@bendyline/gilde 0.1.26 → 0.1.28

package/data/craftbook-templates/co/content-accuracy-review/versions/1.0.1/test.json

@@ -0,0 +1,149 @@
+{
+  "schemaVersion": 1,
+  "title": "Claim-level accuracy review from evidence pack",
+  "objective": "Measure whether the content-accuracy-review craftbook extracts checkable claims and writes sourced true/false/misleading/unverifiable verdicts.",
+  "tags": [
+    "corpus"
+  ],
+  "prompt": "In the `Content Accuracy Review Eval` project, read only `source/article.md` and `source/evidence-pack.md`, then write `accuracy-review.md`. Extract the six checkable claims in the article, rate each true/false/misleading/unverifiable, cite the evidence pack, give corrections, and include an overall trust rating. Do not add claims, companies, sources, or metrics beyond the seeded article and evidence pack.",
+  "setup": {
+    "projectName": "Content Accuracy Review Eval",
+    "about": "Self-contained eval project for content-accuracy-review. The draft article and evidence pack are seeded in workspace/source.",
+    "missionObjectives": "Produce a claim-by-claim accuracy review with verdicts, corrections, sources, and a trust rating.",
+    "files": [
+      {
+        "path": "source/article.md",
+        "content": "# Draft article\n\nHarborCRM launched guided returns intake in February 2026. The pilot covered all SKUs and reduced median first response from 18 hours to 4 hours. Refund leakage fell from 14.2% to 8.9%. Support escalations dropped 21% after agents received recommended resolution scripts. The company guarantees national rollout in July.\n"
+      },
+      {
+        "path": "source/evidence-pack.md",
+        "content": "# Evidence pack\n\n- Launch memo: guided returns intake pilot launched in March 2026, not February.\n- Pilot scope: 18 home-office accessory SKUs, not all SKUs.\n- Metrics dashboard: median first response improved from 18 hours to 6 hours.\n- Metrics dashboard: refund leakage fell from 14.2% to 8.9%.\n- Support report: escalations dropped 21% after recommended resolution scripts were added.\n- Roadmap note: national rollout is proposed for July but not guaranteed.\n"
+      }
+    ],
+    "worker": {
+      "name": "Kiki",
+      "role": "Fact Checker"
+    }
+  },
+  "mocks": [],
+  "success": {
+    "summary": "workspace/accuracy-review.md extracts seeded claims, assigns verdicts, cites evidence, corrects false/misleading claims, and gives an overall trust rating.",
+    "deliverables": [
+      {
+        "path": "accuracy-review.md",
+        "kind": "markdown-report",
+        "minBytes": 1600,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "accuracy-review.md",
+            "pattern": "(?:^|\\n)#{1,3}\\s+\\S",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "accuracy-review.md",
+            "pattern": "TRUE|FALSE|MISLEADING|UNVERIFIABLE|trust rating",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "accuracy-review.md",
+            "pattern": "February|March|all SKUs|18 home-office",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "accuracy-review.md",
+            "pattern": "18\\s*hours|4\\s*hours|6\\s*hours|14\\.2%|8\\.9%|21%",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "accuracy-review.md",
+            "pattern": "correction|correct statement|source|evidence",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "accuracy-review.md",
+            "pattern": "guarantees national rollout|proposed for July",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "accuracy-review.md",
+            "pattern": "February[\\s\\S]{0,250}False|False[\\s\\S]{0,250}February",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "accuracy-review.md",
+            "pattern": "all SKUs[\\s\\S]{0,250}False|False[\\s\\S]{0,250}all SKUs",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "accuracy-review.md",
+            "pattern": "4\\s*hours[\\s\\S]{0,250}(False|Misleading)|(False|Misleading)[\\s\\S]{0,250}4\\s*hours",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "accuracy-review.md",
+            "pattern": "14\\.2%[\\s\\S]{0,250}True|8\\.9%[\\s\\S]{0,250}True|True[\\s\\S]{0,250}14\\.2%",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "accuracy-review.md",
+            "pattern": "21%[\\s\\S]{0,250}True|True[\\s\\S]{0,250}21%",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "accuracy-review.md",
+            "pattern": "guarantees[\\s\\S]{0,300}(False|Misleading)|(False|Misleading)[\\s\\S]{0,300}guarantees",
+            "flags": "i"
+          },
+          {
+            "kind": "notContains",
+            "file": "accuracy-review.md",
+            "pattern": "Q3|AlphaCorp|enterprise software|cash reserve|dominant market|CEO announced|RESTful API|stress tests",
+            "flags": "i",
+            "label": "do not invent review claims or evidence beyond the seeded article"
+          }
+        ]
+      }
+    ]
+  },
+  "rubric": {
+    "artifact": {
+      "path": "accuracy-review.md",
+      "kind": "markdown"
+    },
+    "axes": [
+      {
+        "name": "grounding",
+        "description": "Seeded facts appear faithfully and the source is cited."
