@fenglimg/fabric-cli 1.8.0-rc.2 → 2.0.0-rc.1

package/templates/codex-skills/fabric-init/SKILL.md

@@ -1,162 +0,0 @@
----
-name: fabric-init
-description: Use this skill when .fabric/forensic.json exists and this repository still needs the remaining Fabric initialization steps.
----
-
-# fabric-init — Canonical Skill Source
-
-> This file is the single source of truth for the fabric-init skill.
-> Do NOT edit the per-client SKILL.md files directly.
-> Run `packages/cli/scripts/derive-skills.ts` to regenerate them from this source.
-
-## Precondition
-
-MUST: Read `.fabric/forensic.json` before taking any other action. If the file does not
-exist, stop immediately and tell the user: run `fab init` first to generate the evidence
-package.
-
-MUST: Check `.fabric/init-context.json`. If it already exists, stop and report that this
-repository appears to have completed initialization already.
-
-MUST: Treat `.fabric/bootstrap/README.md` as the authoritative initialization guide for
-the current repository.
-
-MUST: Use `.fabric/forensic.json` and repository structure as evidence when deciding what
-to do next.
-
-MUST: Preserve protected tokens exactly as written — see the Protected Tokens section.
-
-NEVER: Claim initialization is complete without having checked `.fabric/init-context.json`.
-
-NEVER: Rewrite or translate protected tokens.
-
-NEVER: Ignore `.fabric/bootstrap/README.md` when determining the next initialization step.
-
-Treat the following state as initialization pending:
-
-- `.fabric/forensic.json` exists
-- `.fabric/init-context.json` does not exist
-
-## Execution Flow — 3 Phases / 3 Rounds
-
-### Phase 1 — Framework Confirmation (1 round, efficient)
-
-Display a summary of `framework`, `topology.by_ext`, and `entry_points` from
-`.fabric/forensic.json`. Ask the user 1–2 clarifying questions about the framework
-architecture.
-
-Example (Cocos Creator 3.x):
-
-> I detected a Cocos Creator 3.8 project. Main scripts are in `assets/scripts` using the
-> `@ccclass + extends Component` pattern. Please confirm: (1) Is this a TypeScript project
-> (not JavaScript)? (2) Are node references injected mainly via `@property(Node)`, or via
-> `find/getChildByName`?
-
-Store the user's answers as verified framework assumptions before proceeding to Phase 2.
-
-### Phase 2 — Invariant Extraction (1 round, critical)
-
-Based on the `recommendations_for_skill` list in `.fabric/forensic.json`, ask the user
-3–5 invariant questions covering three categories:
-
-- `ban`: things that must never appear — e.g. `any`, `async` in `update()`, find-by-name
-- `require`: things that must always be present — e.g. strict TypeScript, `@ccclass`
-  decorator, imports only from `cc`
-- `protect`: directories or files that AI must not modify — typically
-  `assets/prefabs/**`, `assets/scenes/**`, `**/*.meta`
-
-Principles:
-
-- Ask only about invariants, not about preferences.
-- Each question accepts only yes / no / a concrete rule — never accept vague answers.
-- Do not auto-infer hard constraints the user has not confirmed.
-
-### Phase 3 — Construction and Landing (1 round, automated)
-
-#### 3.1 Write `.fabric/init-context.json`
-
-Fields required:
-
-- `framework`
-- `architecture_patterns`
-- `invariants`
-- `domain_groups`
-- `interview_trail`
-- `forensic_ref`
-
-Writing rules:
-
-- `invariants[].type` MUST be one of `ban`, `require`, `protect`.
-- `domain_groups` is inferred from `entry_points` and interview results.
-- `interview_trail[]` MUST record the raw Q&A from Phase 1 and Phase 2.
-- `forensic_ref` MUST be `.fabric/forensic.json`.
-
-#### 3.2 Generate layered `AGENTS.md`
-
-Root `AGENTS.md` requirements:
-
-- MUST be within 300 lines.
-- Structure:
-  - `# {projectName} — L0 AGENTS.md`
-  - `<!-- fab:index -->`: populated with the `domain_groups` index
-  - `## L0 AI Constraints`: derived from invariants, grouped by `ban`, `require`, `protect`
-  - `## @HUMAN`: protect paths and any human-declared protection rules
-  - `## L1 Candidate Notes`: candidate sub-module descriptions for each domain group
-
-If `domain_groups.length >= 2`, generate a `{group_path}/AGENTS.md` for each group.
-Maximum depth is L3; total nesting MUST NOT exceed 4 levels.
-
-#### 3.3 Update `.fabric/agents.meta.json`
-
-- The `nodes` tree MUST match the generated AGENTS hierarchy.
-- Update the hash of every AGENTS.md file that was written.
-- Maintain a consistent internal revision hash chain.
-
-#### 3.4 Final output
-
-List all generated files for the user and recommend running `fabric doctor --fix` for
-ongoing maintenance.
-
-## Hard Rules
-
-- Zero TODO: never generate `TODO`, `TBD`, placeholders, or stubs in output files.
-- No YAML frontmatter in outputs: generated `AGENTS.md` files MUST NOT contain YAML
-  frontmatter.
-- Root `AGENTS.md` MUST be <= 300 lines.
-- Total AGENTS nesting MUST be <= 4 levels.
-- Do not auto-infer invariants the user has not confirmed.
-- When content is uncertain, omit it — do not leave placeholders.
-
-## Output Contract
-
-On successful completion the following files exist or are updated:
-
-| File | Action |
-|------|--------|
-| `.fabric/init-context.json` | Created with all required fields |
-| `AGENTS.md` | Created (root L0) |
-| `{group_path}/AGENTS.md` | Created for each domain group (when applicable) |
-| `.fabric/agents.meta.json` | Updated nodes tree + hashes |
-
-On failure or early termination the skill MUST leave no partial files. If a write fails
-mid-sequence, report the failure and the exact file that was not written.
-
-## Protected Tokens
-
-The following tokens MUST be preserved exactly as shown — same casing, same punctuation,
-never translated:
-
-| Token | Type |
-|-------|------|
-| `AGENTS.md` | Filename |
-| `FABRIC.md` | Filename |
-| `.fabric/agents.meta.json` | Path |
-| `.fabric/init-context.json` | Path |
-| `.fabric/forensic.json` | Path |
-| `.fabric/bootstrap/README.md` | Path |
-| `MUST` | Keyword |
-| `NEVER` | Keyword |
-| `fab init` | CLI command |
-| `fabric doctor --fix` | CLI command |
-| `<!-- fab:index -->` | HTML comment marker |
-| `@HUMAN` | Section marker |

