@bendyline/gilde 0.1.26 → 0.1.28

package/data/craftbook-templates/an/annotated-bibliography/versions/1.0.1/test.json

@@ -0,0 +1,102 @@
+{
+  "schemaVersion": 1,
+  "title": "Annotated bibliography from seeded source cards",
+  "objective": "Measure whether the annotated-bibliography craftbook locks style/rubric and produces uniform summary, evaluation, and relevance annotations.",
+  "tags": [
+    "data"
+  ],
+  "prompt": "In the `Annotated Bibliography Eval` project, read `source/source-cards.md` and write `bibliography.md` for a research question about rural telehealth follow-up programs. Use APA-style entries and give every source a Summary, Critical evaluation, and Relevance annotation. Do not invent sources beyond the six seeded cards.",
+  "setup": {
+    "projectName": "Annotated Bibliography Eval",
+    "about": "Self-contained eval project for annotated-bibliography. Six source cards are in workspace/source/source-cards.md.",
+    "missionObjectives": "Write a six-entry APA-style annotated bibliography with uniform three-part annotations and no invented sources.",
+    "files": [
+      {
+        "path": "source/source-cards.md",
+        "content": "# Source cards\n\nResearch question: What makes rural telehealth follow-up programs effective after hospital discharge?\n\n1. Allen, Priya. 2022. Rural Telehealth Follow-up and Readmission Risk. Journal of Rural Care. Finds video follow-up within 72 hours lowered 30-day readmissions by 11% across four clinics; limitation: observational design.\n2. Chen, Marco. 2021. Broadband Gaps in Remote Patient Monitoring. Health Systems Review. Surveys 900 households; shows connectivity failures explain many missed visits; limitation: self-reported bandwidth.\n3. Dunn, Elise. 2023. Nurse-led Coaching After Discharge. Care Transitions Quarterly. Randomized pilot, n=240; nurse coaching improved medication adherence by 18 points; limitation: single region.\n4. Garcia, Alma. 2020. Telehealth Trust in Older Adults. Community Health Methods. Qualitative interviews with 42 older adults; highlights trust, language access, and caregiver presence.\n5. Patel, Neel. 2024. Cost Model for Hybrid Follow-up. Rural Health Economics. Compares phone, video, and in-person follow-ups; finds hybrid model breaks even after 1,600 annual encounters.\n6. Williams, June. 2019. Privacy Concerns in Home Monitoring. Nursing Informatics. Reviews consent and device-sharing risks; older but still useful for privacy checklist.\n"
+      }
+    ],
+    "worker": {
+      "name": "Bente",
+      "role": "Researcher"
+    }
+  },
+  "mocks": [],
+  "success": {
+    "summary": "workspace/bibliography.md has six source-grounded entries with APA-style citation data plus summary, credibility/method evaluation, and relevance notes.",
+    "deliverables": [
+      {
+        "path": "bibliography.md",
+        "kind": "markdown-doc",
+        "minBytes": 2600,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "bibliography.md",
+            "pattern": "(?:^|\\n)#{1,3}\\s+\\S",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "bibliography.md",
+            "pattern": "Allen|Chen|Dunn|Garcia|Patel|Williams",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "bibliography.md",
+            "pattern": "Summary|Critical evaluation|Relevance",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "bibliography.md",
+            "pattern": "APA|telehealth|rural|readmission",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "bibliography.md",
+            "pattern": "observational|limitation|bias|method",
+            "flags": "i"
+          },
+          {
+            "kind": "notContains",
+            "file": "bibliography.md",
+            "pattern": "Retrieved from Wikipedia|unknown author",
+            "flags": "i"
+          }
+        ]
+      }
+    ]
+  },
+  "rubric": {
+    "artifact": {
+      "path": "bibliography.md",
+      "kind": "markdown"
+    },
+    "axes": [
+      {
+        "name": "accuracy",
+        "description": "Computed figures are correct against the seeded records."
+      },
+      {
+        "name": "grounding",
+        "description": "Every claim traces to the seeded data — nothing invented."
+      },
+      {
+        "name": "structure",
+        "description": "Tables and sections make the analysis checkable at a glance."
+      },
+      {
+        "name": "actionability",
+        "description": "Recommendations follow from the findings and are concretely doable."
+      }
+    ]
+  },
+  "qualityFocus": [
+    "rubric uniformity",
+    "source coverage",
+    "critical evaluation"
+  ]
+}

package/data/craftbook-templates/an/annual-document-review/versions/1.0.1/craftbook.json

