@rig-c/create 0.0.0-stage → 1.0.1

package/templates/agent-guide.md

@@ -0,0 +1,108 @@
+# Agent guide
+
+One painting in, one rig and its takes out. Start here.
+
+## Start here
+
+1. Read `project.json` first. Only its decisions count. Copy chat decisions
+   verbatim, with speaker and time.
+2. Run `rig-create next` to derive the next moves and whose they are.
+3. Resume per {{RESUME_LINK}}.
+
+Update stale pins with `rig-create packages upgrade --by <operator>`; see PROJECT.md.
+
+Declare the project's painting with `init --painting <file> --origin <words>`
+or `input add <path> --painting --origin <words>`. `project.json.painting` binds
+one input and survives restart; only it checks siblings for duplicate artwork.
+Older records may omit it.
+
+## The order, and what you may not do
+
+`interview` → `plan-approved` → `parts` → `parts-review` (the parts, by eye) →
+`rigging-interview` → `rigging` → `rig-review` (the rig, still frame) → `post-interview` → `post` →
+`direction-review` → `accepted`. Any stage can become `paused`.
+
+- Only a person's `signoffs[]` advances `stage` or passes a plan checkpoint.
+  You may draft, propose, build and measure in any stage; production rigging
+  waits for `rigging-interview` ({{STAGES_LINK}}).
+- Never derive answers from the image. Offer a pick per ask, with one reasoned
+  recommendation, Other and auto. Retain the shown set; never preselect.
+- Record "auto" as delegated, with your decision on the item. It is never a
+  blank for you to fill later.
+- {{INTERVIEW_CHECK}}
+
+## The user's four steps
+
+1. **Parts and meshes**. Ask framing/source. Review a labelled draft's parts and
+   mesh/chest structure. `parts-review` records its Review API response.
+   The latest receipt must approve the draft for `structure-snapshot --draft <id> --out <f>`
+   and `signoff parts-review --draft <id> --structure-file <f>`; record component triage with `diff-classes --diff-report <p> --by agent` ({{DRAFTS_LINK}}).
+2. **Rigging interview**. Ask fresh bones, deformation, physics and motion
+   against approved parts/structure; use interview-options/answer/decision.
+3. **Rig and animation**. Keep rig feedback.
+   Ingest: `review <version> --from-review <actId>`.
+4. **Direction**. `job open --rig <post>` asks the verdict only.
+   Ingest `review <post> --from-review <actId>`, then `job open --plan` for
+   record-page sign-off (source-linked Review).
+
+Ask: `job open --reviewer <id> --role owner --question <id>` or `--plan`;
+ingest `--from-review <actId>`, then advance.
+
+Use `advance <stage>`, `step-status <id>`; see `rig-create --help`.
+`advance parts-review` needs `--parts-check <report> --parts <assembled dir>`,
+refusing {{PARTS_CHECK_LINK}}.
+In `parts`, record an `intake-report` preflight and offer its choice, both advisory:
+{{INTAKE_LINK}}.
+
+Drafts stay under `derived/drafts/<id>/`, never versions/handoff. Old records
+warn: preserve history and dates; adopt choices explicitly.
+
+## Before you ask, plan or show
+
+- **Prior words.** In adopted projects, show them verbatim as a tailored option
+  with source/date; require a pick. Legacy derivation: {{DERIVED_LINK}}.
+- **Projection.** Ask and record it per {{PROJECTION_LINK}}.
+- **Upscale first** a painting smaller than the rig needs (SOURCE.md, step 2 below).
+- **The crux is a bar.** Before rigging, name every risk band, count background
+  or doubled pixels over every frame and both physics/breath extremes, and
+  report the maximum and its frame. Cover every band crossing another bone.
+- **Bars before sign-off.** Each reading names its command, checked against
+  installed `--help`. A reading with no command is a STOP before sign-off.
+  Loop bars: INTERVIEW.md, "Bars have reading commands".
+- **Appearance needs a person**, not a step bar: {{APPEARANCE_LINK}}.
+- **Nothing unnamed.** Add no motion, chain or physics the user did not name. A
+  reference ("move like X") is read for what it shows; anything more is a
+  question, not a default.
+- **Re-run first, cut last**, never a polygon in code or config: the ladder in
+  SOURCE.md, step 2 below.
