@bendyline/gilde 0.1.51 → 0.1.52

package/data/craftbook-templates/te/test-suite-backfill/versions/2.0.1/test.json

@@ -0,0 +1,195 @@
+{
+  "schemaVersion": 1,
+  "title": "Test backfill proven by mutants",
+  "objective": "The craftbook drives a risk-ranked backfill of an untested date-range module. The oracle proves the new tests bite: three seeded mutations (an exclusive boundary, a duration off-by-one, overlap collapsed to containment) are applied in copies and the model's suite must fail at least two of them while staying green on the real tree — with the module under test byte-identical to the fixture.",
+  "tags": [
+    "code",
+    "tests",
+    "coverage",
+    "tactical-fleet"
+  ],
+  "prompt": "Use the Test Suite Backfill craftbook: src/date-range.js schedules bookings across the product and has no tests at all — tests/ only covers the formatter. Backfill a meaningful suite for the date-range module in the existing test layout, highest-risk behavior first: its bounds are inclusive on both ends and an inverted range must throw.",
+  "setup": {
+    "projectName": "Booking Ranges",
+    "about": "A tiny scheduling library. Tests run with `npm run test` (node --test). src/date-range.js is untested; src/format.js has an existing test that shows the layout and style.",
+    "files": [
+      {
+        "path": "package.json",
+        "content": "{\n  \"name\": \"booking-ranges\",\n  \"private\": true,\n  \"version\": \"1.0.0\",\n  \"type\": \"module\",\n  \"scripts\": {\n    \"test\": \"node --test\"\n  }\n}\n"
+      },
+      {
+        "path": "src/date-range.js",
+        "content": "// Inclusive [start, end] day ranges. Days are ISO strings like '2026-03-14'.\nconst DAY_MS = 86400000;\n\nfunction dayValue(day) {\n  if (typeof day !== 'string' || !/^\\d{4}-\\d{2}-\\d{2}$/.test(day)) {\n    throw new Error(`invalid day: ${day}`);\n  }\n  const value = Date.parse(`${day}T00:00:00Z`);\n  if (Number.isNaN(value)) throw new Error(`invalid day: ${day}`);\n  return value;\n}\n\nfunction assertRange(range) {\n  const start = dayValue(range.start);\n  const end = dayValue(range.end);\n  if (start > end) {\n    throw new Error(`invalid range: ${range.start} is after ${range.end}`);\n  }\n  return { start, end };\n}\n\nexport function overlaps(a, b) {\n  const ra = assertRange(a);\n  const rb = assertRange(b);\n  return ra.start <= rb.end && rb.start <= ra.end;\n}\n\nexport function containsDay(range, day) {\n  const r = assertRange(range);\n  const d = dayValue(day);\n  return d >= r.start && d <= r.end;\n}\n\nexport function durationDays(range) {\n  const r = assertRange(range);\n  return Math.round((r.end - r.start) / DAY_MS) + 1;\n}\n"
+      },
+      {
+        "path": "src/format.js",
+        "content": "export function slugify(text) {\n  return String(text)\n    .toLowerCase()\n    .trim()\n    .replace(/[^a-z0-9]+/g, '-')\n    .replace(/^-+|-+$/g, '');\n}\n"
+      },
+      {
+        "path": "tests/format.test.mjs",
+        "content": "import assert from 'node:assert/strict';\nimport { test } from 'node:test';\nimport { slugify } from '../src/format.js';\n\ntest('slugify collapses spaces and punctuation', () => {\n  assert.equal(slugify('Walnut writing desk!'), 'walnut-writing-desk');\n});\n\ntest('slugify trims stray dashes', () => {\n  assert.equal(slugify('  --Oak shelf--  '), 'oak-shelf');\n});\n"
+      },
+      {
+        "path": "tests/verify-backfill.mjs",