@@ -0,0 +1,140 @@
+{
+  "id": "annual-document-review",
+  "name": "Annual Document Review",
+  "description": "Walk a life binder's document roster once a year: inventory what is present and missing, confirm each present document is still current (right beneficiaries, right address, not expired), record the review so staleness tracking resets, and produce a prioritized to-do for the gaps. Locks the currency criteria per document kind FIRST, then verifies each record, then reports — so the binder's completeness claim is re-earned annually instead of rotting silently. Use for annual life-binder reviews, estate document checkups, 'is our paperwork still right', and insurance/beneficiary reviews.\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 review (reviewer) — inventory the roster, currency criteria per document kind → gated on artifact `notes/scope.md` (markdown-notes)\n2. Verify each document (reviewer) — walk the roster with the owner, record reviews, list gaps → gated on artifact `notes/verify.md` (markdown-notes)\n3. Write the binder report (copywriter) — binder health report with a ranked to-do → gated on artifact `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": [
+    "annual document review",
+    "review the binder",
+    "check our documents",
+    "is our paperwork current",
+    "beneficiary review"
+  ],
+  "steps": [
+    {
+      "id": "scope",
+      "name": "Scope the review",
+      "description": "inventory the roster, currency criteria per document kind",
+      "prompt": "Define the review before checking anything. Step 1: read documents.json from the workspace and inventory the roster — which documents are present, which are missing, and when each present one was last reviewed. Step 2: define the currency criteria per document kind — e.g. a will is current if it reflects today's family situation and executor; insurance is current if coverage amounts and beneficiaries are right; contact lists and password locations are current if verified within the year; anything expired or superseded is not current. Step 3: define what counts as done for a missing document — a concrete next step (who to call, what to draft), not just 'get one'. Step 4: write an acceptance-criteria checklist ('every roster record verified with the owner or explicitly deferred', 'every verified document gets its review recorded so staleness resets', 'every missing document has a named next step', 'the report ranks gaps by consequence, not alphabetically'). Call write_task_note with the inventory counts + criteria + checklist and write notes/scope.md.\n\nThe deliverable `notes/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": "reviewer",
+      "advanceWhen": {
+        "file": "notes/scope.md",
+        "minBytes": 1,
+        "sniff": "nonempty",
+        "artifact": true
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "notes/scope.md",
+            "bytes": 120,
+            "artifact": true
+          },
+          {
+            "kind": "sniff",
+            "file": "notes/scope.md",
+            "sniff": "nonempty",
+            "artifact": true
+          }
+        ],
+        "onReject": "scope",
+        "maxAttempts": 3
+      },
+      "next": "verify"
+    },
+    {
+      "id": "verify",
+      "name": "Verify each document",
+      "description": "walk the roster with the owner, record reviews, list gaps",
+      "prompt": "Verify the roster against the locked criteria. Step 1: walk each present document with the owner — ask the kind-specific currency questions from the scope and record the answer. Step 2: for each document confirmed current, record the review through the project's binder capabilities (or note the update for documents.json) so the reviewedAt tracking resets. Step 3: for each document that fails its criteria, record what specifically is out of date. Step 4: for each missing document, confirm the next step with the owner and capture it. Step 5: write the structured results — verified, out-of-date (with why), missing (with next step) — to notes/verify.md, and call write_task_note with the counts per bucket.\n\nThe deliverable `notes/verify.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": "reviewer",
+      "advanceWhen": {
+        "file": "notes/verify.md",
+        "minBytes": 1,
+        "sniff": "nonempty",
+        "artifact": true
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "notes/verify.md",
+            "bytes": 120,
+            "artifact": true
+          },
+          {
+            "kind": "sniff",
+            "file": "notes/verify.md",
+            "sniff": "nonempty",
+            "artifact": true
+          }
+        ],
+        "onReject": "verify",
+        "maxAttempts": 3
+      },
+      "next": "report"
+    },
+    {
+      "id": "report",
+      "name": "Write the binder report",
+      "description": "binder health report with a ranked to-do",
+      "prompt": "Write the binder health report from the verification. Step 1: open with the headline — how many documents are in place and current, out of the full roster. Step 2: list what was verified current this review, one line each. Step 3: give the ranked to-do — out-of-date and missing documents ordered by consequence (what happens to the family if this one is needed and wrong), each with its concrete next step and who acts. Step 4: note the next review date one year out. Step 5: keep the tone calm and practical — the binder exists to make a hard day easier, and the report should read that way. Write report.md and call write_task_note with the report path and the to-do count.\n\nThe deliverable `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": "report.md",
+        "minBytes": 1,
+        "sniff": "nonempty",
+        "artifact": true
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "report.md",
+            "bytes": 1500,
+            "artifact": true
+          },
+          {
+            "kind": "contains",
+            "file": "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 report.md and notes/verify.md and verify every criterion from notes/scope.md: every roster record was verified or explicitly deferred, every verified document had its review recorded (spot-check two against the roster), every missing or out-of-date document carries a concrete named next step, and the to-do is ranked by consequence with a stated rationale. 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 `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": "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": "1.0.1",
+  "releasedAt": "2026-08-14T00:00:00Z"
+}

