source-review.md

@@ -0,0 +1,54 @@
+# Source decisions
+
+This Skill contains original guidance for skilld CLI work.
+Two upstream Skills informed the design. Their application scaffolds and scripts were not incorporated.
+
+## tui-design
+
+Author: Diego Marino.
+Repository: `diegomarino/tui-design`.
+Revision: `57c199f22dbf04f057c279431d45346cc1fd5dcd`.
+[Reviewed SKILL.md](https://github.com/diegomarino/tui-design/blob/57c199f22dbf04f057c279431d45346cc1fd5dcd/skills/tui-design/SKILL.md).
+License: MIT, declared in the reviewed Skill frontmatter.
+
+Adopted ideas:
+
+- Classify static output, pickers, and full-screen interactions separately.
+- Use semantic colours with independent selection and status cues.
+- Design around terminal cells and inspect actual rendered evidence.
+- Preserve selection and distinguish unknown totals from measured progress.
+- Treat plain output and terminal restoration as product behavior.
+
+Excluded prescriptions:
+
+- Mandatory concept counts, large token schemas, and subjective craft scores.
+- Python mock tooling and copied starter applications.
+- Forced dark themes or universal shortcut sets.
+- Mandatory NDJSON, additive-only schemas, and errors-last ordering.
+
+These prescriptions conflict with skilld's existing contracts or add work without improving the user's decision.
+
+## tui-development
+
+Author: NVIDIA.
+Repository: `NVIDIA/OpenShell`.
+Revision: `fb8f6c0885127fe4ca29993739131a3bdbaed477`.
+[Reviewed SKILL.md](https://github.com/NVIDIA/OpenShell/blob/fb8f6c0885127fe4ca29993739131a3bdbaed477/.agents/skills/tui-development/SKILL.md).
+License: [Apache-2.0 at the reviewed revision](https://github.com/NVIDIA/OpenShell/blob/fb8f6c0885127fe4ca29993739131a3bdbaed477/LICENSE).
+
+Adopted ideas:
+
+- Keep long operations outside the input loop.
+- Represent lifecycle states explicitly and route background results through events.
+- Make focus, valid actions, scope, and outcomes visible.
+- Review file changes before applying them.
+
+Excluded prescriptions:
+
+- OpenShell branding, topology, deployment commands, and splash screens.
+- Mandatory Tokio, unbounded channels, and fixed network timeouts.
+- Independent pending flags and forced interruption during writes.
+
+The synthesis uses skilld's current host boundaries, shared rendering, and terminal lifecycle.
+It adds durable shell summaries and actual terminal journey verification beyond the upstream build guidance.
+It introduces no deployment, privilege, or installation requirements.

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "skilld-harness",
   "type": "module",
-  "version": "3.6.7",
+  "version": "3.6.8",
   "description": "Run skilld-maintained Skills with checked, atomic output",
   "author": {
     "name": "Harlan Wilton",