@bendyline/gilde 0.1.38 → 0.1.39

package/data/craftbook-templates/qu/quarterly-summary/versions/1.0.2/test.json

@@ -0,0 +1,116 @@
+{
+  "schemaVersion": 1,
+  "title": "Quarterly Shoebox Summary smoke eval",
+  "objective": "Task-specific eval for the Quarterly Shoebox Summary craftbook: Q2 totals computed from the seeded ledger entries, the two entries lacking source documents named, and a plain-words estimated-payment reminder.",
+  "tags": [
+    "data"
+  ],
+  "prompt": "how did Q2 shake out for the shoebox? want the summery in tasks/eval/report.md with whats still missing paperwork",
+  "setup": {
+    "projectName": "Tax Shoebox — Quarterly Summary Eval",
+    "about": "Self-contained eval project for quarterly-summary. The household expense ledger lives at ledger.json in the workspace and covers the first two quarters of 2026; two Q2 entries are missing their source documents. The quarter close goes to tasks/eval/report.md.",
+    "missionObjectives": "Use the Quarterly Shoebox Summary craftbook/template, compute everything from the seeded ledger.json, and write tasks/eval/report.md — no network calls, real credentials, or live services.",
+    "files": [
+      {
+        "path": "ledger.json",
+        "content": "[\n  { \"date\": \"2026-01-15\", \"vendor\": \"Office Depot\", \"amountCents\": 21000, \"category\": \"office-supplies\", \"source\": \"inbox/2026-01-15-officedepot.txt\" },\n  { \"date\": \"2026-02-06\", \"vendor\": \"Cascade Air\", \"amountCents\": 29800, \"category\": \"travel\", \"source\": \"inbox/2026-02-06-cascadeair.txt\" },\n  { \"date\": \"2026-02-20\", \"vendor\": \"Peak Web Hosting\", \"amountCents\": 4900, \"category\": \"software\", \"source\": \"inbox/2026-02-20-peakweb.txt\" },\n  { \"date\": \"2026-03-12\", \"vendor\": \"Harbor Grill\", \"amountCents\": 5140, \"category\": \"meals\", \"source\": \"inbox/2026-03-12-harborgrill.txt\" },\n  { \"date\": \"2026-04-04\", \"vendor\": \"Peak Web Hosting\", \"amountCents\": 4900, \"category\": \"software\", \"source\": \"inbox/2026-04-04-peakweb.txt\" },\n  { \"date\": \"2026-04-18\", \"vendor\": \"Office Depot\", \"amountCents\": 8450, \"category\": \"office-supplies\", \"source\": \"inbox/2026-04-18-officedepot.txt\" },\n  { \"date\": \"2026-05-02\", \"vendor\": \"Cascade Air\", \"amountCents\": 41260, \"category\": \"travel\", \"source\": \"inbox/2026-05-02-cascadeair.txt\" },\n  { \"date\": \"2026-05-11\", \"vendor\": \"Harbor Grill\", \"amountCents\": 6220, \"category\": \"meals\" },\n  { \"date\": \"2026-05-30\", \"vendor\": \"Office Depot\", \"amountCents\": 14950, \"category\": \"office-supplies\", \"source\": \"inbox/2026-05-30-officedepot.txt\" },\n  { \"date\": \"2026-06-08\", \"vendor\": \"Peak Web Hosting\", \"amountCents\": 4900, \"category\": \"software\", \"source\": \"inbox/2026-06-08-peakweb.txt\" },\n  { \"date\": \"2026-06-21\", \"vendor\": \"Office Depot\", \"amountCents\": 12999, \"category\": \"office-supplies\" },\n  { \"date\": \"2026-06-27\", \"vendor\": \"Blue Line Transit\", \"amountCents\": 18875, \"category\": \"travel\", \"source\": \"inbox/2026-06-27-bluelinetransit.txt\" }\n]\n"
+      }
+    ],
+    "worker": {
+      "name": "Nora",
+      "role": "Data Analyst"
+    },
+    "craftbookParams": {
+      "workPath": "tasks/eval"
+    }
+  },
+  "mocks": [],
+  "success": {
+    "summary": "tasks/eval/report.md carries a category table whose totals re-add from the Q2 entries, names both entries missing paperwork, and compares against Q1 in plain words.",
+    "deliverables": [
+      {
+        "path": "tasks/eval/report.md",
+        "kind": "markdown-report",
+        "minBytes": 800,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "tasks/eval/report.md",
+            "pattern": "(?:^|\\n)#{1,3}\\s+\\S",
+            "flags": "i"
+          },
+          {
+            "kind": "tableShape",
+            "file": "tasks/eval/report.md",
+            "requiredColumns": [
+              "Category",
+              "Total"
+            ],
+            "minRows": 3
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/report.md",
+            "pattern": "363[.,]99",
+            "label": "office-supplies Q2 total computed correctly (363.99)"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/report.md",
+            "pattern": "601[.,]35",