+- **A full-screen scene is a CG build.** For `rig_type` `fullscreen.*`, each
+  rig version declares the gates in {{CG_GATES_LINK}}. Refusals are the art's to fix.
+- **The bust is meshed** unless the `meshes` guide excludes it with a reason
+  ({{BUST_LINK}}).
+- **Overlaps.** A piece over a part with its own animation (glasses over
+  eyelids) holds none of that part's pixels, or the animation shows its copy.
+- **Pairs.** Measure left/right symmetry against the painting before calling a
+  pair uneven ({{GATES_LINK}}).
+- **Look before the user does**, at native resolution and every boundary the
+  crux names: a whole-figure |composite − painting| map, every component listed
+  by box and part with a status (in the painting / fixed / left, why). Use the
+  true stack of part images, not a render flattened on white; crop on white and
+  on mid-grey, as white hides light fringes. Eye boxes a
+  reviewer chose are not the check: a seam outside them got through.
+
+## Where things go
+
+Each folder's lifetime and git status: {{LIFETIME_LINK}}.
+
+## What to read, and why
+
+{{READ_FIRST}}
+
+Read installed docs here at `package.json`'s pins to match tools, not memory.
+Run refusal `(see … guide …)` commands before changes.
+
+{{PACKAGE_DOCS}}
+
+Project organisation: rig-c/create docs (not installed).
+
+{{CREATE_DOCS}}

package/templates/collection-agent-guide.md

@@ -0,0 +1,72 @@
+# Agent guide
+
+This folder is a rig collection: several rigs that belong together — cut from
+one group painting, sharing outfits, a pose and a lineup. Read this page before
+you touch anything. What a person decided is in `collection.json` and in each
+member's `members/<name>/project.json`, and nowhere else.
+
+## Start here
+
+1. Read `collection.json` first. Its `interview` holds the answers given once
+   for every member; `members[]` lists the members; `versions[]` holds each
+   collection version with the member builds it pins.
+2. A member is an ordinary project under `members/<name>/`, with its own
+   `AGENTS.md`. Work on one member from inside its folder.
+3. The layout, the inherited answers and the pins are explained in
+   {{COLLECTIONS_LINK}}.
+
+## The order, and what you may not do
+
+`interview` → `plan-approved` → `reference` (one member goes first and sets the
+direction) → `members` (the rest, in parallel once their interviews are signed)
+→ `post-interview` → `post` → `direction-review` → `accepted`. Any stage can
+become `paused`.
+
+- Never move `stage` forward yourself. Only a person's `signoffs[]` entry does.
+- A member's answer that the collection gave is inherited, not copied. Never
+  copy a collection answer into a member; an override names what it overrides
+  and why, in the user's words.
+- Composition (lineup, overlaps, collection takes) belongs here, never in a
+  member.
+- A collection output is made from pinned member builds. Make it, then record
+  it: `rig-create collection pin`. Before showing one, run
+  `rig-create collection --check`: a member rebuilt since the pins is refused
+  by name, and its output is stale.
+- Derive before asking: take every answer the record already gives (earlier
+  answers, an intake note, a reviewer's past verdicts), recorded as
+  {{DERIVED_LINK}}, and ask only what is left, in one short message. This holds
+  for the collection interview and for each member's.
+- A group painting smaller than the members need is upscaled before it goes
+  into `inputs/`.
+- The rest of what to do before asking, planning or showing (the crux bar, no
+  unnamed motion, re-run the split first, overlaps, pairs, looking before the
+  user does) is in each member's `AGENTS.md` and applies to the member's work
+  unchanged.
