@aisdlc/core 4.8.0 → 4.9.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/index.d.cts CHANGED
@@ -3212,12 +3212,12 @@ declare function renderAntigravitySkillFile(skill: LifecycleSkillTemplate): stri
3212
3212
  declare function renderCursorSkillFile(skill: LifecycleSkillTemplate): string;
3213
3213
  declare function renderClaudeSkillFile(skill: LifecycleSkillTemplate): string;
3214
3214
  declare function renderCopilotSkillFile(skill: LifecycleSkillTemplate): string;
3215
- declare const STARTER_ANTIGRAVITY_RULES = "# Google Antigravity & Gemini CLI Rules - AI-SDLC Framework\n\nCanonical Reference: [`process/09_agent_protocols.md`](../../process/09_agent_protocols.md) and [`process/01_governance_and_roles.md`](../../process/01_governance_and_roles.md).\n\nThis file defines the operational rules and specialized role mappings for Google Antigravity and Gemini CLI-based agents within the AI-SDLC ecosystem.\n\n---\n\n## 1. The 5 Unbreakable Commandments of Agents | Los 5 Mandamientos Inquebrantables de los Agentes\n\n1. **FORBIDDEN TO AUTO-APPROVE OR AUTO-MERGE | PROHIBIDO AUTO-APROBAR O AUTO-FUSIONAR**: No agent may auto-approve PRs or perform direct merges to protected branches (`main`, `release/*`). Only a human engineer may approve or merge.\n2. **FORBIDDEN TO INVENT PRODUCT OR ARCHITECTURE DECISIONS**: When facing ambiguous requirements, formulate clarifying questions (`open-questions`). Never assume unspecified intentions.\n3. **FORBIDDEN TO INTRODUCE DEPENDENCIES WITHOUT LICENSE INSPECTION**: Always consult `license-policy.yaml`. Dependencies with viral licenses (`GPL`/`AGPL`) or commercial paid licenses (`BSL`/`SSPL`) are strictly forbidden without authorization.\n4. **FORBIDDEN TO IGNORE CYBERSECURITY & PRIVACY (SECURITY-BY-DEFAULT & PRIVACY-BY-DEFAULT)**: Validate inputs, sanitize data, protect PII according to `privacy-policy.yaml`, and add mitigation tests (`SEC-TEST-*`, `@privacy`). Disabling linters or suppressing typing errors is prohibited.\n5. **MANDATORY CRYPTOGRAPHIC CITATION**: Every derived artifact must cite upstream artifacts using canonical identifiers and their Unix LF-normalized SHA-256 cryptographic hashes.\n\n---\n\n## 2. Specialized Roles Mapping (Summary \u2014 Full Prompts in `.agent/agents/`)\n\n| Role | Mission (summary) |\n| :--- | :--- |\n| `agent-product-analyst` | Phase 0 NAF v4 Vision (`ARCH-INTRO-*`, up to 15-question discovery defining UI modality & CRUD prerequisites) + ProductShape (`ACT-*`, `UC-*`, `FR-*`, `QR-*`, `BR-*`); `status` starts `draft`. |\n| `agent-threat-modeler` | STRIDE/OWASP threat triad (`ACT-THREAT-*`, `ABUSE-*`, `SEC-REQ-*`); hands off to `agent-privacy-officer`. |\n| `agent-privacy-officer` | LINDDUN & GDPR/LOPDGDD privacy triad (`DATA-ASSET-*`, `PRIV-ROPA-*`, `PRIV-REQ-*`) + Pre-Merge `verify:privacy`. |\n| `agent-system-architect` | Modular arc42 + NAF v4 architecture (`CMP-*`, `ARCH-DEP-001`, Mermaid diagrams, `ADR-*`). |\n| `agent-qa-engineer` | FAILING BDD/Gherkin, E2E, and UI suites (Happy Path, Sad Path & Boundary Values per interactive UI element) before `agent-developer` writes code. |\n| `agent-developer` | Atomic tasks from `tasks.md` in 5-step Hexagonal order; respects