@bendyline/gilde 0.1.38 → 0.1.39

package/data/craftbook-templates/mo/morning-brief/versions/1.0.2/test.json

@@ -0,0 +1,228 @@
+{
+  "schemaVersion": 1,
+  "title": "Morning Topic Brief smoke eval",
+  "objective": "Task-specific eval for the Morning Topic Brief craftbook: two watched topics come from a live fake feed service (eight items, one duplicate of yesterday, one single-sourced rumor), and the dated brief lands in briefs/ with sources cited and the rumor flagged.",
+  "tags": [
+    "external"
+  ],
+  "prompt": "morning — can you pull the feeds and write todays breif? its july 16 so it goes in briefs/2026-07-16.md next to yesterdays. the fake feed service is wired up with a probe script",
+  "setup": {
+    "projectName": "Topic Watch — Morning Brief Eval",
+    "about": "Self-contained eval project for morning-brief. The feeds are wired as a live fake service — endpoints in mocks/services.md, and the ready-made mock-probe project script fetches every watched topic's items. Watched topics are seeded at topics.json and yesterday's brief sits in briefs/; today's brief goes to briefs/2026-07-16.md.",
+    "missionObjectives": "Use the Morning Topic Brief craftbook/template, pull the current items for each topic in topics.json from the fake feed service through the provided probe, and write briefs/2026-07-16.md. No real feeds, services, or credentials — only the wired mock.",
+    "files": [
+      {
+        "path": "topics.json",
+        "content": "[\n  {\n    \"id\": \"municipal-broadband\",\n    \"name\": \"Municipal Broadband\",\n    \"note\": \"the city fiber question - council, funding, pilots\"\n  },\n  {\n    \"id\": \"battery-recycling\",\n    \"name\": \"Battery Recycling\",\n    \"note\": \"collection programs, recovery tech, local depots\"\n  }\n]\n"
+      },
+      {
+        "path": "briefs/2026-07-15.md",
+        "content": "# Morning brief - 2026-07-15\n\n## Municipal Broadband\n- Council delays the broadband vote to August, citing the unfinished cost study. (Newsline - https://newsline.example/broadband/vote-delay)\n- Westport's pilot posts its first uptime report: 99.2% over 90 days. (Capitol Wire - https://capitolwire.example/fiber/westport-uptime)\n\n## Battery Recycling\n- County adds two weekend drop-off depots after the spring pilot. (Harbor Bulletin - https://harborbulletin.example/recycling/depots-added)\n"
+      }
+    ],
+    "worker": {
+      "name": "Isa",
+      "role": "Editor"
+    },
+    "craftbookParams": {
+      "workPath": "tasks/eval"
+    }
+  },
+  "mocks": [
+    {
+      "kind": "http",
+      "id": "feeds",
+      "description": "Fake feed service for this eval. GET /feeds/municipal-broadband and GET /feeds/battery-recycling each return the topic's current items with title, source, url, and confidence notes.",
+      "credential": {
+        "name": "mock.feeds"
+      },
+      "routes": [
+        {
+          "method": "GET",
+          "path": "/feeds/municipal-broadband",
+          "body": {
+            "topic": "municipal-broadband",
+            "fetched": "2026-07-16T06:30:00Z",
+            "items": [
+              {
+                "title": "Council pushes broadband vote to August",
+                "source": "Newsline",
+                "url": "https://newsline.example/broadband/vote-delay",
+                "published": "2026-07-15T21:10:00Z",
+                "note": "recap of yesterday's delay story"
+              },
+              {
+                "title": "State grant window opens for municipal fiber pilots",
+                "source": "Capitol Wire",
+                "url": "https://capitolwire.example/fiber/grant-window",
+                "published": "2026-07-16T05:40:00Z"
+              },
+              {
+                "title": "County said to be in talks to sell its dark fiber",
+                "source": "The Daily Splash",
+                "url": "https://dailysplash.example/fiber/dark-fiber-talks",
+                "published": "2026-07-16T04:15:00Z",
+                "note": "single source, unconfirmed"
+              },
+              {
+                "title": "What the pilot towns learned in year one",
+                "source": "Newsline",
+                "url": "https://newsline.example/broadband/pilot-lessons",
+                "published": "2026-07-16T02:00:00Z"
+              }
+            ]
+          }
+        },
+        {
+          "method": "GET",
+          "path": "/feeds/battery-recycling",
+          "body": {
+            "topic": "battery-recycling",