+
+## Where things go
+
+| Path | Lifetime | In git |
+| --- | --- | --- |
+| `collection.json`, `shared/` | forever | yes |
+| `members/<name>/` | as in a project | as in a project |
+| `inputs/` | forever, never edited: the group painting, shared references | yes |
+| `evidence/v<N>/` | forever: what a collection verdict rested on | yes |
+| `derived/` | until regenerated | no |
+| `outputs/v<N>/` | until discarded; frozen when pinned | no |
+| `scratch/` | none; deleting it must lose nothing | no |
+
+## What to read, and why
+
+Read in this order, for the collection and in each member, each before the step
+it names:
+
+{{READ_FIRST}}
+
+These are installed here at the versions `package.json` pins, once for every
+member:
+
+{{PACKAGE_DOCS}}
+
+How a collection and its members are organised is documented in rig-c/create:
+
+{{CREATE_DOCS}}

package/templates/collection.json

@@ -0,0 +1,252 @@
+{
+  "schema": "rig-create/collection@0",
+  "name": null,
+  "title": null,
+  "stage": "interview",
+  "stage_values": [
+    "interview",
+    "plan-approved",
+    "reference",
+    "members",
+    "post-interview",
+    "post",
+    "direction-review",
+    "accepted",
+    "paused"
+  ],
+  "signoffs": [],
+  "interview": {
+    "question_set": "rig-create/interview@2 (waypoints read from @rig-c/parts 2.37.0)",
+    "held": false,
+    "date": null,
+    "interviewer": null,
+    "respondent": null,
+    "questions": [
+      {
+        "id": "members",
+        "scope": "collection",
+        "text": "Who is in it, and where does each member come from (its own painting, or a crop of a shared one)?",
+        "source": "rig-create",
+        "answer": null,
+        "history": []
+      },
+      {
+        "id": "reference_member",
+        "scope": "collection",
+        "text": "Which member goes first and sets the direction for the rest?",
+        "source": "rig-create",
+        "answer": null,
+        "history": []
+      },
+      {
+        "id": "style",
+        "scope": "collection",
+        "text": "What must look and move the same on every member (rig style, physics character, line and colour handling)?",
+        "source": "rig-create",
+        "answer": null,
+        "history": []
+      },
+      {
+        "id": "outfits",
+        "scope": "collection",
+        "text": "Which outfits (or other variants) does every member wear; what changes with the outfit and what stays the same?",
+        "source": "rig-create",
+        "answer": null,
+        "history": []
+      },
+      {
+        "id": "composition",
+        "scope": "collection",
+        "text": "How are the members put together: lineup, who overlaps whom, and which collection outputs (lineup take, outfit-switch take)?",
+        "source": "rig-create",
+        "answer": null,
+        "history": []
+      }
+    ],
+    "framing": [
+      {
+        "id": "rig_type",
+        "text": "What is this: a character rig (standing idle / complex motion / a monster with its own moves) or a full-screen scene (minimal motion: breath, blink / motion tied to game logic)?",
+        "source": "rig-create",
+        "required_for": ["all"],
+        "answer": null,
+        "history": []
+      },
+      {
+        "id": "reference",
+        "text": "'Make it feel like this': a video, an existing demo, a game you have in mind?",
+        "source": "rig-create",
+        "required_for": ["all"],
+        "answer": null,
+        "history": []
+      }
+    ],
+    "waypoints": [
+      {
+        "id": "seethrough",
+        "config": ["seethrough"],
+        "prompts": [
+          {
+            "id": "x_layers",
+            "text": "Is there anything the layer split must keep together or keep apart: a hat with the head, a held object on its own?",
+            "source": "rig-create",
+            "required_for": ["all"]
+          }
+        ],
+        "guide": null,
+        "history": []
+      },
+      {
+        "id": "assemble",
+        "config": ["assemble", "limbs", "scene", "waist"],
+        "prompts": [
+          {
+            "id": "split_granularity",
+            "text": "How finely should it be cut?",
+            "source": "rig-create",
+            "required_for": ["all"]
+          },
+          {
+            "id": "split_emphasis",
+            "text": "Which parts matter most: more detail, or kept exactly as painted?",
+            "source": "rig-create",
+            "required_for": ["all"]
+          }
+        ],
+        "guide": null,
+        "history": []
+      },
+      {
+        "id": "bones",
+        "config": ["bones"],
+        "prompts": [
+          {
+            "id": "x_chains",
+            "text": "What moves on its own: hair strands, coat tails, sleeves, held things?",
+            "source": "rig-create",