+            "label": "travel Q2 total computed correctly (601.35)"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/report.md",
+            "pattern": "missing|documentation|paperwork|no source",
+            "flags": "i",
+            "label": "missing-documentation list present"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/report.md",
+            "pattern": "Harbor Grill",
+            "label": "the undocumented Harbor Grill entry is named"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/report.md",
+            "pattern": "Q2|second quarter",
+            "flags": "i",
+            "label": "the target quarter is stated"
+          }
+        ]
+      }
+    ]
+  },
+  "rubric": {
+    "artifact": {
+      "path": "tasks/eval/report.md",
+      "kind": "markdown"
+    },
+    "axes": [
+      {
+        "name": "accuracy",
+        "description": "Every figure re-adds from the ledger entries — no estimated or invented totals."
+      },
+      {
+        "name": "completeness",
+        "description": "Every Q2 entry is counted once, and both entries lacking source documents are named."
+      },
+      {
+        "name": "actionability",
+        "description": "The missing-paperwork list and next-quarter action are concrete enough to act on this week."
+      },
+      {
+        "name": "tone",
+        "description": "Calm and plain-words — a household ritual, not an audit finding."
+      }
+    ]
+  },
+  "qualityFocus": [
+    "computed category totals",
+    "missing-documentation sweep",
+    "plain-words payment reminder"
+  ]
+}

package/data/craftbook-templates/re/reading-list/versions/1.0.2/craftbook.json

@@ -0,0 +1,119 @@
+{
+  "id": "reading-list",
+  "name": "Curated Annotated Reading List",
+  "description": "Curate an annotated reading list on a topic for a defined reader: a planner sets the reader, their goal, and the selection and ordering criteria, then a researcher curates and annotates each pick with why it matters, what it covers, and a suggested reading order. The ordering matters because a list assembled without a reader and a goal becomes a random link dump, and one without annotations forces the reader to guess what to read first; fixing who it is for and what 'good for this reader' means up front turns it into a guided learning path. Use this for curated reading lists, a syllabus or learning path, 'what should I read to learn X', resource roundups, and annotated bibliographies of recommended materials.\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 list (planner) — reader, goal, and selection + ordering criteria → gated on artifact `{{workPath}}/scope.md` (markdown-notes)\n2. Curate & annotate (researcher) — selected, ordered, annotated picks for the reader → gated on `reading-list.md` (markdown-doc)\n\nThe gates never advance with an unmet criterion, and loop back to the\nowning phase to fix named gaps.\n",
+  "entryStepId": "scope",
+  "triggers": [
+    "curate a reading list",
+    "what should I read to learn",
+    "build a learning path",
+    "resource roundup on",
+    "annotated reading list"
+  ],
+  "steps": [
+    {
+      "id": "scope",
+      "name": "Scope the list",
+      "description": "reader, goal, and selection + ordering criteria",
+      "prompt": "1. Pin the reader and their goal in one sentence (e.g. 'a working dev who wants a practical intro to distributed systems') and their current level. 2. Set selection criteria: what makes a resource worth including for THIS reader (authority, level fit, recency, format mix) and how many items (aim 6-12). 3. Decide the ordering principle — a reading ORDER (foundational -> advanced, or a sequenced path), not just an unordered set — and any grouping (e.g. start here / go deeper / reference). 4. Lock the per-item annotation template: title, author, type/length, why it is on the list, what it covers, level, and where it fits in the order. 5. Write an acceptance-criteria checklist (e.g. 'each item has a full annotation', 'items are sequenced into a sensible reading order', 'a mix of levels/formats appropriate to the reader', 'each pick is justified for this reader specifically', 'links/locators are included'). 6. Record decisions. Call write_task_note with the reader, criteria, ordering, template, and checklist; write the same to {{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": "planner",