package/data/craftbook-templates/an/annual-document-review/versions/1.0.1/test.json

@@ -0,0 +1,110 @@
+{
+  "schemaVersion": 1,
+  "title": "Annual Document Review smoke eval",
+  "objective": "Self-contained smoke eval for the Annual Document Review craftbook using the corpus generic harness.",
+  "tags": [
+    "corpus"
+  ],
+  "prompt": "We need a annual document review 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": "Annual Document Review Eval",
+    "about": "Self-contained eval project for annual-document-review. Seeded inputs are under workspace/source or workspace/fixtures; final deliverable is workspace/report.md.",
+    "missionObjectives": "Use the Annual Document Review 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": "# Annual Document Review 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: annual-document-review - Annual Document Review.\n"
+      }
+    ],
+    "worker": {
+      "name": "Jules",
+      "role": "Writer"
+    }
+  },
+  "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/an/anomaly-scan/versions/1.0.1/craftbook.json

@@ -0,0 +1,141 @@
+{
+  "id": "anomaly-scan",
+  "name": "Anomaly / Outlier Scan",
+  "description": "Scan a dataset for anomalies and outliers and report what is genuinely unusual versus expected noise. Defines the detection criteria and thresholds FIRST (statistical and rule-based, with an explicit definition of 'anomalous') so findings are reproducible, then runs the scan to flag and rank suspect records, then writes a prioritized report a human can triage. Use this for outlier detection, anomaly scanning, finding bad records, fraud/error spotting, or surfacing data points that don't fit.\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. Define detection criteria (data-analyst) — lock what counts as anomalous and the thresholds → gated on artifact `notes/criteria.md` (markdown-notes)\n2. Run the scan (developer) — apply the criteria and emit ranked flagged records → gated on `data/anomalies.json` (json)\n3. Report the findings (copywriter) — write a prioritized, triage-ready anomaly report → gated on artifact `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": "criteria",
+  "triggers": [
+    "find anomalies",
+    "detect outliers",
+    "scan for bad records",
+    "spot unusual data",
+    "outlier detection"
+  ],
+  "steps": [
+    {
+      "id": "criteria",
+      "name": "Define detection criteria",
+      "description": "lock what counts as anomalous and the thresholds",
+      "prompt": "Define exactly what 'anomalous' means before scanning, so results are reproducible. Step 1: choose which columns/metrics to scan and the expected normal behavior of each. Step 2: pick detection methods per column — statistical (z-score > 3, IQR fence at 1.5x, percentile cutoffs) for numerics; rule-based (impossible/contradictory values, future dates, out-of-domain categories, broken referential integrity) where rules apply. Step 3: set concrete thresholds and a severity scale (e.g. high = impossible value or z > 4; medium = z 3-4). Step 4: decide how to rank flagged records. Step 5: write an acceptance-criteria checklist ('every method has a stated threshold', 'each flag records the rule it tripped and a severity', 'findings are ranked', 'false-positive rate is considered'). Call write_task_note and write the methods + thresholds + checklist to notes/criteria.md. No scanning yet.\n\nThe deliverable `notes/criteria.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": "notes/criteria.md",
+        "minBytes": 1,
+        "sniff": "nonempty",
+        "artifact": true
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "notes/criteria.md",
+            "bytes": 120,
+            "artifact": true
+          },
+          {