+            "fetched": "2026-07-16T06:30:00Z",
+            "items": [
+              {
+                "title": "Weekend depots triple battery drop-off volume",
+                "source": "Harbor Bulletin",
+                "url": "https://harborbulletin.example/recycling/depot-volume",
+                "published": "2026-07-16T05:05:00Z"
+              },
+              {
+                "title": "Recycler expands second-life pack line at the port",
+                "source": "Industry Ledger",
+                "url": "https://industryledger.example/recycling/second-life-line",
+                "published": "2026-07-16T03:30:00Z"
+              },
+              {
+                "title": "Fire-code update targets battery storage depots",
+                "source": "Harbor Bulletin",
+                "url": "https://harborbulletin.example/recycling/fire-code",
+                "published": "2026-07-15T23:45:00Z"
+              },
+              {
+                "title": "Study puts LFP recovery rates at 91 percent",
+                "source": "Materials Weekly",
+                "url": "https://materialsweekly.example/research/lfp-recovery",
+                "published": "2026-07-16T01:20:00Z"
+              }
+            ]
+          }
+        }
+      ]
+    },
+    {
+      "kind": "cli",
+      "id": "probe",
+      "description": "Provenance-trusted project script `mock-probe` that fetches every watched topic's current items from the fake feed service through http.authed.",
+      "shim": {
+        "path": "scripts/mock-probe.ts",
+        "content": "import { defineScript, gezel } from '@bendyline/gezel-sdk';\n\nexport const meta = defineScript({\n  name: 'mock-probe',\n  description:\n    'Probe the live fake feed service for this eval: fetches the current items for every topic in topics.json and returns the raw JSON bodies as evidence for the brief.',\n  inputs: {},\n  outputs: {\n    feeds: { type: 'string', description: 'JSON object mapping each topic id to the raw feed response body.' },\n  },\n  requires: ['workspace.read', 'network', 'credential:mock.feeds'],\n});\n\nconst services = JSON.parse(await gezel.fs.read('mocks/services.json'));\nconst base = (id) => {\n  const service = services.find((entry) => entry.id === id);\n  if (!service) throw new Error(`mock service \"${id}\" is not listed in mocks/services.json`);\n  return service.baseUrl;\n};\n\nconst topics = JSON.parse(await gezel.fs.read('topics.json'));\nconst out = {};\nfor (const topic of topics) {\n  const res = await gezel.http.authed(`${base('feeds')}/feeds/${topic.id}`, {\n    credential: 'mock.feeds',\n  });\n  if (!res.ok) throw new Error(`feed fetch failed for ${topic.id}: ${res.status} ${res.body}`);\n  out[topic.id] = res.body;\n}\n\ngezel.output({ feeds: JSON.stringify(out) });\n"
+      }
+    }
+  ],
+  "success": {
+    "summary": "briefs/2026-07-16.md is a five-minute brief grounded in the live fake feeds: a section per topic, sources cited per item, the dark-fiber rumor flagged as unconfirmed, and yesterday's council-vote story not re-reported as new.",
+    "deliverables": [
+      {
+        "path": "briefs/2026-07-16.md",
+        "kind": "markdown-report",
+        "minBytes": 700,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "briefs/2026-07-16.md",
+            "pattern": "(?:^|\\n)#{1,3}\\s+\\S",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "briefs/2026-07-16.md",
+            "pattern": "Broadband",
+            "flags": "i",
+            "label": "a section for the broadband topic"
+          },
+          {
+            "kind": "contains",
+            "file": "briefs/2026-07-16.md",
+            "pattern": "Battery|Recycling",
+            "flags": "i",
+            "label": "a section for the recycling topic"
+          },
+          {
+            "kind": "contains",
+            "file": "briefs/2026-07-16.md",
+            "pattern": "rumor|unconfirmed|single[- ]?source",
+            "flags": "i",
+            "label": "the single-sourced dark-fiber item is flagged, not reported as fact"
+          },
+          {
+            "kind": "contains",
+            "file": "briefs/2026-07-16.md",
+            "pattern": "grant window|fiber pilot",
+            "flags": "i",
+            "label": "the fresh grant-window story from the live feed is in the brief"
+          },
+          {
+            "kind": "contains",
+            "file": "briefs/2026-07-16.md",
+            "pattern": "depot|drop[- ]?off",
+            "flags": "i",