+            "required_for": ["all"]
+          }
+        ],
+        "guide": null,
+        "history": []
+      },
+      {
+        "id": "meshes",
+        "config": ["meshes", "regions"],
+        "prompts": [
+          {
+            "id": "x_deform",
+            "text": "What must bend smoothly as one piece, and what must stay rigid?",
+            "source": "rig-create",
+            "required_for": ["all"]
+          }
+        ],
+        "guide": null,
+        "history": []
+      },
+      {
+        "id": "constraints",
+        "config": ["constraints", "stops", "cloth"],
+        "prompts": [
+          {
+            "id": "split_effects",
+            "text": "Which parts have special effects (sway on physics, move in wind, jiggle, glow, flicker), and how far do they move?",
+            "source": "rig-create",
+            "required_for": ["all"]
+          }
+        ],
+        "guide": null,
+        "history": []
+      },
+      {
+        "id": "motion",
+        "config": ["motion"],
+        "prompts": [
+          {
+            "id": "motions",
+            "text": "Which moves? Name each one.",
+            "source": "rig-create",
+            "required_for": [
+              "character.idle",
+              "character.complex",
+              "character.monster",
+              "fullscreen.minimal"
+            ]
+          },
+          {
+            "id": "game_states",
+            "text": "Which states, and what moves it from one to the next?",
+            "source": "rig-create",
+            "required_for": ["fullscreen.game_logic"]
+          }
+        ],
+        "guide": null,
+        "history": []
+      }
+    ],
+    "post": {
+      "held": false,
+      "date": null,
+      "items": [
+        {
+          "id": "camera_depth",
+          "config": ["turn", "depth"],
+          "text": "Does the view move (a 2.5D head-turn with parallax, a push-in) or is it fixed?",
+          "source": "rig-create",
+          "guide": null,
+          "history": []
+        },
+        {
+          "id": "post_timing",
+          "config": [],
+          "text": "How long, does it loop, and where are the beats?",
+          "source": "rig-create",
+          "guide": null,
+          "history": []
+        },
+        {
+          "id": "post_composition",
+          "config": [],
+          "text": "What is around the figures: background, framing, light?",
+          "source": "rig-create",
+          "guide": null,
+          "history": []
+        }
+      ]
+    }
+  },
+  "plan": {
+    "drafted_by": null,
+    "drafted_at": null,
+    "order": [
+      {
+        "phase": 1,
+        "members": [],
+        "runs": "alone",
+        "rests_on": ["reference_member"],
+        "ends_with": "the reference member's rig is accepted at its rig-review and the user signs { \"stage\": \"reference\" }"
+      },
+      {
+        "phase": 2,
+        "members": [],
+        "runs": "parallel",
+        "rests_on": ["members"],
+        "ends_with": "every member's rig accepted at rig-review and every cross bar met"
+      }
+    ],
+    "cross_bars": [],
+    "steps": []
+  },
+  "members": [],
+  "shared": "shared/conventions.json",
+  "composition": {
+    "lineup": null,
+    "layers": [],
+    "outputs_wanted": []
+  },
+  "inputs": [],
+  "derived": [],
+  "versions": [],
+  "archive": null,
+  "notes": []
+}

package/templates/example-chain/AGENTS.md

@@ -0,0 +1,134 @@
+# Agent guide
+
+One painting in, one rig and its takes out. Start here.
+
+## Start here
+
+1. Read `project.json` first. Only its decisions count. Copy chat decisions
+   verbatim, with speaker and time.
+2. Run `rig-create next` to derive the next moves and whose they are.
+3. Resume per [PROJECT.md § How an agent resumes a project](https://github.com/rig-c/create/blob/main/docs/PROJECT.md#how-an-agent-resumes-a-project).
+
+Update stale pins with `rig-create packages upgrade --by <operator>`; see PROJECT.md.
+
+Declare the project's painting with `init --painting <file> --origin <words>`
+or `input add <path> --painting --origin <words>`. `project.json.painting` binds
+one input and survives restart; only it checks siblings for duplicate artwork.
+Older records may omit it.
+
+## The order, and what you may not do
+
+`interview` → `plan-approved` → `parts` → `parts-review` (the parts, by eye) →
+`rigging-interview` → `rigging` → `rig-review` (the rig, still frame) → `post-interview` → `post` →
+`direction-review` → `accepted`. Any stage can become `paused`.
+
+- Only a person's `signoffs[]` advances `stage` or passes a plan checkpoint.
+  You may draft, propose, build and measure in any stage; production rigging
+  waits for `rigging-interview` ([PROJECT.md § Stages](https://github.com/rig-c/create/blob/main/docs/PROJECT.md#stages)).