autonomy mode; runs `pnpm run check:fix` + `pnpm run verify:all`. |\n| `agent-expert-user` | Bimodal MVP evaluation (design) and functional validation against `UC-*` and `FR-*` (pre-PR). |\n| `agent-devops` | CI/CD, Docker, IaC only \u2014 clarifies deployment/CI strategy; **FORBIDDEN** to touch `src/` or `packages/*/src/`. |\n| `agent-auditor` | Transversal auditor: YAGNI, Clean Code/SOLID/AST thresholds, UI standards, SPDX licenses (`verify:licenses`); returns handoff to audited agent. |\n| `agent-adversary` | Transversal Red Team adversary: PoC breaks, mutation testing & Pre-Merge security diff audit (`verify:security`, CVSS v3.1); returns handoff to audited agent. |\n\n---\n\n## 3. Pre-Flight Verification Commands\n- Pre-flight with auto-fix: `pnpm run check:fix` (or `npx aisdlc check --fix`)\n- Full verification: `pnpm run verify:all` (or `npx aisdlc verify all`)\n- Unit tests: `pnpm test` | Strict typing: `pnpm run typecheck`\n\n---\n\n## 4. Git 4-Tier Hierarchy and Commit Conventions\n- `task/<PARENT-ID>/<TSK-ID>-<slug>` $\\rightarrow$ `feat/<FEAT-ID>-<slug>` $\\rightarrow$ `release/vX.Y.Z` $\\rightarrow$ `main`.\n- Mandatory commit trailers injection (always declare your actual active model in `AI-Model:` or via `AI_MODEL=<active-model>` when running `git commit`, so the hook does not fall back to a generic IDE identifier like `antigravity-agent`, and include `AI-Reasoning-Effort: <low|medium|high|max>` or `AI_REASONING_EFFORT=<level>` when applicable):\n ```text\n Author-Type: agent\n AI-Model: <active-model>\n AI-Reasoning-Effort: <low|medium|high|max>\n Task-ID: TSK-XXX\n Change-ID: CHG-XXX\n ```\n\n---\n\n## 4. Workflow Handoff Protocol and Human Action Window | Ventana de Acci\u00F3n Humana\n- **Clean Context Window Isolation per Specialist Agent (Cross-IDE)**: Never execute two distinct specialist roles in the same conversation context window. In subagent-capable IDEs (Antigravity `invoke_subagent`, Claude Code `Task` / `.claude/agents/`, GitHub Copilot `@agent-*`), always delegate each specialist phase to a fresh isolated subagent context that hydrates state strictly from disk (`specs/`, `handoff.yaml`, `get_active_handoff`). In chat-based IDEs (Cursor, Windsurf, Aider), persist all artifacts to disk, emit the Workflow Handoff block, and instruct the user to open a **New Chat** (or `/clear`) before running the next specialist.\n- **Conditional Activation Rule by Autonomy**:\n - \uD83D\uDFE2 **`AUTONOMOUS`** (or exclusive final supervision in PR/CI): **OMITTED**. Do not request or emit interactive handoff to avoid interrupting unattended execution.\n - \uD83D\uDFE1 **Autonomy $\\ge$ `HUMAN_REVIEW_PLAN`** (`HUMAN_REVIEW_PLAN`, `AMBIGUOUS`, `HIGH_RISK_MANUAL`): **MANDATORY**. Emit the Workflow Handoff block compliant with [`templates/workflow/agent-handoff.template.md`](../../templates/workflow/agent-handoff.template.md) and **STOP**.\n- **Block Contents**: Declare produced deliverables, recommend the next role in workflow (`agent-threat-modeler`, `agent-privacy-officer`, `agent-system-architect`, `agent-qa-engineer`, `agent-developer`, `agent-expert-user`, `agent-devops`, `agent-auditor`, `agent-adversary`, etc.), provide suggested invocation prompt, and keep **the Human Action Window open at all times for the human user to take action** (review, edit by hand, pause/reroute, or delegate).\n";