+      "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": "curate"
+    },
+    {
+      "id": "curate",
+      "name": "Curate & annotate",
+      "description": "selected, ordered, annotated picks for the reader",
+      "prompt": "1. Find resources that meet the selection criteria for the specified reader, preferring authoritative and well-regarded materials at the right level. 2. Write reading-list.md grouped/ordered per the locked ordering principle, with each item fully annotated to the template: title, author, type & rough length, why it earns a spot for THIS reader, what it covers, its level, and a working link or locator. 3. Sequence the items into an explicit reading order (number them or use 'start here / next / deeper / reference' groupings) so the reader knows what to read first. 4. Justify each pick against the reader's goal — drop anything you cannot justify rather than padding the list. 5. Include a deliberate mix of levels and formats as scoped, and verify each link/locator is real. 6. On a loop-back, fix only the named gaps. 7. Call write_task_note with the list path and item count.",
+      "suggestedRole": "researcher",
+      "advanceWhen": {
+        "file": "reading-list.md",
+        "minBytes": 1,
+        "sniff": "nonempty"
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "reading-list.md",
+            "bytes": 500
+          }
+        ],
+        "scripts": [
+          {
+            "name": "checkContains",
+            "scope": "standard",
+            "inputs": {
+              "file": "reading-list.md",
+              "pattern": "(?:^|\\n)#{1,3}\\s+\\S",
+              "flags": "i"
+            }
+          }
+        ],
+        "onReject": "curate",
+        "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": "Open reading-list.md and check EACH criterion PASS/FAIL: (1) every item has a full annotation following the template from {{workPath}}/scope.md; (2) items are sequenced into an explicit, sensible reading order (not an unordered dump); (3) each pick is justified for the specified reader and goal, not generic; (4) the mix of levels/formats matches the scope and suits the reader; (5) every item has a working link or locator (spot-check 3); (6) the list is the scoped length with no padding. A random link dump, or items with no annotation or no justification for this reader, FAILS.\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: \"curate\" })` 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": "reading-list.md"
+        }
+      ],
+      "next": "curate"
+    },
+    {
+      "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": "1.0.2",
+  "releasedAt": "2026-08-21T00:00:00Z",
+  "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}}"
+      }
+    }
+  },
+  "minGezelVersion": "1.26233"
+}

package/data/craftbook-templates/re/reading-list/versions/1.0.2/test.json

@@ -0,0 +1,113 @@
+{
+  "schemaVersion": 1,
+  "title": "Curated Annotated Reading List smoke eval",
+  "objective": "Self-contained smoke eval for the Curated Annotated Reading List craftbook using the corpus generic harness.",
+  "tags": [
+    "corpus"
+  ],
+  "prompt": "We need a curated annotated reading list for the returns-desk work. The backgound info is in source/brief.md — have a look and write it up in report.md.",
+  "setup": {
+    "projectName": "Curated Annotated Reading List Eval",
+    "about": "Self-contained eval project for reading-list. Seeded inputs are under workspace/source or workspace/fixtures; final deliverable is workspace/report.md.",
+    "missionObjectives": "Use the Curated Annotated Reading List craftbook/template, read the seeded local fixtures, and write report.md without network calls, real credentials, or live services.",
+    "files": [
+      {
+        "path": "source/brief.md",
+        "content": "# Curated Annotated Reading List Eval Brief\n\nClient: Boreal Desk, a home-office accessories company.\nAudience: operations leads who need an artifact they can use this week.\n\nFixed source facts for grounding:\n- The returns desk pilot covered 18 SKUs.\n- Median first response improved from 18 hours to 6 hours.\n- Preventable refund leakage fell from 14.2% to 8.9%.\n- The top unresolved complaint is status silence after photo submission.\n- Required next actions are automated status emails, barcode-exception training, and a weekly Finance exception export.\n\nUse these facts when the task asks for prose, analysis, copy, UI content, or test data. Do not use live web services, real credentials, or current outside data.\n\nCraftbook under test: reading-list - Curated Annotated Reading List.\n"