+        "content": "import assert from 'node:assert/strict';\nimport { execFileSync } from 'node:child_process';\nimport { cpSync, mkdirSync, readFileSync, readdirSync, rmSync, writeFileSync } from 'node:fs';\nimport { join } from 'node:path';\n\nconst root = process.cwd();\nconst ORACLE = 'verify-backfill.mjs';\n\n// Bare `node --test` discovery: finds tests/**/*.test.* from cwd and skips\n// dot-directories, so the .oracle-* mutant copies never leak into a run.\n// (A literal `--test tests/` directory argument no longer resolves on\n// modern Node - it is treated as a module path.)\nfunction runSuite(cwd) {\n  try {\n    execFileSync(process.execPath, ['--test'], { cwd, stdio: 'pipe', timeout: 30000 });\n    return 0;\n  } catch (err) {\n    return typeof err.status === 'number' ? err.status : 1;\n  }\n}\n\n// 1) New tests exist in the real layout and target the date-range module.\nconst testFiles = readdirSync(join(root, 'tests'), { recursive: true })\n  .map(String)\n  .filter((f) => /\\.test\\.(mjs|js|cjs)$/.test(f) && !f.endsWith('format.test.mjs'));\nassert.ok(testFiles.length >= 1, 'no new test file found in tests/');\nconst newTestText = testFiles.map((f) => readFileSync(join(root, 'tests', f), 'utf8')).join('\\n');\nassert.ok(/date-range/.test(newTestText), 'the new tests never reference the date-range module');\nconsole.log('BACKFILL_ORACLE new tests found');\n\n// 2) The suite is green on the real tree.\nassert.equal(runSuite(root), 0, 'the suite must pass on the real tree');\nconsole.log('BACKFILL_ORACLE suite green');\n\n// 3) Seeded defects in copies must make the new suite fail - proof the tests bite.\nconst source = readFileSync(join(root, 'src', 'date-range.js'), 'utf8');\nfunction mutantCaught(name, mutate) {\n  const mroot = join(root, `.oracle-${name}`);\n  rmSync(mroot, { recursive: true, force: true });\n  mkdirSync(mroot, { recursive: true });\n  cpSync(join(root, 'src'), join(mroot, 'src'), { recursive: true });\n  cpSync(join(root, 'tests'), join(mroot, 'tests'), { recursive: true });\n  rmSync(join(mroot, 'tests', ORACLE), { force: true });\n  cpSync(join(root, 'package.json'), join(mroot, 'package.json'));\n  const mutated = mutate(source);\n  assert.notEqual(mutated, source, `mutation ${name} failed to apply - src/date-range.js was altered`);\n  writeFileSync(join(mroot, 'src', 'date-range.js'), mutated);\n  const caught = runSuite(mroot) !== 0;\n  console.log(`BACKFILL_ORACLE mutant ${name} ${caught ? 'caught' : 'MISSED'}`);\n  return caught;\n}\n\nconst caught = [\n  mutantCaught('boundary', (s) => s.replace('rb.start <= ra.end', 'rb.start < ra.end')),\n  mutantCaught('duration', (s) => s.replace('/ DAY_MS) + 1', '/ DAY_MS)')),\n  mutantCaught('containment', (s) => s.replace('ra.start <= rb.end && rb.start <= ra.end', 'ra.start <= rb.start && rb.end <= ra.end')),\n].filter(Boolean).length;\nassert.ok(caught >= 2, `the new tests caught only ${caught}/3 seeded defects - they do not pin real behavior`);\nconsole.log(`BACKFILL_ORACLE mutants ${caught}/3`);\nconsole.log('BACKFILL_ORACLE done');\n",
+        "surface": "harness"
+      }
+    ]
+  },
+  "mocks": [],
+  "success": {
+    "summary": "New tests for the date-range module live in the real layout, the whole suite is green on the untouched tree, and the tests catch at least two of three seeded mutations — with every fleet artifact present and honestly cited.",
+    "deliverables": [
+      {
+        "path": "{{task.dir}}/risk-map.md",
+        "kind": "markdown-notes",
+        "artifact": true,
+        "minBytes": 500,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "{{task.dir}}/risk-map.md",
+            "pattern": "^##\\s+Modules[\\s\\S]*^##\\s+Risks ranked[\\s\\S]*^##\\s+Chosen targets",
+            "flags": "im",
+            "label": "risk-map sections in order"
+          },
+          {
+            "kind": "contains",
+            "file": "{{task.dir}}/risk-map.md",
+            "pattern": "src/date-range\\.js",
+            "label": "names the real target module"
+          },
+          {
+            "kind": "citationsResolve",
+            "file": "{{task.dir}}/risk-map.md",