+      },
+      {
+        "name": "structure",
+        "description": "Sections progress logically for the artifact class."
+      },
+      {
+        "name": "completeness",
+        "description": "Required elements (risks, next actions, scope) are all present."
+      },
+      {
+        "name": "tone",
+        "description": "The register fits the audience and artifact class."
+      }
+    ]
+  },
+  "qualityFocus": [
+    "claim extraction",
+    "evidence-backed verdicts",
+    "corrections"
+  ]
+}

package/data/craftbook-templates/co/content-deck/versions/1.2.1/craftbook.json

@@ -0,0 +1,206 @@
+{
+  "id": "content-deck",
+  "name": "Slide Deck from Content",
+  "description": "Turn a body of source content into a presentation deck. Outlines the narrative first, designs a consistent slide look, then builds the deck (HTML slides), evaluating for narrative flow and visual consistency.\n\nA gallery craftbook generated from an archetype spec. It runs\n`phase → (per-phase gate) → … → evaluate → (loop) → finish`. Each build\nphase that produces a checkable artifact is followed by a **runtime\ngate-checkpoint** — the runtime verifies the artifact and routes with no\nmodel turn, looping back to redo the phase on a miss. The final `evaluate`\nstep holds a static deliverable gate plus a reviewer QA pass. What it adds\nover the generic `build-loop`: a specialist role per phase, a\ndomain-correct ordering, and a concrete per-phase quality bar.\n\nDeliverables marked \"artifact\" land in the project's artifacts drawer (`write_artifact` / `read_artifact`), not the shipped workspace — review output is not product source.\n\nPhases:\n\n1. Outline the deck (planner) — narrative arc + per-slide bullets from the source → gated on artifact `deck/outline.md` (markdown-notes)\n2. Slide design (designer) — a consistent slide template + palette → gated on artifact `deck/design-system.md` (markdown-notes)\n3. Build the deck (developer) — generate the slides to the template → gated on `index.html` (slide-deck)\n\nThe gates never advance with an unmet criterion, and loop back to the\nowning phase to fix named gaps.\n",
+  "entryStepId": "outline",
+  "triggers": [
+    "slide deck",
+    "powerpoint",
+    "presentation",
+    "build a deck"
+  ],
+  "steps": [
+    {
+      "id": "outline",
+      "name": "Outline the deck",
+      "description": "narrative arc + per-slide bullets from the source",
+      "prompt": "Read every supplied source file with workspace `read_file` before outlining. Build a source-grounded deck plan: a title slide, a clear narrative arc, 5-7 content slides with one idea and 2-4 bullets each, and a takeaways/CTA ending. Attach a source filename or source heading to every factual claim. Write an **acceptance-criteria checklist** (\"opens with a title slide\", \"one idea per slide\", \"ends with takeaways/CTA\", \"every claim traces to the source\"). Save the outline + checklist to artifact `deck/outline.md` with `write_artifact`; note the path in task notes.",
+      "suggestedRole": "planner",
+      "advanceWhen": {
+        "file": "deck/outline.md",
+        "minBytes": 1,
+        "sniff": "nonempty",
+        "artifact": true
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "deck/outline.md",
+            "bytes": 700,
+            "artifact": true
+          },
+          {
+            "kind": "sniff",
+            "file": "deck/outline.md",
+            "sniff": "nonempty",
+            "artifact": true
+          },
+          {
+            "kind": "contains",
+            "file": "deck/outline.md",
+            "pattern": "title slide|slide 1",
+            "flags": "i",
+            "label": "title slide",
+            "artifact": true
+          },
+          {
+            "kind": "contains",
+            "file": "deck/outline.md",
+            "pattern": "takeaway|CTA|call.to.action|recommendation",
+            "flags": "i",
+            "label": "closing slide",
+            "artifact": true
+          }
+        ],
+        "onReject": "outline",
+        "maxAttempts": 3
+      },
+      "next": "slide-design"