+      }
+    ],
+    "worker": {
+      "name": "Jules",
+      "role": "Writer"
+    },
+    "craftbookParams": {
+      "workPath": "tasks/eval"
+    }
+  },
+  "mocks": [],
+  "success": {
+    "summary": "report.md is a source-grounded markdown artifact with required facts and actions.",
+    "deliverables": [
+      {
+        "path": "report.md",
+        "kind": "markdown-report",
+        "minBytes": 1000,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "report.md",
+            "pattern": "(?:^|\\n)#{1,3}\\s+\\S",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "report.md",
+            "pattern": "source/brief\\.md|Brief",
+            "flags": "i",
+            "label": "cite or name the seeded source"
+          },
+          {
+            "kind": "contains",
+            "file": "report.md",
+            "pattern": "18\\b[\\s\\S]{0,30}\\bSKUs",
+            "flags": "i",
+            "label": "include the 18-SKU pilot scope"
+          },
+          {
+            "kind": "contains",
+            "file": "report.md",
+            "pattern": "18\\s*hours|6\\s*hours",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "report.md",
+            "pattern": "14\\.2%|8\\.9%",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "report.md",
+            "pattern": "status silence|photo submission",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "report.md",
+            "pattern": "automated status emails|barcode|Finance",
+            "flags": "i"
+          }
+        ]
+      }
+    ]
+  },
+  "rubric": {
+    "artifact": {
+      "path": "report.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": [
+    "source-grounded prose",
+    "section structure",
+    "next-action clarity"
+  ]
+}

package/data/craftbook-templates/re/receipt-intake/versions/1.0.2/craftbook.json

@@ -0,0 +1,152 @@
+{
+  "id": "receipt-intake",
+  "name": "Receipt Intake",
+  "description": "Empty the shoebox inbox properly: every waiting receipt becomes a dated, categorized ledger entry that names its source file, likely duplicates get flagged instead of double-counted, and the genuinely ambiguous ones become owner questions instead of guesses. Locks the intake rules FIRST, then works the inbox receipt by receipt, then reports the totals and the question list — the year-round habit that makes April an afternoon. Use for receipt intake, emptying the receipt inbox, 'get these into the ledger', and monthly shoebox catch-ups.\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 intake (clerk) — inbox listing, ledger categories, intake rules locked → gated on artifact `{{workPath}}/scope.md` (markdown-notes)\n2. Work the inbox (data-analyst) — one entry per receipt, duplicates flagged, questions parked → gated on artifact `{{workPath}}/intake.md` (markdown-notes)\n3. Write the intake report (copywriter) — counts, totals by category, the question list → 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": [
+    "process the receipt inbox",
+    "get these receipts into the ledger",
+    "empty the shoebox",
+    "log this month's receipts",
+    "receipt intake"
+  ],
+  "steps": [
+    {
+      "id": "scope",
+      "name": "Scope the intake",
+      "description": "inbox listing, ledger categories, intake rules locked",
+      "prompt": "Define the intake before recording a single entry. Step 1: list the receipt notes waiting in inbox/ and read ledger.json from the workspace — the existing categories, the recent entries, and how entries are shaped. Step 2: lock the intake rules — one ledger entry per receipt; every entry records the source file it came from; a receipt that looks like one already in the ledger (same vendor, date, and amount) is flagged as a likely duplicate, never silently entered again; a receipt whose category is genuinely unclear becomes a question for the owner instead of a guess. Step 3: write an acceptance-criteria checklist ('every inbox receipt produced an entry, a duplicate flag, or an owner question', 'every new entry carries date, amount, category, and source', 'amounts copied exactly from the receipt, never rounded or invented', 'the report totals what was processed and lists every question'). Call write_task_note with the inbox count + rules + checklist and write {{workPath}}/scope.md. No entries yet.\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": "clerk",