+            "minCitations": 2,
+            "artifact": true
+          }
+        ]
+      },
+      {
+        "path": "{{task.dir}}/test-notes.md",
+        "kind": "markdown-notes",
+        "artifact": true,
+        "minBytes": 600,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "{{task.dir}}/test-notes.md",
+            "pattern": "^##\\s+Tests added[\\s\\S]*^##\\s+Behaviors pinned[\\s\\S]*^##\\s+How each test would catch a realistic break",
+            "flags": "im",
+            "label": "test-notes sections in order"
+          },
+          {
+            "kind": "citationsResolve",
+            "file": "{{task.dir}}/test-notes.md",
+            "minCitations": 2,
+            "artifact": true
+          }
+        ]
+      },
+      {
+        "path": "{{task.dir}}/verification.md",
+        "kind": "markdown-notes",
+        "artifact": true,
+        "minBytes": 400,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "{{task.dir}}/verification.md",
+            "pattern": "^##\\s+Suite[\\s\\S]*^##\\s+Coverage summary[\\s\\S]*^##\\s+Result",
+            "flags": "im",
+            "label": "verification sections in order"
+          }
+        ]
+      },
+      {
+        "path": "{{task.dir}}/review.md",
+        "kind": "markdown-report",
+        "artifact": true,
+        "minBytes": 400,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "{{task.dir}}/review.md",
+            "pattern": "Verdict:\\s*(?:PASS|REVISE)",
+            "flags": "i",
+            "label": "explicit reviewer verdict"
+          }
+        ]
+      }
+    ],
+    "checks": [
+      {
+        "kind": "nodeRuns",
+        "file": "tests/format.test.mjs"
+      },
+      {
+        "kind": "nodeScriptPasses",
+        "script": "tests/verify-backfill.mjs",
+        "timeoutMs": 120000,
+        "requiredOutput": [
+          {
+            "pattern": "BACKFILL_ORACLE new tests found",
+            "label": "new tests exist in the real layout"
+          },
+          {
+            "pattern": "BACKFILL_ORACLE suite green",
+            "label": "the whole suite passes on the real tree"
+          },
+          {
+            "pattern": "BACKFILL_ORACLE mutants [23]/3",
+            "label": "the new tests catch seeded defects"
+          }
+        ]
+      }
+    ],
+    "taskNotes": {
+      "minBytes": 120,
+      "checks": [
+        {
+          "kind": "contains",
+          "file": "task-notes.md",
+          "pattern": "\\bDONE\\b[\\s\\S]*(npm\\s+run\\s+test|node\\s+--test)",
+          "flags": "i",
+          "label": "DONE note names the real test command"
+        }
+      ]
+    },
+    "taskGraph": {
+      "requireCraftbookTask": true,
+      "requireTerminalStep": true
+    },
+    "unchangedFixtures": [
+      "package.json",
+      "src/date-range.js",
+      "src/format.js",
+      "tests/format.test.mjs"
+    ]
+  },
+  "rubric": {
+    "artifact": {
+      "path": "{{task.dir}}/test-notes.md",
+      "kind": "markdown"
+    },
+    "axes": [
+      {
+        "name": "Risk-driven targeting",
+        "description": "The tests pin the behaviors whose silent breakage costs most — inclusive bounds, duration arithmetic, the invalid-range throw — following the risk map rather than the easiest paths."
+      },
+      {
+        "name": "Tests that bite",
+        "description": "Each test asserts real values, errors, and edge cases a wrong implementation would fail; no tautologies, and seeded mutations to the module are actually caught."
+      },
+      {
+        "name": "Layout fidelity and honesty",
+        "description": "Tests follow the existing layout and style, the suite result comes from real runs, the module under test is untouched, and uncovered targets are deferred honestly."