+- Never derive answers from the image. Offer a pick per ask, with one reasoned
+  recommendation, Other and auto. Retain the shown set; never preselect.
+- Record "auto" as delegated, with your decision on the item. It is never a
+  blank for you to fill later.
+- Before asking the user to sign the interview, run `rig-create interview --check`: it names every `null` answer.
+
+## The user's four steps
+
+1. **Parts and meshes**. Ask framing/source. Review a labelled draft's parts and
+   mesh/chest structure. `parts-review` records its Review API response.
+   The latest receipt must approve the draft for `structure-snapshot --draft <id> --out <f>`
+   and `signoff parts-review --draft <id> --structure-file <f>`; record component triage with `diff-classes --diff-report <p> --by agent` ([PROJECT.md § Labelled draft rigs](https://github.com/rig-c/create/blob/main/docs/PROJECT.md#labelled-draft-rigs-and-structure-decisions)).
+2. **Rigging interview**. Ask fresh bones, deformation, physics and motion
+   against approved parts/structure; use interview-options/answer/decision.
+3. **Rig and animation**. Keep rig feedback.
+   Ingest: `review <version> --from-review <actId>`.
+4. **Direction**. `job open --rig <post>` asks the verdict only.
+   Ingest `review <post> --from-review <actId>`, then `job open --plan` for
+   record-page sign-off (source-linked Review).
+
+Ask: `job open --reviewer <id> --role owner --question <id>` or `--plan`;
+ingest `--from-review <actId>`, then advance.
+
+Use `advance <stage>`, `step-status <id>`; see `rig-create --help`.
+`advance parts-review` needs `--parts-check <report> --parts <assembled dir>`,
+refusing [`PARTS_CHECK_MISSING`, `PARTS_CHECK_UNREAD`, `PARTS_CHECK_INVALID`, `PARTS_CHECK_CHANGED`, `PARTS_CHECK_FAILED`, `PARTS_CHECK_STALE`](https://github.com/rig-c/create/blob/main/docs/CHECKS.md#parts-stage-check-at-the-parts--parts-review-move).
+In `parts`, record an `intake-report` preflight and offer its choice, both advisory:
+[PROJECT.md § Advisory intake reports](https://github.com/rig-c/create/blob/main/docs/PROJECT.md#advisory-intake-reports).
+
+Drafts stay under `derived/drafts/<id>/`, never versions/handoff. Old records
+warn: preserve history and dates; adopt choices explicitly.
+
+## Before you ask, plan or show
+
+- **Prior words.** In adopted projects, show them verbatim as a tailored option
+  with source/date; require a pick. Legacy derivation: [INTERVIEW.md § Derived answers](https://github.com/rig-c/create/blob/main/docs/INTERVIEW.md#derived-answers).
+- **Projection.** Ask and record it per [CHECKS.md § Projection and ground](https://github.com/rig-c/create/blob/main/docs/CHECKS.md#projection-and-ground).
+- **Upscale first** a painting smaller than the rig needs (SOURCE.md, step 2 below).
+- **The crux is a bar.** Before rigging, name every risk band, count background
+  or doubled pixels over every frame and both physics/breath extremes, and
+  report the maximum and its frame. Cover every band crossing another bone.
+- **Bars before sign-off.** Each reading names its command, checked against
+  installed `--help`. A reading with no command is a STOP before sign-off.
+  Loop bars: INTERVIEW.md, "Bars have reading commands".
+- **Appearance needs a person**, not a step bar: [CHECKS.md § Appearance bar wording](https://github.com/rig-c/create/blob/main/docs/CHECKS.md#appearance-bar-wording).
+- **Nothing unnamed.** Add no motion, chain or physics the user did not name. A
+  reference ("move like X") is read for what it shows; anything more is a
+  question, not a default.
+- **Re-run first, cut last**, never a polygon in code or config: the ladder in
+  SOURCE.md, step 2 below.
+- **A full-screen scene is a CG build.** For `rig_type` `fullscreen.*`, each
+  rig version declares the gates in [CHECKS.md § CG gates](https://github.com/rig-c/create/blob/main/docs/CHECKS.md#cg-gates). Refusals are the art's to fix.