+      "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": "extract"
+    },
+    {
+      "id": "extract",
+      "name": "Work the inbox",
+      "description": "one entry per receipt, duplicates flagged, questions parked",
+      "prompt": "Work the inbox against the locked rules, one receipt at a time. Step 1: read each receipt note in inbox/ and pull the vendor, date, and exact amount, and pick the most sensible category from the ledger's existing set. Step 2: record each clear receipt as a ledger entry through the project's ledger capabilities — date, amount, category, and the source file it came from. Step 3: where a receipt matches an existing entry on vendor, date, and amount, flag it as a likely duplicate and do not enter it again; where the category is genuinely ambiguous, park it as an owner question with your best two candidate categories instead of guessing. Step 4: keep a running log in {{workPath}}/intake.md — per receipt: what it was, what you did (entered, flagged, question), and why. Call write_task_note with the entered/flagged/question counts.\n\nThe deliverable `{{workPath}}/intake.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": "data-analyst",
+      "advanceWhen": {
+        "file": "{{workPath}}/intake.md",
+        "minBytes": 1,
+        "sniff": "nonempty",
+        "artifact": true
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "{{workPath}}/intake.md",
+            "bytes": 120,
+            "artifact": true
+          },
+          {
+            "kind": "sniff",
+            "file": "{{workPath}}/intake.md",
+            "sniff": "nonempty",
+            "artifact": true
+          }
+        ],
+        "onReject": "extract",
+        "maxAttempts": 3
+      },
+      "next": "report"
+    },
+    {
+      "id": "report",
+      "name": "Write the intake report",
+      "description": "counts, totals by category, the question list",
+      "prompt": "Write the intake report from the run log. Step 1: open with the headline — how many receipts came in, how many became entries, how many were flagged or parked as questions. Step 2: give the totals of what was entered this run, by category, computed from the entries themselves. Step 3: list every owner question with the receipt it belongs to and your best candidate answers, so the owner can clear the list in one pass. Step 4: note anything that looked off — a likely duplicate, an amount that seems wrong for its vendor, a receipt too faded to trust — plainly and without alarm. Step 5: keep it short and warm; this is a chore done well, not an audit. Write {{workPath}}/report.md and call write_task_note with the entered/flagged/question 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": "copywriter",
+      "advanceWhen": {
+        "file": "{{workPath}}/report.md",
+        "minBytes": 1,
+        "sniff": "nonempty",
+        "artifact": true
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "{{workPath}}/report.md",
+            "bytes": 1000,
+            "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": "report",
+        "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": "Open {{workPath}}/report.md, {{workPath}}/intake.md, and the ledger and verify every criterion from {{workPath}}/scope.md: every inbox receipt is accounted for as an entry, a duplicate flag, or an owner question; every new entry carries date, amount, category, and its source file; every entered amount matches its receipt exactly (spot-check three against the inbox notes); the report's totals trace to the entries; and every owner question appears in the report. Write PASS/FAIL per criterion; on any failure, name the gap and loop back to report.\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: \"report\" })` 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": "report"
+    },
+    {
+      "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": "1.0.2",
+  "releasedAt": "2026-08-21T00:00:00Z",
+  "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}}"
+      }
+    }
+  },
+  "minGezelVersion": "1.26233"
+}