+            "kind": "sniff",
+            "file": "notes/criteria.md",
+            "sniff": "nonempty",
+            "artifact": true
+          }
+        ],
+        "onReject": "criteria",
+        "maxAttempts": 3
+      },
+      "next": "scan"
+    },
+    {
+      "id": "scan",
+      "name": "Run the scan",
+      "description": "apply the criteria and emit ranked flagged records",
+      "prompt": "Apply the detection criteria to the full dataset deterministically. Step 1: compute the per-column statistics the thresholds need (mean/stdev, quartiles, value-domain sets). Step 2: evaluate every record against each method and collect the flags it trips, with the measured value and the threshold it breached. Step 3: assign each flagged record a severity and a reason string. Step 4: rank flagged records by severity then magnitude. Step 5: emit data/anomalies.json — a ranked list of { record id/row, column, value, rule, threshold, severity } plus a summary count by severity. Do not flag records that don't breach a stated threshold. Call write_task_note with the counts by severity and the top findings.",
+      "suggestedRole": "developer",
+      "advanceWhen": {
+        "file": "data/anomalies.json",
+        "minBytes": 1,
+        "sniff": "json-valid"
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "data/anomalies.json",
+            "bytes": 2
+          }
+        ],
+        "scripts": [
+          {
+            "name": "checkJsonValid",
+            "scope": "standard",
+            "inputs": {
+              "file": "data/anomalies.json"
+            }
+          }
+        ],
+        "onReject": "scan",
+        "maxAttempts": 3
+      },
+      "next": "report"
+    },
+    {
+      "id": "report",
+      "name": "Report the findings",
+      "description": "write a prioritized, triage-ready anomaly report",
+      "prompt": "Turn the flagged records into a report a human can triage — use only what the scan found. Step 1: open with a summary: how many records scanned, how many flagged, and the breakdown by severity. Step 2: list the top findings highest-severity first, each with the record, the value, what's wrong, and why it tripped (the rule + threshold). Step 3: group repeated patterns ('14 rows have a negative price') rather than listing each identically. Step 4: separate likely-real anomalies from probable noise/false-positives and say which deserve action. Step 5: end with recommended next steps. Write the report to report.md and call write_task_note with the path. On a loop-back, fix only the named gaps.\n\nThe deliverable `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": "report.md",
+        "minBytes": 1,
+        "sniff": "nonempty",
+        "artifact": true
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "report.md",
+            "bytes": 1500,
+            "artifact": true
+          },
+          {
+            "kind": "contains",
+            "file": "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": "Read report.md and cross-check it against notes/criteria.md and data/anomalies.json. Verify: every detection method named a concrete threshold; each reported flag records the rule it tripped, the value, and a severity; findings are ranked highest-severity first; repeated patterns are grouped, not duplicated; and likely false-positives are separated from real anomalies. Re-derive two flagged records from the criteria to confirm they genuinely breach a threshold (no spurious flags). Write PASS/FAIL per criterion; any FAIL loops back to report.\n\nThe deliverable lives in the project's artifacts drawer — open `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": "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": "1.0.1",
+  "releasedAt": "2026-08-14T00:00:00Z"
+}