+      }
+    ]
+  },
+  "qualityFocus": [
+    "mutation-proof regression guards",
+    "risk-ranked coverage",
+    "enforced review of assertions"
+  ]
+}

package/data/craftbook-templates/th/thank-you-batch/versions/1.0.3/craftbook.json

@@ -0,0 +1,280 @@
+{
+  "id": "thank-you-batch",
+  "name": "Thank-You Note Batch",
+  "description": "Clear the owed-thanks list with notes that sound like you: read who is owed and for what from the correspondence roster, draft one note per person that names the actual gift or kindness — no template smell — then log every note so nobody who mattered goes unanswered. Locks the household voice and the coverage rules FIRST, then drafts the batch, then records it in the roster with a short run summary. Use for thank-you notes, clearing the owed-thanks list, 'we owe thank yous from the party', and post-holiday gratitude sweeps.\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. Scope the batch (letter-writer) — owed list from the roster, voice guide, coverage rules → gated on artifact `{{workPath}}/scope.md` (markdown-notes)\n2. Draft the notes (copywriter) — one personal note per owed person, in the household voice → gated on `letters/thank-you.md` (markdown-doc)\n3. Log the batch (archivist) — record each note in the roster, write the run summary → gated on artifact `{{workPath}}/report.md` (markdown-report)\n\nThe gates never advance with an unmet criterion, and loop back to the\nowning phase to fix named gaps.\n",
+  "entryStepId": "scope",
+  "triggers": [
+    "draft the thank you notes",
+    "clear the owed thanks list",
+    "we owe thank yous",
+    "write the thank you cards",
+    "thank you note batch"
+  ],
+  "paramSchema": {
+    "type": "object",
+    "properties": {
+      "workPath": {
+        "type": "string",
+        "title": "Working folder",
+        "description": "Per-task working folder in the artifacts drawer. Defaults to this task's own folder so runs never collide; override with a stable name when you deliberately want runs to share files.",
+        "default": "{{task.dir}}"
+      }
+    }
+  },
+  "steps": [
+    {
+      "id": "scope",
+      "name": "Scope the batch",
+      "description": "owed list from the roster, voice guide, coverage rules",
+      "prompt": "Define the batch before drafting a word. Step 1: read correspondents.json from the workspace — who is owed thanks, for what gift or kindness, and how long each has been owed. Step 2: read the household's voice notes if they exist ({{workPath}}/voice.md or a voice field on the roster) — the phrases this family actually uses, how they open and sign off, and what they would never say. Step 3: confirm with the owner anyone to add or defer, and how the notes go out (card, email, hand-delivered) since that shapes length. Step 4: write an acceptance-criteria checklist ('every owed person gets a note or an explicit deferral with a reason', 'every note names the actual gift or kindness — a generic thanks is a miss', 'the voice guide is respected and no two notes are interchangeable', 'every drafted note is logged against the roster'). Call write_task_note with the owed count + checklist and write {{workPath}}/scope.md.\n\nThe deliverable `{{workPath}}/scope.md` lands in the project's artifacts drawer — write it with `write_artifact` and read it back with `read_artifact`; the shipped workspace stays untouched.",
+      "suggestedRole": "letter-writer",
+      "advanceWhen": {
+        "file": "{{workPath}}/scope.md",
+        "minBytes": 1,
+        "sniff": "nonempty",
+        "artifact": true
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "{{workPath}}/scope.md",
+            "bytes": 120,
+            "artifact": true
+          },
+          {
+            "kind": "sniff",
+            "file": "{{workPath}}/scope.md",
+            "sniff": "nonempty",
+            "artifact": true
+          }
+        ],
+        "onReject": "scope",
+        "maxAttempts": 3
+      },
+      "next": "draft",
+      "toolPolicy": {
+        "disallowBuiltinToolsets": [
+          "ai-apps",
+          "archives",
+          "audio",
+          "browser-automation",
+          "code-execution",
+          "craftbooks",
+          "data-tables",
+          "entity-intel",
+          "git",
+          "image-intel",
+          "images",
+          "role-delegation",
+          "role-delegation-escalation",
+          "security-intel",
+          "team-management",
+          "videos",
+          "web",
+          "workspace-fs-write"
+        ],
+        "outputMedium": "artifact",
+        "additionalOutputMedia": [
+          "task-note"
+        ]
+      }
+    },
+    {