+    },
+    {
+      "id": "slide-design",
+      "name": "Slide design",
+      "description": "a consistent slide template + palette",
+      "prompt": "Read artifact `deck/outline.md` with `read_artifact`. Design one reusable slide system: title position, body/grid layout, exact palette, type scale, spacing, progress affordance, and focused/inactive slide states. Include desktop and narrow-screen behavior plus keyboard/button navigation treatment. Consistency matters more than flourish. Save the template spec to artifact `deck/design-system.md` with `write_artifact`; note the path in task notes. No slide content yet.",
+      "suggestedRole": "designer",
+      "consumes": [
+        {
+          "file": "deck/outline.md",
+          "artifact": true
+        }
+      ],
+      "advanceWhen": {
+        "file": "deck/design-system.md",
+        "minBytes": 1,
+        "sniff": "nonempty",
+        "artifact": true
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "deck/design-system.md",
+            "bytes": 500,
+            "artifact": true
+          },
+          {
+            "kind": "sniff",
+            "file": "deck/design-system.md",
+            "sniff": "nonempty",
+            "artifact": true
+          },
+          {
+            "kind": "contains",
+            "file": "deck/design-system.md",
+            "pattern": "#[0-9a-f]{3,8}|rgb\\(|hsl\\(",
+            "flags": "i",
+            "label": "concrete palette",
+            "artifact": true
+          },
+          {
+            "kind": "contains",
+            "file": "deck/design-system.md",
+            "pattern": "keyboard|arrow|previous|next|navigation",
+            "flags": "i",
+            "label": "navigation treatment",
+            "artifact": true
+          }
+        ],
+        "onReject": "slide-design",
+        "maxAttempts": 3
+      },
+      "next": "build"
+    },
+    {
+      "id": "build",
+      "name": "Build the deck",
+      "description": "generate the slides to the template",
+      "prompt": "Read artifacts `deck/outline.md` and `deck/design-system.md` with `read_artifact`. Build the deck in workspace `index.html` with `write_file`: one semantic section per slide, inline CSS/JS, visible slide count/progress, previous/next controls, and ArrowLeft/ArrowRight keyboard navigation. Apply the template uniformly and use only claims in the outline/source. Keep it self-contained with no external assets. On a loop-back, fix only the named gaps. `write_task_note` the workspace path and slide count.",
+      "suggestedRole": "developer",
+      "consumes": [
+        {
+          "file": "deck/outline.md",
+          "artifact": true
+        },
+        {
+          "file": "deck/design-system.md",
+          "artifact": true
+        }
+      ],
+      "advanceWhen": {
+        "file": "index.html",
+        "minBytes": 1,
+        "sniff": "nonempty"
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "index.html",
+            "bytes": 3600
+          },
+          {
+            "kind": "htmlLint",
+            "file": "index.html"
+          },
+          {
+            "kind": "cssMinBytes",
+            "bytes": 700,
+            "file": "index.html"
+          },
+          {
+            "kind": "jsParses",
+            "file": "index.html"
+          },
+          {
+            "kind": "contains",
+            "file": "index.html",
+            "pattern": "ArrowLeft|ArrowRight|keydown",
+            "label": "keyboard navigation"
+          }
+        ],
+        "scripts": [
+          {
+            "name": "checkContains",
+            "scope": "standard",
+            "inputs": {
+              "file": "index.html",
+              "pattern": "(?:slide|section|^---$)",
+              "flags": "im"
+            }
+          }
+        ],
+        "onReject": "build",
+        "maxAttempts": 4
+      },
+      "next": "evaluate"
+    },
+    {
+      "id": "evaluate",
+      "name": "Evaluate",
+      "description": "Grade the deliverable against every acceptance criterion. All pass → finish; any fail → loop back and fix the gap.",