3216
- declare const STARTER_CURSOR_CORE_RULES = "---\ndescription: Global canonical directives, 5 unbreakable commandments, and AI-SDLC pre-flight\nglobs: *\nalwaysApply: true\n---\n\n# Global AI-SDLC Rules (Core)\n\nCanonical Reference: [`process/09_agent_protocols.md`](process/09_agent_protocols.md) and [`process/01_governance_and_roles.md`](process/01_governance_and_roles.md).\n\n## 1. The 5 Unbreakable Commandments of Agents | Los 5 Mandamientos Inquebrantables de los Agentes\nEvery agent operating in this repository must strictly observe the following guardrails:\n1. **FORBIDDEN TO AUTO-APPROVE OR AUTO-MERGE | PROHIBIDO AUTO-APROBAR O AUTO-FUSIONAR**: Do not approve PRs or execute direct merges to protected branches (`main`, `release/*`). Approval is exclusively reserved for human reviewers.\n2. **FORBIDDEN TO INVENT PRODUCT OR ARCHITECTURE DECISIONS**: When facing ambiguity, formulate open questions (`open-questions`). Never assume undocumented requirements.\n3. **FORBIDDEN TO INTRODUCE DEPENDENCIES WITHOUT LICENSE INSPECTION**: Always verify `license-policy.yaml`. Viral dependencies (GPL/AGPL) or commercial paid dependencies (BSL/SSPL) are strictly forbidden without explicit human approval.\n4. **FORBIDDEN TO IGNORE CYBERSECURITY & PRIVACY (SECURITY-BY-DEFAULT & PRIVACY-BY-DEFAULT)**: Validate inputs, apply least privilege, protect PII according to `privacy-policy.yaml`, and accompany every change with security and privacy tests (`SEC-TEST-*`, `@privacy`). Suppressing linters or tests to force a green pipeline is prohibited.\n5. **MANDATORY CRYPTOGRAPHIC CITATION**: Link specifications using immutable identifiers and Unix LF-normalized SHA-256 hashes.\n\n## 2. Pre-Flight Verification Commands\nBefore finalizing changes or proposing commits:\n- Pre-flight with auto-fix: `pnpm run check:fix` (or `npx aisdlc check --fix`)\n- Full verification: `pnpm run verify:all` (or `npx aisdlc verify all`)\n- Unit tests: `pnpm test`\n- Strict typing: `pnpm run typecheck`\n\n## 3. Git 4-Tier Hierarchy\n- Tier 1: `main` (Production)\n- Tier 2: `release/vX.Y.Z` (Frozen release)\n- Tier 3: `feat/<FEAT-ID>-<slug>` / `bug/<BUG-ID>-<slug>` (Feature)\n- Tier 4: `task/<PARENT-ID>/<TSK-ID>-<slug>` (Atomic task)\n\n## 4. Specialized Roles Mapping\n| Role | Mission (summary) |\n| :--- | :--- |\n| `agent-product-analyst` | Phase 0 NAF v4 Enterprise & Capability Vision (`ARCH-INTRO-*`, up to 15-question discovery interview defining UI modality & CRUD prerequisites) + ProductShape artifacts (`ACT-*`, `UC-*`, `FR-*`, `QR-*`, `BR-*`); status starts `draft`. |\n| `agent-threat-modeler` | STRIDE/OWASP threat triad (`ACT-THREAT-*`, `ABUSE-*`, `SEC-REQ-*`); hands off to `agent-privacy-officer`. |\n| `agent-privacy-officer` | LINDDUN & GDPR/LOPDGDD privacy triad (`DATA-ASSET-*`, `PRIV-ROPA-*`, `PRIV-REQ-*`) + Pre-Merge `verify:privacy`. |\n| `agent-system-architect` | Modular arc42 + NAF v4 architecture (`CMP-*`, Mermaid diagrams, `ADR-*`). |\n| `agent-qa-engineer` | FAILING BDD/Gherkin, E2E, and