+- **The bust is meshed** unless the `meshes` guide excludes it with a reason
+  ([CHECKS.md § The bust at meshes](https://github.com/rig-c/create/blob/main/docs/CHECKS.md#the-bust-at-meshes)).
+- **Overlaps.** A piece over a part with its own animation (glasses over
+  eyelids) holds none of that part's pixels, or the animation shows its copy.
+- **Pairs.** Measure left/right symmetry against the painting before calling a
+  pair uneven ([CRAFT.md G8](https://github.com/rig-c/create/blob/main/docs/CRAFT.md#for-the-gates-not-the-interview)).
+- **Look before the user does**, at native resolution and every boundary the
+  crux names: a whole-figure |composite − painting| map, every component listed
+  by box and part with a status (in the painting / fixed / left, why). Use the
+  true stack of part images, not a render flattened on white; crop on white and
+  on mid-grey, as white hides light fringes. Eye boxes a
+  reviewer chose are not the check: a seam outside them got through.
+
+## Where things go
+
+Each folder's lifetime and git status: [PROJECT.md § Data is split by lifetime](https://github.com/rig-c/create/blob/main/docs/PROJECT.md#data-is-split-by-lifetime).
+
+## What to read, and why
+
+1. **Before the interview**: INTERVIEW.md, and CRAFT.md for the default at each
+   waypoint (both linked below); parts AUTHORING §3, *`config.json`, field by
+   field*; [CHECKS.md](https://github.com/rig-c/create/blob/main/docs/CHECKS.md) before a sign-off.
+2. **Before `seethrough` and `assemble`**: SOURCE.md, the order of the split and
+   the cut: [re-running a bad split](https://github.com/rig-c/create/blob/main/docs/SOURCE.md#re-run-first),
+   [where a cut is allowed](https://github.com/rig-c/create/blob/main/docs/SOURCE.md#where-a-cut-is-allowed);
+   then parts AUTHORING *See-through head-run source reading*, *Source before
+   scissors* and *Limbs: cut at their joints*.
+3. **Before rigging**: [CRAFT.md § For the gates](https://github.com/rig-c/create/blob/main/docs/CRAFT.md#for-the-gates-not-the-interview)
+   (G1–G9); core RIGGING and MOTION; parts AUTHORING §7, the bars `check` enforces.
+4. **Before showing the user**: [LESSONS.md § The self-check](https://github.com/rig-c/create/blob/main/docs/LESSONS.md#the-self-check),
+   the [per-frame seam count](https://github.com/rig-c/create/blob/main/docs/LESSONS.md#count-the-seam-bar-over-every-frame),
+   [what to show first](https://github.com/rig-c/create/blob/main/docs/LESSONS.md#what-to-show-the-user-first).
+
+Read installed docs here at `package.json`'s pins to match tools, not memory.
+Run refusal `(see … guide …)` commands before changes.
+
+- [@rig-c/parts README.md](node_modules/@rig-c/parts/README.md): painting-to-rig pipeline and command order.
+- [@rig-c/parts docs/AUTHORING.md](node_modules/@rig-c/parts/docs/AUTHORING.md): config fields, interview waypoints and refusal remedies.
+- [@rig-c/parts docs/PAINTING.md](node_modules/@rig-c/parts/docs/PAINTING.md): what makes a painting riggable, and the check to run on it before See-through.
+- [@rig-c/parts docs/DEPTH.md](node_modules/@rig-c/parts/docs/DEPTH.md): when the plan has a head turn driven by depth.
+- [@rig-c/parts docs/STABILITY.md](node_modules/@rig-c/parts/docs/STABILITY.md): stable commands, fields and outputs.
+- [@rig-c/parts docs/MIGRATION.md](node_modules/@rig-c/parts/docs/MIGRATION.md): only when the pinned version is bumped.
+- [@rig-c/core README.md](node_modules/@rig-c/core/README.md): rigc scope, commands and documentation.
+- [@rig-c/core docs/AUTHORING.md](node_modules/@rig-c/core/docs/AUTHORING.md): rig/motion formats, build loop and failure remedies.
+- [@rig-c/core docs/RIGGING.md](node_modules/@rig-c/core/docs/RIGGING.md): bones and pivots at the bones waypoint.