package/templates/husky/pre-commit

@@ -1,9 +0,0 @@
-#!/bin/sh
-# Fabric pre-commit hook: block manual edits to generated metadata.
-
-if git diff --cached --name-only | grep -q '^\.fabric/agents\.meta\.json$'; then
-  if [ "$FAB_ALLOW_META_EDIT" != '1' ]; then
-    echo '.fabric/agents.meta.json cannot be manually edited; update .fabric/rules and run fabric doctor --fix, or set FAB_ALLOW_META_EDIT=1' >&2
-    exit 1
-  fi
-fi

package/templates/skill-source/fabric-init/SOURCE.md

@@ -1,157 +0,0 @@
-# fabric-init — Canonical Skill Source
-
-> This file is the single source of truth for the fabric-init skill.
-> Do NOT edit the per-client SKILL.md files directly.
-> Run `packages/cli/scripts/derive-skills.ts` to regenerate them from this source.
-
-## Precondition
-
-MUST: Read `.fabric/forensic.json` before taking any other action. If the file does not
-exist, stop immediately and tell the user: run `fab init` first to generate the evidence
-package.
-
-MUST: Check `.fabric/init-context.json`. If it already exists, stop and report that this
-repository appears to have completed initialization already.
-
-MUST: Treat `.fabric/bootstrap/README.md` as the authoritative initialization guide for
-the current repository.
-
-MUST: Use `.fabric/forensic.json` and repository structure as evidence when deciding what
-to do next.
-
-MUST: Preserve protected tokens exactly as written — see the Protected Tokens section.
-
-NEVER: Claim initialization is complete without having checked `.fabric/init-context.json`.
-
-NEVER: Rewrite or translate protected tokens.
-
-NEVER: Ignore `.fabric/bootstrap/README.md` when determining the next initialization step.
-
-Treat the following state as initialization pending:
-
-- `.fabric/forensic.json` exists
-- `.fabric/init-context.json` does not exist
-
-## Execution Flow — 3 Phases / 3 Rounds
-
-### Phase 1 — Framework Confirmation (1 round, efficient)
-
-Display a summary of `framework`, `topology.by_ext`, and `entry_points` from
-`.fabric/forensic.json`. Ask the user 1–2 clarifying questions about the framework
-architecture.
-
-Example (Cocos Creator 3.x):
-
-> I detected a Cocos Creator 3.8 project. Main scripts are in `assets/scripts` using the
-> `@ccclass + extends Component` pattern. Please confirm: (1) Is this a TypeScript project
-> (not JavaScript)? (2) Are node references injected mainly via `@property(Node)`, or via
-> `find/getChildByName`?
-
-Store the user's answers as verified framework assumptions before proceeding to Phase 2.
-
-### Phase 2 — Invariant Extraction (1 round, critical)
-
-Based on the `recommendations_for_skill` list in `.fabric/forensic.json`, ask the user
-3–5 invariant questions covering three categories:
-
-- `ban`: things that must never appear — e.g. `any`, `async` in `update()`, find-by-name
-- `require`: things that must always be present — e.g. strict TypeScript, `@ccclass`
-  decorator, imports only from `cc`
-- `protect`: directories or files that AI must not modify — typically
-  `assets/prefabs/**`, `assets/scenes/**`, `**/*.meta`
-
-Principles:
-
-- Ask only about invariants, not about preferences.
-- Each question accepts only yes / no / a concrete rule — never accept vague answers.
-- Do not auto-infer hard constraints the user has not confirmed.
-
-### Phase 3 — Construction and Landing (1 round, automated)
-
-#### 3.1 Write `.fabric/init-context.json`
-
-Fields required:
-
-- `framework`
-- `architecture_patterns`
-- `invariants`
-- `domain_groups`
-- `interview_trail`
-- `forensic_ref`
-
-Writing rules:
-
-- `invariants[].type` MUST be one of `ban`, `require`, `protect`.