+      "id": "draft",
+      "name": "Draft the notes",
+      "description": "one personal note per owed person, in the household voice",
+      "prompt": "Draft the whole batch against the locked rules. Step 1: write one section per person in letters/thank-you.md — a heading with their name, then the note. Step 2: open each note with the specific gift or kindness from the roster in the first two sentences — the detail IS the thanks; 'thank you for the lovely gift' is a rejection. Step 3: vary the shape — if two notes could swap recipients without anyone noticing, rewrite one of them. Step 4: keep the household's voice throughout — warm, theirs, sized for how each note will be sent. Step 5: call write_task_note with the note count and anyone you had to defer.",
+      "suggestedRole": "copywriter",
+      "advanceWhen": {
+        "file": "letters/thank-you.md",
+        "minBytes": 1,
+        "sniff": "nonempty"
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "letters/thank-you.md",
+            "bytes": 1200
+          }
+        ],
+        "scripts": [
+          {
+            "name": "checkContains",
+            "scope": "standard",
+            "inputs": {
+              "file": "letters/thank-you.md",
+              "pattern": "(?:^|\\n)#{1,3}\\s+\\S",
+              "flags": "i"
+            }
+          }
+        ],
+        "onReject": "draft",
+        "maxAttempts": 3
+      },
+      "next": "log",
+      "toolPolicy": {
+        "disallowBuiltinToolsets": [
+          "ai-apps",
+          "archives",
+          "artifacts",
+          "audio",
+          "browser-automation",
+          "code-execution",
+          "craftbooks",
+          "data-tables",
+          "entity-intel",
+          "git",
+          "image-intel",
+          "images",
+          "role-delegation",
+          "role-delegation-escalation",
+          "security-intel",
+          "team-management",
+          "videos",
+          "web"
+        ],
+        "outputMedium": "workspace",
+        "additionalOutputMedia": [
+          "task-note"
+        ]
+      }
+    },
+    {
+      "id": "log",
+      "name": "Log the batch",
+      "description": "record each note in the roster, write the run summary",
+      "prompt": "Close the batch. Step 1: for each drafted note, record it against the person's entry through the project's correspondence capabilities (or note the exact roster update) so the owed-thanks tracking clears. Step 2: write the run summary — who got a note and what it names, who was deferred and why, and what still has to physically happen (addresses to find, stamps to buy). Step 3: keep it short and factual — this is a household ritual, not paperwork. Write {{workPath}}/report.md and call write_task_note with the drafted and deferred counts.\n\nThe deliverable `{{workPath}}/report.md` lands in the project's artifacts drawer — write it with `write_artifact` and read it back with `read_artifact`; the shipped workspace stays untouched.",
+      "suggestedRole": "archivist",
+      "advanceWhen": {
+        "file": "{{workPath}}/report.md",
+        "minBytes": 1,
+        "sniff": "nonempty",
+        "artifact": true
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "{{workPath}}/report.md",
+            "bytes": 600,
+            "artifact": true
+          },
+          {
+            "kind": "contains",
+            "file": "{{workPath}}/report.md",
+            "pattern": "(?:^|\\n)#{1,3}\\s+\\S",
+            "flags": "i",
+            "label": "at least one markdown heading",
+            "artifact": true
+          }
+        ],
+        "onReject": "log",
+        "maxAttempts": 4
+      },
+      "next": "evaluate",
+      "toolPolicy": {
+        "disallowBuiltinToolsets": [
+          "ai-apps",
+          "archives",
+          "audio",
+          "browser-automation",
+          "code-execution",
+          "craftbooks",
+          "data-tables",
+          "entity-intel",
+          "git",
+          "image-intel",
+          "images",
+          "role-delegation",
+          "role-delegation-escalation",
+          "security-intel",
+          "team-management",
+          "videos",
+          "web",
+          "workspace-fs-write"
+        ],
+        "outputMedium": "artifact",
+        "additionalOutputMedia": [
+          "task-note"
+        ]
+      }
+    },
+    {
+      "id": "evaluate",
+      "name": "Evaluate",
+      "description": "Grade the deliverable against every acceptance criterion. All pass → finish; any fail → loop back and fix the gap.",