+            "label": "the fresh depot-volume story from the live feed is in the brief"
+          },
+          {
+            "kind": "citationsResolve",
+            "file": "briefs/2026-07-16.md",
+            "minCitations": 3,
+            "label": "items cite their sources"
+          },
+          {
+            "kind": "judge",
+            "file": "briefs/2026-07-16.md",
+            "rubric": "Yesterday's council-vote delay story is skipped or explicitly marked as already covered rather than re-reported as new, and the single-sourced dark-fiber item is flagged as rumor or unconfirmed, not presented as settled reporting.",
+            "sourceFiles": [
+              "briefs/2026-07-15.md"
+            ],
+            "label": "dedupe and skepticism"
+          }
+        ]
+      }
+    ],
+    "mocks": [
+      {
+        "service": "feeds",
+        "minRequests": 2,
+        "requiredPaths": [
+          "^/feeds/"
+        ]
+      }
+    ]
+  },
+  "rubric": {
+    "artifact": {
+      "path": "briefs/2026-07-16.md",
+      "kind": "markdown"
+    },
+    "axes": [
+      {
+        "name": "selection",
+        "description": "The kept items are the ones worth the owner's morning; duplicates of yesterday and noise are cut, with nothing important missing."
+      },
+      {
+        "name": "sourcing",
+        "description": "Every item names its source, linked where the feed gave one."
+      },
+      {
+        "name": "skepticism",
+        "description": "Single-sourced material is flagged as rumor or unconfirmed and kept visually separate from reporting."
+      },
+      {
+        "name": "brevity",
+        "description": "Five minutes of reading — tight sentences, no filler, sections that scan."
+      }
+    ]
+  },
+  "qualityFocus": [
+    "feed grounding",
+    "dedupe against yesterday",
+    "rumor flagged, never dressed as fact"
+  ]
+}

package/data/craftbook-templates/mo/morning-report/versions/1.0.2/craftbook.json

@@ -0,0 +1,164 @@
+{
+  "id": "morning-report",
+  "name": "Morning House Report",
+  "description": "The house says good morning properly: every zone's state read through the device tools, anything unusual surfaced first with one suggested action apiece, and the boring rest compressed to a line or two — glanceable before the first coffee. Locks the report criteria FIRST (every zone covered, anomalies before normals, a suggested action per oddity, short enough to read standing up), then polls the house, then writes the report — so the one morning the garage door stayed open all night is the morning it gets noticed. Use for morning house checks, daily home status, 'anything wrong with the house', overnight anomaly sweeps, and a glanceable device roundup.\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 report (huisautomaat) — coverage, anomalies-first ordering, and brevity locked as criteria → gated on artifact `{{workPath}}/scope.md` (markdown-notes)\n2. Poll the house (developer) — every device's state read and recorded, oddities flagged in the evidence → gated on artifact `{{workPath}}/states.md` (markdown-notes)\n3. Write the morning report (copywriter) — anomalies first with suggested actions, the calm rest in two lines → 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": [
+    "morning house report",
+    "how is the house",
+    "anything wrong with the house",
+    "what happened overnight",
+    "house status check"
+  ],
+  "steps": [
+    {
+      "id": "scope",
+      "name": "Scope the report",
+      "description": "coverage, anomalies-first ordering, and brevity locked as criteria",
+      "prompt": "Decide what a good morning report is before reading a single sensor. Step 1: pull the device inventory through the project's home capabilities and note the zones the house actually has — the report must account for every one of them, so nothing quietly falls off the page. Step 2: define what counts as unusual against what is typical for the hour — a door or window open overnight, a battery running low, a sensor that has gone quiet, anything switched on that should not be. Step 3: lock the shape — anything unusual comes FIRST, each oddity with the device, what is off, and one concrete suggested action; the nominal rest is compressed to a line or two, never a device-by-device inventory dump; the whole report reads standing up, before coffee. Step 4: write an acceptance-criteria checklist ('every zone accounted for', 'anomalies surfaced before anything normal', 'each anomaly carries one suggested action', 'every device mentioned exists in the states evidence — nothing invented', 'short and warm'). Call write_task_note with the zone 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": "huisautomaat",