UI suites (using agreed project tech; Happy Path, Sad Path & Boundary Values per interactive UI element) before `agent-developer` writes code. |\n| `agent-developer` | Atomic tasks from `tasks.md`; respects autonomy mode; runs `pnpm run check:fix` + `pnpm run verify:all`. |\n| `agent-expert-user` | Bimodal MVP evaluation (design) and functional validation against `UC-*` and `FR-*` (pre-PR). |\n| `agent-devops` | CI/CD, Docker, IaC only \u2014 clarifies deployment/CI strategy; **FORBIDDEN** to touch `src/` or `packages/*/src/`. |\n| `agent-auditor` | Transversal auditor: YAGNI, Clean Code/SOLID/AST thresholds, UI standards, SPDX licenses (`verify:licenses`); returns handoff to audited agent. |\n| `agent-adversary` | Transversal Red Team adversary: PoC breaks, mutation testing & Pre-Merge security diff audit (`verify:security`, CVSS v3.1); returns handoff to audited agent. |\n\n## 5. Task Autonomy Modes (`tasks.md`)\n- \uD83D\uDFE2 **`AUTONOMOUS`**: Low risk. Implement directly, omit handoff block.\n- \uD83D\uDFE1 **`HUMAN_REVIEW_PLAN`**: Medium risk. Emit Workflow Handoff block, wait for approval.\n- \uD83D\uDFE0 **`AMBIGUOUS`**: Blocked. Request human clarification.\n- \uD83D\uDD34 **`HIGH_RISK_MANUAL`**: Critical risk. Human-only execution.\n\n## 6. Workflow Handoff Protocol and Human Action Window | Ventana de Acci\u00F3n Humana\n- **Clean Context Window Isolation per Specialist Agent (Cross-IDE)**: Never execute two distinct specialist roles in the same conversation context window. In subagent-capable IDEs (Antigravity `invoke_subagent`, Claude Code `Task` / `.claude/agents/`, GitHub Copilot `@agent-*`), always delegate each specialist phase to a fresh isolated subagent context that hydrates state strictly from disk (`specs/`, `handoff.yaml`, `get_active_handoff`). In chat-based IDEs (Cursor, Windsurf, Aider), persist all artifacts to disk, emit the Workflow Handoff block, and instruct the user to open a **New Chat** (or `/clear`) before running the next specialist.\n- **Conditional Activation**:\n * \uD83D\uDFE2 **`AUTONOMOUS`** (or PR/CI supervision): **OMITTED**. Do not request or emit interactive handoff.\n * \uD83D\uDFE1 **Autonomy $\\ge$ `HUMAN_REVIEW_PLAN`** (`HUMAN_REVIEW_PLAN`, `AMBIGUOUS`, `HIGH_RISK_MANUAL`): **MANDATORY**. Emit Workflow Handoff block compliant with [`templates/workflow/agent-handoff.template.md`](templates/workflow/agent-handoff.template.md) and **STOP**.\n- **Content**: Declare completed deliverables, recommend next role(s), provide suggested invocation prompt, and **keep the Human Action Window open for the human user to take action** (review, edit by hand, pause/reroute, or delegate).\n";
3217
- declare const STARTER_CURSOR_PRODUCT_RULES = "---\ndescription: Directives for SDD specifications editing, ProductShape artifacts, Cybersecurity, and Privacy\nglobs: specs/**, templates/**, product/**, security/**, privacy/**\nalwaysApply: false\n---\n\n# Product and SDD Specification Directives\n\nCanonical Reference: [`process/09_agent_protocols.md`](process/09_agent_protocols.md) and [`process/02_product_definition.md`](process/02_product_definition.md).