-- `domain_groups` is inferred from `entry_points` and interview results.
-- `interview_trail[]` MUST record the raw Q&A from Phase 1 and Phase 2.
-- `forensic_ref` MUST be `.fabric/forensic.json`.
-
-#### 3.2 Generate layered `AGENTS.md`
-
-Root `AGENTS.md` requirements:
-
-- MUST be within 300 lines.
-- Structure:
-  - `# {projectName} — L0 AGENTS.md`
-  - `<!-- fab:index -->`: populated with the `domain_groups` index
-  - `## L0 AI Constraints`: derived from invariants, grouped by `ban`, `require`, `protect`
-  - `## @HUMAN`: protect paths and any human-declared protection rules
-  - `## L1 Candidate Notes`: candidate sub-module descriptions for each domain group
-
-If `domain_groups.length >= 2`, generate a `{group_path}/AGENTS.md` for each group.
-Maximum depth is L3; total nesting MUST NOT exceed 4 levels.
-
-#### 3.3 Update `.fabric/agents.meta.json`
-
-- The `nodes` tree MUST match the generated AGENTS hierarchy.
-- Update the hash of every AGENTS.md file that was written.
-- Maintain a consistent internal revision hash chain.
-
-#### 3.4 Final output
-
-List all generated files for the user and recommend running `fabric doctor --fix` for
-ongoing maintenance.
-
-## Hard Rules
-
-- Zero TODO: never generate `TODO`, `TBD`, placeholders, or stubs in output files.
-- No YAML frontmatter in outputs: generated `AGENTS.md` files MUST NOT contain YAML
-  frontmatter.
-- Root `AGENTS.md` MUST be <= 300 lines.
-- Total AGENTS nesting MUST be <= 4 levels.
-- Do not auto-infer invariants the user has not confirmed.
-- When content is uncertain, omit it — do not leave placeholders.
-
-## Output Contract
-
-On successful completion the following files exist or are updated:
-
-| File | Action |
-|------|--------|
-| `.fabric/init-context.json` | Created with all required fields |
-| `AGENTS.md` | Created (root L0) |
-| `{group_path}/AGENTS.md` | Created for each domain group (when applicable) |
-| `.fabric/agents.meta.json` | Updated nodes tree + hashes |
-
-On failure or early termination the skill MUST leave no partial files. If a write fails
-mid-sequence, report the failure and the exact file that was not written.
-
-## Protected Tokens
-
-The following tokens MUST be preserved exactly as shown — same casing, same punctuation,
-never translated:
-
-| Token | Type |
-|-------|------|
-| `AGENTS.md` | Filename |
-| `FABRIC.md` | Filename |
-| `.fabric/agents.meta.json` | Path |
-| `.fabric/init-context.json` | Path |
-| `.fabric/forensic.json` | Path |
-| `.fabric/bootstrap/README.md` | Path |
-| `MUST` | Keyword |
-| `NEVER` | Keyword |
-| `fab init` | CLI command |
-| `fabric doctor --fix` | CLI command |
-| `<!-- fab:index -->` | HTML comment marker |
-| `@HUMAN` | Section marker |

package/templates/skill-source/fabric-init/clients.json

@@ -1,17 +0,0 @@
-{
-  "claude": {
-    "outputName": "fabric-init",
-    "frontmatter": {
-      "name": "fabric-init",
-      "description": "Use this skill when fab init just completed, when forensic.json was generated, or when the user is asking to initialize AGENTS.md. This skill runs a 3-phase initialization interview, writes .fabric/init-context.json, generates layered AGENTS.md, and updates .fabric/agents.meta.json.",
-      "allowed-tools": ["Read", "Write", "Glob", "Grep", "Bash"]
-    }
-  },
-  "codex": {
-    "outputName": "fabric-init",
-    "frontmatter": {
-      "name": "fabric-init",
-      "description": "Use this skill when .fabric/forensic.json exists and this repository still needs the remaining Fabric initialization steps."
-    }
-  }
-}