+      "prompt": "Read workspace `index.html` with `read_file` and artifacts `deck/outline.md` and `deck/design-system.md` with `read_artifact`. Check narrative flow (title opening, one idea per slide, takeaways ending), visual consistency, source fidelity, and both button and keyboard navigation. Render and exercise it if browser QA is available; otherwise trace the event handlers. Write PASS/FAIL per criterion.\n\nThen route — this is the whole point of the loop:\n\n- **Every criterion PASSES →** call `advance_task_step({ ref, stepId: \"evaluate\", next: \"finish\" })`.\n- **Any criterion FAILS →** write the specific gaps to notes, then call `advance_task_step({ ref, stepId: \"evaluate\", next: \"build\" })` to loop back. The builder fixes exactly those gaps.\n\nNever route to `finish` while any criterion is unmet. The build phase's completion gate already blocked a grossly-incomplete deliverable; your job is the judgment an automated check cannot make (does it actually work, read well, look right). After ~3 unproductive loops, stop and report DONE_WITH_CONCERNS so the user can step in.",
+      "suggestedRole": "reviewer",
+      "consumes": [
+        {
+          "file": "index.html"
+        }
+      ],
+      "next": "build"
+    },
+    {
+      "id": "finish",
+      "name": "Finish",
+      "description": "All acceptance criteria met. Stamp a short summary and report DONE.",
+      "prompt": "Every acceptance criterion passed. Write a one-paragraph DONE summary to task notes via `write_task_note`: what was built, the deliverable path(s), and a one-line confirmation that each criterion is met. Then report DONE.",
+      "suggestedRole": "developer",
+      "terminal": true
+    }
+  ],
+  "version": "1.2.1",
+  "releasedAt": "2026-08-14T00:00:00Z"
+}

package/data/craftbook-templates/co/content-deck/versions/1.2.1/test.json

@@ -0,0 +1,169 @@
+{
+  "schemaVersion": 1,
+  "title": "Source-grounded HTML slide-deck workflow",
+  "objective": "Run the Slide Deck from Content craftbook and verify a source-cited outline and explicit design system precede a complete navigable HTML deck.",
+  "tags": [
+    "html-page",
+    "workflow"
+  ],
+  "prompt": "Use the Slide Deck from Content craftbook. Read workspace `source/field-notes.md`, then build workspace `index.html` for an operations leadership review. The deck needs a title slide, 5-7 content slides, a takeaways/CTA slide, consistent visual treatment, source-grounded claims, visible progress, previous/next controls, and ArrowLeft/ArrowRight navigation. Vanilla HTML/CSS/JS only, with no external assets or build step.",
+  "setup": {
+    "projectName": "Content Deck Eval",
+    "about": "Self-contained workflow eval for the Slide Deck from Content craftbook. Source notes are in the workspace, outline/design deliverables belong in artifacts, and the final deck belongs at workspace/index.html.",
+    "missionObjectives": "Turn the seeded field notes into a concise, source-grounded, navigable HTML slide deck with a locked slide design and reliable keyboard/button navigation.",
+    "files": [
+      {
+        "path": "source/field-notes.md",
+        "content": "# Field Notes: Returns Desk Pilot\n\nAudience: operations leadership deciding whether to expand the pilot.\n\n## Context\n- Harbor & Finch ships home-office accessories through three regional warehouses.\n- A March returns-desk pilot used guided intake forms, photo evidence, and a triage queue.\n- The goal was to cut preventable refunds without frustrating customers.\n\n## Results\n- Median first response improved from 18 hours to 6 hours.\n- Refund leakage fell from 14.2% to 8.9% on pilot SKUs.\n- Support escalations dropped 21% after agents received recommended resolution scripts.\n- Customers still complained about status silence after photo submission.\n\n## Risks\n- Warehouse teams need barcode exception training before national rollout.\n- Finance wants weekly exception exports before approving automation.\n- The photo review queue can exceed SLA on Monday mornings.\n\n## Recommendation\nExpand to the West warehouse for 30 days, add automated status emails, and hold national rollout until exception exports are reliable.\n"