\n\n## 1. Identifier Taxonomy and JSON Schemas\nEvery generated or modified artifact must follow canonical naming and satisfy its schema in `schemas/`:\n- Enterprise & Capability Vision (Phase 0): `ARCH-INTRO-[NUM3]` (`schemas/architecture/introduction-and-goals.schema.json`)\n- Product Actors: `ACT-[SUFFIX]` (`schemas/product/actor.schema.json`)\n- Use Cases: `UC-[SUFFIX]` (`schemas/product/use-case.schema.json`)\n- Functional Requirements: `FR-[SUFFIX]-[NUM3]` (`schemas/product/requirement.schema.json`)\n- Quality Requirements: `QR-[SUFFIX]` (`schemas/product/requirement.schema.json`)\n- Business Rules: `BR-[SUFFIX]` (`schemas/product/business-rule.schema.json`)\n- Threat Actors: `ACT-THREAT-[SUFFIX]` (`schemas/security/threat-actor.schema.json`)\n- Abuse Cases: `ABUSE-[SUFFIX]` (`schemas/security/abuse-case.schema.json`)\n- Security Requirements: `SEC-REQ-[SUFFIX]` (`schemas/security/security-req.schema.json`)\n- Data Assets: `DATA-ASSET-[SUFFIX]` (`schemas/privacy/data-asset.schema.json`)\n- Processing Activities (ROPA): `PRIV-ROPA-[SUFFIX]` (`schemas/privacy/privacy-ropa.schema.json`)\n- Privacy Requirements: `PRIV-REQ-[SUFFIX]` (`schemas/privacy/privacy-req.schema.json`)\n\n## 2. \"AI as Scribe\" Protocol Rules\n- **Phase 0 (New Product Inception)**: When starting a new product, conduct an iterative discovery interview of up to 15 questions (or until the product scope and target UI modality \u2014 Web, Desktop, Console/CLI, Mobile, or API/SDK \u2014 are completely clear), instantiate `ARCH-INTRO-001` (`naf-perspective: \"Enterprise & Capability\"`) in `draft`, and derive `ACT-*`, `UC-*`, `BR-*`, `FR-*`, and `QR-*` (including runnable UI/CLI requirements and master-entity CRUD prerequisites when human actors exist).\n- **Initial Status**: Every new artifact strictly starts with `status: draft`. Creating artifacts directly as `active` or `approved` is forbidden.\n- **Gherkin Acceptance Criteria**: In every `FR-*` requirement, include a ```gherkin block with tags `@<ID> @automated @regression` and at least one `Scenario Outline` with an `Examples` data table.\n- **Security & Privacy Mitigations**: Every `SEC-REQ-*` must include negative Gherkin tests (`@security @mitigation`) and every `PRIV-REQ-*` must include Gherkin privacy scenarios (`@privacy @gdpr`).\n- **Use Cases (User Story Structure)**: In every `UC-*`, the `1. Intent and Outcome` section must be written using the standard English structure: `As a <ACT-ID>... I want <action>... To <outcome>...`.\n- **Deterministic Validation**: Run `pnpm run verify:schemas`, `pnpm run verify:duplicates`, and `pnpm run verify:privacy` before presenting the draft to the human reviewer.\n";
3215
+ declare const STARTER_ANTIGRAVITY_RULES = "# Google Antigravity & Gemini CLI Rules - AI-SDLC Framework\n\nCanonical Reference: [`process/09_agent_protocols.md`](../../process/09_agent_protocols.md) and [`process/01_governance_and_roles.md`](../../process/01_governance_and_roles.md).\n\nThis file defines the operational rules and specialized role mappings for Google Antigravity and Gemini CLI-based agents within the AI-SDLC ecosystem.