+      "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": "poll"
+    },
+    {
+      "id": "poll",
+      "name": "Poll the house",
+      "description": "every device's state read and recorded, oddities flagged in the evidence",
+      "prompt": "Read the whole house, honestly. Step 1: through the project's home capabilities, read the inventory and the current state of every device — every zone, not just the interesting ones. Step 2: record the evidence in {{workPath}}/states.md — for each device: its name and id, its zone, the state you read, and the battery level or last-seen time where the house reports one — verbatim from the tool responses, never from memory. Step 3: as you go, flag which readings look unusual against the criteria locked in {{workPath}}/scope.md, and mark each flagged reading with a short why ('open since 23:41', 'battery 9 percent'). Step 4: end the notes with a one-line tally — devices read, anomalies flagged. Call write_task_note with that tally.\n\nBefore working, open the artifact input `{{workPath}}/scope.md` with `read_artifact`; do not look for it in the workspace.\n\nThe deliverable `{{workPath}}/states.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": "developer",
+      "consumes": [
+        {
+          "file": "{{workPath}}/scope.md",
+          "artifact": true
+        }
+      ],
+      "advanceWhen": {
+        "file": "{{workPath}}/states.md",
+        "minBytes": 1,
+        "sniff": "nonempty",
+        "artifact": true
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "{{workPath}}/states.md",
+            "bytes": 120,
+            "artifact": true
+          },
+          {
+            "kind": "sniff",
+            "file": "{{workPath}}/states.md",
+            "sniff": "nonempty",
+            "artifact": true
+          }
+        ],
+        "onReject": "poll",
+        "maxAttempts": 3
+      },
+      "next": "report"
+    },
+    {
+      "id": "report",
+      "name": "Write the morning report",
+      "description": "anomalies first with suggested actions, the calm rest in two lines",
+      "prompt": "Write the report the owner reads standing up. Step 1: open with a 'Needs attention' section — every anomaly from {{workPath}}/states.md, each as one short line: the device, what is off, and one concrete suggested action ('close it', 'charge it today', 'worth a look'). The most urgent first. If nothing is unusual, say so in one honest, pleasant line instead. Step 2: follow with the all-clear — the rest of the house compressed to a line or two by zone, never a device-by-device dump. Step 3: keep every device name and reading traceable to {{workPath}}/states.md — the report never mentions a device the evidence does not carry. Step 4: keep it warm and short — headings, plain words, no filler; if it does not read in under a minute, trim it. Write {{workPath}}/report.md and call write_task_note with the anomaly count and the top item.\n\nBefore working, open the artifact input `{{workPath}}/states.md` with `read_artifact`; do not look for it in the workspace.\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",
+      "consumes": [
+        {
+          "file": "{{workPath}}/states.md",
+          "artifact": true
+        }
+      ],
+      "advanceWhen": {
+        "file": "{{workPath}}/report.md",
+        "minBytes": 1,
+        "sniff": "nonempty",
+        "artifact": true
+      },
+      "gate": {
+        "at": "completion",
+        "checks": [
+          {
+            "kind": "minBytes",
+            "file": "{{workPath}}/report.md",
+            "bytes": 800,
+            "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}}/states.md, and {{workPath}}/scope.md and verify every criterion from the checklist: every zone in the states evidence is accounted for in the report; every anomaly flagged in the evidence appears in the 'Needs attention' section, before anything normal, each with one concrete suggested action; every device the report mentions exists in {{workPath}}/states.md with the state the report claims — nothing invented, nothing embellished; and the nominal remainder is compressed rather than dumped. Read it once at speed: if it does not scan before coffee, that is a FAIL on brevity. 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"
+}

package/data/craftbook-templates/mo/morning-report/versions/1.0.2/test.json