+      "prompt": "Open letters/thank-you.md and {{workPath}}/report.md and verify every criterion from {{workPath}}/scope.md: every owed person in correspondents.json has a note or an explicit deferral with a reason, every note names the actual gift or kindness from the roster — spot-check each against the record, a note that thanks someone for the wrong thing is worse than none — no two notes are interchangeable, the voice matches the household's guide, and every drafted note is logged. Write PASS/FAIL per criterion; on any failure, name the person and the gap and loop back to draft.\n\nThe deliverable lives in the project's artifacts drawer — open `{{workPath}}/report.md` with `read_artifact`, not `read_file`.\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: \"draft\" })` 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": "{{workPath}}/report.md",
+          "artifact": true
+        }
+      ],
+      "next": "draft",
+      "toolPolicy": {
+        "disallowBuiltinToolsets": [
+          "ai-apps",
+          "archives",
+          "audio",
+          "browser-automation",
+          "code-execution",
+          "craftbooks",
+          "data-tables",
+          "entity-intel",
+          "git",
+          "image-intel",
+          "images",
+          "role-delegation",
+          "role-delegation-escalation",
+          "security-intel",
+          "team-management",
+          "videos",
+          "web",
+          "workspace-fs-write"
+        ],
+        "outputMedium": "task-note"
+      }
+    },
+    {
+      "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,
+      "toolPolicy": {
+        "disallowBuiltinToolsets": [
+          "ai-apps",
+          "archives",
+          "artifacts",
+          "audio",
+          "browser-automation",
+          "code-execution",
+          "craftbooks",
+          "data-tables",
+          "entity-intel",
+          "git",
+          "image-intel",
+          "images",
+          "role-delegation",
+          "role-delegation-escalation",
+          "security-intel",
+          "team-management",
+          "videos",
+          "web",
+          "workspace-fs-write"
+        ],
+        "outputMedium": "task-note"
+      }
+    }
+  ],
+  "version": "1.0.3",
+  "releasedAt": "2026-09-01T20:00:00Z",
+  "minGezelVersion": "1.26233"
+}

package/data/craftbook-templates/th/thank-you-batch/versions/1.0.3/test.json

@@ -0,0 +1,144 @@
+{
+  "schemaVersion": 1,
+  "title": "Thank-You Note Batch — owed list to logged notes",
+  "objective": "Measure whether the thank-you-batch craftbook turns a seeded correspondence roster and voice guide into one personal, specific note per owed person, logged with a run summary.",
+  "tags": [
+    "corpus"
+  ],
+  "prompt": "we owe thank yous from the aniversary party — the list is in correspondents.json and how we like to sound is in tasks/eval/voice.md. can you draft them all and log them?",
+  "setup": {
+    "projectName": "Vandermeer Correspondence Desk",
+    "about": "Correspondence desk for the Vandermeer household. The roster lives at workspace/correspondents.json; drafted letters go under workspace/letters. The household voice guide is workspace/notes/voice.md.",
+    "missionObjectives": "Keep the household's correspondence answered and warm. Notes are grounded in the roster's records — no live services, no real credentials.",
+    "files": [
+      {
+        "path": "correspondents.json",
+        "content": "{\n  \"household\": \"the Vandermeers\",\n  \"correspondents\": [\n    {\n      \"name\": \"Ruth Okafor\",\n      \"relation\": \"next-door neighbor\",\n      \"owedThanks\": true,\n      \"for\": \"the hand-quilted table runner she made for the anniversary\",\n      \"occasion\": \"40th anniversary party\"\n    },\n    {\n      \"name\": \"Stan Pollard\",\n      \"relation\": \"old fishing friend\",\n      \"owedThanks\": true,\n      \"for\": \"lending the big canvas tent and setting it up in the rain the night before\",\n      \"occasion\": \"40th anniversary party\"\n    },\n    {\n      \"name\": \"Mei Chu\",\n      \"relation\": \"friend from the pottery class\",\n      \"owedThanks\": true,\n      \"for\": \"the pair of hand-thrown teacups from her own studio\",\n      \"occasion\": \"40th anniversary party\"\n    },\n    {\n      \"name\": \"Deacon Ellis\",\n      \"relation\": \"nephew\",\n      \"owedThanks\": true,\n      \"for\": \"driving Grandma both ways and staying past midnight to help clean up\",\n      \"occasion\": \"40th anniversary party\"\n    },\n    {\n      \"name\": \"Priya Raman\",\n      \"relation\": \"college roommate\",\n      \"owedThanks\": true,\n      \"for\": \"the framed watercolor she painted of the old lake house\",\n      \"occasion\": \"40th anniversary party\"\n    },\n    {\n      \"name\": \"Tomas Herrera\",\n      \"relation\": \"across-the-street neighbor\",\n      \"owedThanks\": true,\n      \"for\": \"the case of orchard cider and the toast that made everyone cry\",\n      \"occasion\": \"40th anniversary party\"\n    },\n    {\n      \"name\": \"Gwen Fowler\",\n      \"relation\": \"cousin\",\n      \"owedThanks\": false,\n      \"note\": \"thanked in person for the flowers — nothing owed\"\n    }\n  ]\n}\n"