package/data/craftbook-templates/an/anomaly-scan/versions/1.0.1/test.json

@@ -0,0 +1,120 @@
+{
+  "schemaVersion": 1,
+  "title": "Anomaly scan over seeded orders data",
+  "objective": "Measure whether the anomaly-scan craftbook defines reproducible criteria, emits machine-readable flags, and writes a triage-ready report.",
+  "tags": [
+    "data"
+  ],
+  "prompt": "In the `Anomaly Scan Eval` project, read `data/orders.csv`, define concrete anomaly thresholds, write `data/anomalies.json`, and then write `report.md` ranking the findings. The report must separate likely real problems from probable noise.",
+  "setup": {
+    "projectName": "Anomaly Scan Eval",
+    "about": "Self-contained eval project for anomaly-scan. The source dataset is workspace/data/orders.csv.",
+    "missionObjectives": "Find and explain anomalous order rows with explicit rules, JSON flags, severities, and a triage report.",
+    "files": [
+      {
+        "path": "data/orders.csv",
+        "content": "order_id,order_date,sku,quantity,revenue,region\nO-1001,2026-05-01,CHAIR-1,2,240,West\nO-1002,2026-05-01,LAMP-2,1,80,East\nO-1003,2026-05-02,DESK-5,1,620,West\nO-1004,2026-05-02,CHAIR-1,3,360,South\nO-1005,2026-05-03,LAMP-2,2,160,East\nO-1006,2026-05-03,DESK-5,1,610,West\nO-1007,2026-05-04,CHAIR-1,4,-480,West\nO-1008,2026-05-04,LAMP-2,1,82,North\nO-1009,2026-05-05,DESK-5,50,30500,West\nO-1010,2028-01-01,LAMP-2,1,80,East\nO-1011,2026-05-06,CHAIR-1,2,242,South\nO-1012,2026-05-06,DESK-5,1,615,West\nO-1013,2026-05-07,LAMP-2,0,0,East\n"
+      }
+    ],
+    "worker": {
+      "name": "Cato",
+      "role": "Data Analyst"
+    }
+  },
+  "mocks": [],
+  "success": {
+    "summary": "workspace/report.md and data/anomalies.json identify the seeded anomalous rows with rules, thresholds, severity, and triage guidance.",
+    "deliverables": [
+      {
+        "path": "report.md",
+        "kind": "markdown-report",
+        "minBytes": 1800,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "report.md",
+            "pattern": "(?:^|\\n)#{1,3}\\s+\\S",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "report.md",
+            "pattern": "O-1007|O-1009|O-1010|O-1013",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "report.md",
+            "pattern": "negative|future|quantity|outlier|threshold|severity",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "report.md",
+            "pattern": "likely real|probable noise|false positive",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "report.md",
+            "pattern": "recommended next steps|triage|action",
+            "flags": "i"
+          }
+        ]
+      },
+      {
+        "path": "data/anomalies.json",
+        "kind": "json",
+        "minBytes": 300,
+        "checks": [
+          {
+            "kind": "sniff",
+            "file": "data/anomalies.json",
+            "sniff": "json-valid"
+          },
+          {
+            "kind": "contains",
+            "file": "data/anomalies.json",
+            "pattern": "O-1007|O-1009|O-1010|O-1013",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "data/anomalies.json",
+            "pattern": "severity|rule|threshold|summary",
+            "flags": "i"
+          }
+        ]
+      }
+    ]
+  },
+  "rubric": {
+    "artifact": {
+      "path": "report.md",
+      "kind": "markdown"
+    },
+    "axes": [
+      {
+        "name": "accuracy",
+        "description": "Computed figures are correct against the seeded records."
+      },
+      {
+        "name": "grounding",
+        "description": "Every claim traces to the seeded data — nothing invented."
+      },
+      {
+        "name": "structure",
+        "description": "Tables and sections make the analysis checkable at a glance."
+      },
+      {
+        "name": "actionability",
+        "description": "Recommendations follow from the findings and are concretely doable."
+      }
+    ]
+  },
+  "qualityFocus": [
+    "criteria first",
+    "JSON flags",
+    "ranked triage report"
+  ]
+}