@@ -0,0 +1,253 @@
+{
+  "schemaVersion": 1,
+  "title": "Morning House Report smoke eval",
+  "objective": "Task-specific eval for the Morning House Report craftbook: the whole fake house is polled through the wired device tools, the garage door left open overnight and the low sensor battery surface first with suggested actions, and the nominal rest is compressed to a glanceable line or two.",
+  "tags": [
+    "external"
+  ],
+  "prompt": "morning - whats the house looking like today? give me the unusal stuff first",
+  "setup": {
+    "projectName": "Smart Home — Morning Report Eval",
+    "about": "Self-contained eval project for morning-report. The fake house's device tools are wired into the project's tool roster — the device inventory and the live state snapshot come from them, and they are the only source of truth for what the house is doing this morning. The polling evidence goes to tasks/eval/states.md and the morning report to tasks/eval/report.md.",
+    "missionObjectives": "Use the Morning House Report craftbook/template, read every zone's state through the wired device tools, surface anything unusual first with one suggested action each, compress the nominal rest, and write tasks/eval/report.md. No real smart-home services, hubs, or credentials — only the wired fake house.",
+    "files": [],
+    "worker": {
+      "name": "Roos",
+      "role": "Huisautomaat"
+    },
+    "craftbookParams": {
+      "workPath": "tasks/eval"
+    }
+  },
+  "mocks": [
+    {
+      "kind": "mcp",
+      "id": "home",
+      "description": "Fake smart-home MCP for this eval. Its tools surface directly on the tool roster: list the house's device inventory and read the current state snapshot of every device, battery levels included.",
+      "tools": [
+        {
+          "name": "list_devices",
+          "description": "List every device in the house with its id, name, zone, and type",
+          "resultTemplate": {
+            "devices": [
+              {
+                "id": "dev-01",
+                "name": "porch light",
+                "zone": "outside",
+                "type": "light"
+              },
+              {
+                "id": "dev-02",
+                "name": "front hall presence sensor",
+                "zone": "hall",
+                "type": "presence"
+              },
+              {
+                "id": "dev-03",
+                "name": "living room thermostat",
+                "zone": "living room",
+                "type": "thermostat"
+              },
+              {
+                "id": "dev-04",
+                "name": "garage door",
+                "zone": "garage",
+                "type": "cover"
+              },
+              {
+                "id": "dev-05",
+                "name": "kitchen ceiling light",
+                "zone": "kitchen",
+                "type": "light"
+              },
+              {
+                "id": "dev-06",
+                "name": "bedroom lamp",
+                "zone": "bedroom",
+                "type": "light"
+              },
+              {
+                "id": "dev-07",
+                "name": "front door lock",
+                "zone": "hall",
+                "type": "lock"
+              },
+              {
+                "id": "dev-08",
+                "name": "backyard motion sensor",
+                "zone": "outside",
+                "type": "motion"
+              }
+            ]
+          }
+        },
+        {
+          "name": "get_state",
+          "description": "Read the current state snapshot of every device in the house, with battery levels and last-change times where the device reports them",
+          "resultTemplate": {
+            "asOf": "2026-07-19T06:40:00Z",
+            "states": [
+              {
+                "id": "dev-01",
+                "name": "porch light",
+                "state": "off",
+                "reachable": true
+              },
+              {
+                "id": "dev-02",
+                "name": "front hall presence sensor",
+                "state": "home",
+                "reachable": true
+              },
+              {
+                "id": "dev-03",
+                "name": "living room thermostat",
+                "state": "heat 19.5C",
+                "reachable": true
+              },
+              {
+                "id": "dev-04",
+                "name": "garage door",
+                "state": "open",
+                "since": "2026-07-18T23:41:00Z",