+      },
+      {
+        "path": "tasks/eval/voice.md",
+        "content": "# How we sound\n\n- We open every note with \"Dear\" and the person's first name. Always.\n- We sign off \"with love from the whole crew\" or \"yours over the fence\" for neighbors.\n- Short sentences. Warm. One small joke allowed per note, never two.\n- Always name the thing. \"Your lovely gift\" is banned in this house.\n- We never say \"esteemed\", \"heartfelt\", or anything that sounds like a greeting-card rack.\n"
+      }
+    ],
+    "worker": {
+      "name": "Lotte",
+      "role": "Letter writer"
+    },
+    "craftbookParams": {
+      "workPath": "tasks/eval"
+    }
+  },
+  "mocks": [],
+  "success": {
+    "summary": "letters/thank-you.md holds one personal note per owed person, each naming the actual gift or kindness in the household voice; tasks/eval/report.md logs the run.",
+    "deliverables": [
+      {
+        "path": "letters/thank-you.md",
+        "kind": "markdown-doc",
+        "minBytes": 1200,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "letters/thank-you.md",
+            "pattern": "(?:^|\\n)#{1,3}\\s+\\S"
+          },
+          {
+            "kind": "valuesSubsetOf",
+            "file": "letters/thank-you.md",
+            "sourceFiles": [
+              "correspondents.json"
+            ],
+            "pattern": "Dear\\s+([A-Z][A-Za-z'-]+)",
+            "minMatches": 6,
+            "label": "a salutation per owed person, names from the roster"
+          },
+          {
+            "kind": "contains",
+            "file": "letters/thank-you.md",
+            "pattern": "quilt|table runner",
+            "flags": "i",
+            "label": "Ruth's gift named"
+          },
+          {
+            "kind": "contains",
+            "file": "letters/thank-you.md",
+            "pattern": "tent|teacup",
+            "flags": "i",
+            "label": "Stan's or Mei's gift named"
+          },
+          {
+            "kind": "contains",
+            "file": "letters/thank-you.md",
+            "pattern": "watercolor|lake house|cider|toast",
+            "flags": "i",
+            "label": "Priya's or Tomas's gift named"
+          },
+          {
+            "kind": "contains",
+            "file": "letters/thank-you.md",
+            "pattern": "driv|midnight|clean",
+            "flags": "i",
+            "label": "Deacon's kindness named"
+          },
+          {
+            "kind": "judge",
+            "file": "letters/thank-you.md",
+            "rubric": "Each note names the giver's actual gift or kindness from the roster and follows the household voice guide; no two notes are interchangeable.",
+            "sourceFiles": [
+              "tasks/eval/voice.md",
+              "correspondents.json"
+            ]
+          }
+        ]
+      },
+      {
+        "path": "tasks/eval/report.md",
+        "kind": "markdown-report",
+        "minBytes": 600,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "tasks/eval/report.md",
+            "pattern": "(?:^|\\n)#{1,3}\\s+\\S"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/report.md",
+            "pattern": "logged|recorded",
+            "flags": "i",
+            "label": "notes logged against the roster"
+          }
+        ]
+      }
+    ]
+  },
+  "rubric": {
+    "artifact": {
+      "path": "letters/thank-you.md",
+      "kind": "markdown"
+    },
+    "axes": [
+      {
+        "name": "warmth",
+        "description": "The notes read like a person, not a template — warm and sized for how they will be sent."
+      },
+      {
+        "name": "specificity",
+        "description": "Each note names the actual gift or kindness from the roster in its opening lines."
+      },
+      {
+        "name": "voice",
+        "description": "The household voice guide is followed — openings, sign-offs, and the banned phrases stay banned."
+      },
+      {
+        "name": "completeness",
+        "description": "Every owed person is covered or explicitly deferred with a reason; nobody quietly dropped."
+      }
+    ]
+  },
+  "qualityFocus": [
+    "per-note specificity",
+    "household voice",
+    "full coverage"
+  ]
+}