+      }
+    ],
+    "worker": {
+      "name": "Sanne",
+      "role": "Developer"
+    }
+  },
+  "mocks": [],
+  "success": {
+    "summary": "A craftbook task produces source-grounded outline/design artifacts and a complete navigable workspace/index.html deck.",
+    "deliverables": [
+      {
+        "path": "index.html",
+        "kind": "slide-deck",
+        "minBytes": 3600,
+        "checks": [
+          {
+            "kind": "cssMinBytes",
+            "bytes": 700,
+            "file": "index.html"
+          },
+          {
+            "kind": "jsParses",
+            "file": "index.html"
+          },
+          {
+            "kind": "contains",
+            "file": "index.html",
+            "pattern": "Returns Desk Pilot",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "index.html",
+            "pattern": "18\\s*hours[\\s\\S]*6\\s*hours|6\\s*hours[\\s\\S]*18\\s*hours"
+          },
+          {
+            "kind": "contains",
+            "file": "index.html",
+            "pattern": "14\\.2%[\\s\\S]*8\\.9%|8\\.9%[\\s\\S]*14\\.2%"
+          },
+          {
+            "kind": "contains",
+            "file": "index.html",
+            "pattern": "takeaways|recommendation|CTA",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "index.html",
+            "pattern": "previous|prev[\\s\\S]*next|next[\\s\\S]*(?:previous|prev)",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "index.html",
+            "pattern": "ArrowLeft|ArrowRight|keydown"
+          }
+        ]
+      }
+    ],
+    "checks": [
+      {
+        "kind": "minBytes",
+        "file": "deck/outline.md",
+        "bytes": 700,
+        "artifact": true
+      },
+      {
+        "kind": "contains",
+        "file": "deck/outline.md",
+        "pattern": "(?:title slide|slide 1)[\\s\\S]*(?:takeaway|recommendation|CTA)",
+        "flags": "i",
+        "artifact": true
+      },
+      {
+        "kind": "minBytes",
+        "file": "deck/design-system.md",
+        "bytes": 500,
+        "artifact": true
+      },
+      {
+        "kind": "contains",
+        "file": "deck/design-system.md",
+        "pattern": "#[0-9a-f]{3,8}[\\s\\S]*(?:keyboard|arrow|navigation)",
+        "flags": "i",
+        "artifact": true
+      }
+    ],
+    "taskNotes": {
+      "minBytes": 160,
+      "checks": [
+        {
+          "kind": "contains",
+          "file": "task-notes.md",
+          "pattern": "outline\\.md",
+          "flags": "i"
+        },
+        {
+          "kind": "contains",
+          "file": "task-notes.md",
+          "pattern": "design-system\\.md",
+          "flags": "i"
+        },
+        {
+          "kind": "contains",
+          "file": "task-notes.md",
+          "pattern": "index\\.html",
+          "flags": "i"
+        }
+      ],
+      "requireCraftbookTask": true
+    },
+    "taskGraph": {
+      "requireCraftbookTask": true,
+      "requireTerminalStep": true
+    },
+    "unchangedFixtures": [
+      "source/field-notes.md"
+    ]
+  },
+  "rubric": {
+    "artifact": {
+      "path": "index.html",
+      "kind": "html"
+    },
+    "axes": [
+      {
+        "name": "narrative",
+        "description": "The deck has a deliberate source-grounded arc with one clear idea per slide and a decision-oriented close."
+      },
+      {
+        "name": "design consistency",
+        "description": "Slides visibly share the locked palette, type scale, spacing, and active/inactive state treatment."
+      },
+      {
+        "name": "navigation",
+        "description": "Buttons, progress, and keyboard controls make the complete deck reliably traversable."
+      },
+      {
+        "name": "workflow evidence",
+        "description": "The outline and design artifacts are substantive and agree with the final rendered deck."
+      }
+    ]
+  },
+  "qualityFocus": [
+    "craftbook attribution",
+    "source-first outline",
+    "artifact-to-workspace handoff",
+    "keyboard and button navigation"
+  ]
+}