+                "reachable": true
+              },
+              {
+                "id": "dev-05",
+                "name": "kitchen ceiling light",
+                "state": "off",
+                "reachable": true
+              },
+              {
+                "id": "dev-06",
+                "name": "bedroom lamp",
+                "state": "off",
+                "reachable": true
+              },
+              {
+                "id": "dev-07",
+                "name": "front door lock",
+                "state": "locked",
+                "reachable": true
+              },
+              {
+                "id": "dev-08",
+                "name": "backyard motion sensor",
+                "state": "clear",
+                "battery": 9,
+                "reachable": true
+              }
+            ]
+          }
+        }
+      ]
+    }
+  ],
+  "success": {
+    "summary": "tasks/eval/report.md leads with the garage door open since 23:41 and the backyard sensor at 9 percent battery — each with a suggested action — before any all-clear, the evidence in tasks/eval/states.md carries the real device ids from the tools (inventory and state snapshot actually read — request-log assertions), and the nominal rest is compressed rather than dumped.",
+    "deliverables": [
+      {
+        "path": "tasks/eval/states.md",
+        "kind": "markdown-notes",
+        "minBytes": 200,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "tasks/eval/states.md",
+            "pattern": "dev-\\d{2}",
+            "label": "the state evidence carries real device ids from the tools"
+          }
+        ]
+      },
+      {
+        "path": "tasks/eval/report.md",
+        "kind": "markdown-report",
+        "minBytes": 400,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "tasks/eval/report.md",
+            "pattern": "(?:^|\\n)#{1,3}\\s+\\S",
+            "flags": "i"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/report.md",
+            "pattern": "attention|unusual|anomal",
+            "flags": "i",
+            "label": "an anomalies-first section exists"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/report.md",
+            "pattern": "garage",
+            "flags": "i",
+            "label": "the garage door left open overnight is surfaced"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/report.md",
+            "pattern": "batter",
+            "flags": "i",
+            "label": "the low sensor battery is surfaced"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/report.md",
+            "pattern": "close|shut|charge|replace|check",
+            "flags": "i",
+            "label": "a suggested action is offered"
+          },
+          {
+            "kind": "judge",
+            "file": "tasks/eval/report.md",
+            "rubric": "The two anomalies (the garage door open overnight and the backyard sensor's low battery) appear before anything normal and each carries one concrete suggested action, every device mentioned matches the recorded state evidence, the nominal rest of the house is compressed to a line or two rather than a device-by-device dump, and the whole report reads comfortably in under a minute.",
+            "sourceFiles": [
+              "tasks/eval/states.md"
+            ],
+            "label": "anomalies-first and glanceable"
+          }
+        ]
+      }
+    ],
+    "mocks": [
+      {
+        "service": "home",
+        "minRequests": 2,
+        "requiredTools": [
+          "list_devices",
+          "get_state"
+        ]
+      }
+    ]
+  },
+  "rubric": {
+    "artifact": {
+      "path": "tasks/eval/report.md",
+      "kind": "markdown"
+    },
+    "axes": [
+      {
+        "name": "anomaly-detection",
+        "description": "The open garage door and the low battery are both caught from the state snapshot; nothing unusual is missed or invented."
+      },
+      {
+        "name": "prioritization",
+        "description": "Anomalies lead, most urgent first, each with one concrete suggested action; the all-clear never buries them."
+      },
+      {
+        "name": "brevity",
+        "description": "The nominal rest is compressed by zone to a line or two — glanceable before coffee, no inventory dump."
+      },
+      {
+        "name": "warmth",
+        "description": "It reads like the house saying good morning — plain, friendly words, no telemetry jargon."
+      }
+    ]
+  },
+  "qualityFocus": [
+    "anomalies surfaced first",
+    "grounded device states",
+    "glanceable brevity"
+  ]
+}