+- [@rig-c/core docs/MOTION.md](node_modules/@rig-c/core/docs/MOTION.md): key-pose motions at the motion waypoint.
+- [@rig-c/core docs/FACE.md](node_modules/@rig-c/core/docs/FACE.md): blink, gaze and head-turn plans.
+- [@rig-c/core docs/INGEST.md](node_modules/@rig-c/core/docs/INGEST.md): only when you were handed a skeleton this project did not author.
+
+Project organisation: rig-c/create docs (not installed).
+
+- [docs/PROJECT.md](https://github.com/rig-c/create/blob/main/docs/PROJECT.md): the folder, stages and sign-offs, the plan, versions, and `project.json` field by field.
+- [docs/INTERVIEW.md](https://github.com/rig-c/create/blob/main/docs/INTERVIEW.md): how to ask, and the shape every answer, delegation and piece of feedback takes.
+- [docs/CRAFT.md](https://github.com/rig-c/create/blob/main/docs/CRAFT.md): the default to propose at each waypoint.
+- [docs/COLLECTIONS.md](https://github.com/rig-c/create/blob/main/docs/COLLECTIONS.md): when this project is a member of a collection.

package/templates/example-chain/CLAUDE.md

@@ -0,0 +1,3 @@
+@AGENTS.md
+
+The agent guide for this project is AGENTS.md, imported above. It is the only copy: change it there, not here.

package/templates/example-chain/README.md

@@ -0,0 +1,13 @@
+# Synthetic bound chain
+
+The geometric painting and parts come from generateExampleParts. The labelled draft and production model come from generateExampleMeshDraft on those exact parts, with a producer-made torso mesh.
+
+The setup still is the assembly recomposite, retained as synthetic setup evidence. The example demonstrates record binding. Rendered motion and camera behaviour require separate producer validation.
+
+The public create CLI writes stage moves, signoffs, rig plan steps (plan add), parts step progress (step-status), rig/post review verdicts, the review package, immutable input inventory, the sealed scene and authored direction. The direction is direction@2, bound to that scene. The parts verdict is a synthetic review-tool export from tests/parts-package.ts, recorded as a producer step.
+
+Two seed kinds cover plan.steps (setup and status progress) with split.parts, and the rig version declaration. Source inventory is registered by input add. Each seed declares its consuming command and records that command’s step index in chain.json.
+
+checkpoints/ contains frozen prefix projects and adjacent pending-act declarations. Checkpoint 0a-interview-options follows interview-options rig_type and precedes the interview answers, the plan add calls and signoff interview; it has no skeleton.model.json and four expected INTERVIEW_INCOMPLETE refusal codes (interview.framing[rig_type].answer, interview.framing[reference].answer, interview.waypoints[seethrough].guide, interview.waypoints[assemble].guide) with zero warnings. Checkpoint 0b-plan-pending follows advance plan-approved with the plan add steps drafted and precedes signoff plan-approved; it has no skeleton.model.json, zero refusals and zero warnings. Stage 1 (1-parts-verdict, zero refusals, zero warnings) follows the producer parts-stage check report that the parts-review move binds, and precedes the synthetic review-tool parts verdict export and the structure-snapshot writer, which copies the meshes declarations of the labelled draft once that verdict is approved. Stage 2 (2-rigging-answers) precedes the first rigging answer and still needs later option sets to be agent-drafted in declared steps before registration by public writers. Pending answers use the SHA-256 of the file drafted during the walk. Stage 3 (3-rig-review, zero refusals, zero warnings) precedes rig review. Stage 4 (4-direction-post, zero refusals, zero warnings) precedes the background plate registration, the scene write, the declared agent-authored direction@2 input and the direction write. Pending stage moves use advance with the existing authorizing signoffs. Every file authored or copied for a later writer has a declared producer step with output hashes. The chain_step indices are zero-based. Stage 2 has four expected INTERVIEW_INCOMPLETE findings for its unanswered rigging guides (interview.waypoints[bones].guide, interview.waypoints[meshes].guide, interview.waypoints[constraints].guide, interview.waypoints[motion].guide) and zero warnings. Every prefix has zero warnings and records its stage moves in advances[]. tests/example-chain.test.ts checks these states.
+
+Reproduce with `bun scripts/generate-example-chain.ts <fresh-dir> --check` from the create repository.