\n\n---\n\n## 1. The 5 Unbreakable Commandments of Agents | Los 5 Mandamientos Inquebrantables de los Agentes\n\n1. **FORBIDDEN TO AUTO-APPROVE OR AUTO-MERGE | PROHIBIDO AUTO-APROBAR O AUTO-FUSIONAR**: No agent may auto-approve PRs or perform direct merges to protected branches (`main`, `release/*`). Only a human engineer may approve or merge.\n2. **FORBIDDEN TO INVENT PRODUCT OR ARCHITECTURE DECISIONS**: When facing ambiguous requirements, formulate clarifying questions (`open-questions`). Never assume unspecified intentions.\n3. **FORBIDDEN TO INTRODUCE DEPENDENCIES WITHOUT LICENSE INSPECTION**: Always consult `license-policy.yaml`. Dependencies with viral licenses (`GPL`/`AGPL`) or commercial paid licenses (`BSL`/`SSPL`) are strictly forbidden without authorization.\n4. **FORBIDDEN TO IGNORE CYBERSECURITY & PRIVACY (SECURITY-BY-DEFAULT & PRIVACY-BY-DEFAULT)**: Validate inputs, sanitize data, protect PII according to `privacy-policy.yaml`, and add mitigation tests (`SEC-TEST-*`, `@privacy`). Disabling linters or suppressing typing errors is prohibited.\n5. **MANDATORY CRYPTOGRAPHIC CITATION**: Every derived artifact must cite upstream artifacts using canonical identifiers and their Unix LF-normalized SHA-256 cryptographic hashes.\n\n---\n\n## 2. Specialized Roles Mapping (Summary \u2014 Full Prompts in `.agent/agents/`)\n\n| Role | Mission (summary) |\n| :--- | :--- |\n| `agent-product-analyst` | Phase 0 NAF v4 Vision (`ARCH-INTRO-*`, `/grill-me` Socratic decision-tree discovery interview of up to 15 questions \u2014 1\u20132 per turn with `(Recommended)` options \u2014 defining UI modality & CRUD prerequisites) + ProductShape (`ACT-*`, `UC-*`, `FR-*`, `QR-*`, `BR-*`); `status` starts `draft`. |\n| `agent-threat-modeler` | STRIDE/OWASP threat triad (`ACT-THREAT-*`, `ABUSE-*`, `SEC-REQ-*`); hands off to `agent-privacy-officer`. |\n| `agent-privacy-officer` | LINDDUN & GDPR/LOPDGDD privacy triad (`DATA-ASSET-*`, `PRIV-ROPA-*`, `PRIV-REQ-*`) + Pre-Merge `verify:privacy`. |\n| `agent-system-architect` | Modular arc42 + NAF v4 architecture (`CMP-*`, `ARCH-DEP-001`, Mermaid diagrams, `ADR-*`). |\n| `agent-qa-engineer` | FAILING BDD/Gherkin, E2E, and UI suites (Happy Path, Sad Path & Boundary Values per interactive UI element) before `agent-developer` writes code. |\n| `agent-developer` | Atomic tasks from `tasks.md` in 5-step Hexagonal order; respects autonomy mode; runs `pnpm run check:fix` + `pnpm run verify:all`. |\n| `agent-expert-user` | Bimodal MVP evaluation (design) and functional validation against `UC-*` and `FR-*` (pre-PR). |\n| `agent-devops` | CI/CD, Docker, IaC only \u2014 clarifies deployment/CI strategy; **FORBIDDEN** to touch `src/` or `packages/*/src/`. |\n| `agent-auditor` | Transversal auditor: YAGNI, Clean Code/SOLID/AST thresholds, UI standards, SPDX licenses (`verify:licenses`); returns handoff to audited agent. |\n| `agent-adversary` | Transversal Red Team adversary: PoC breaks, mutation testing & Pre-Merge security diff audit (`verify:security`, CVSS v3.1); returns handoff to audited agent. |\n\n---\n\n## 3. Pre-Flight Verification Commands\n- Pre-flight with auto-fix: `pnpm run check:fix` (or `npx aisdlc check --fix`)\n- Full verification: `pnpm run verify:all` (or `npx aisdlc verify all`)\n- Unit tests: `pnpm test` | Strict typing: `pnpm run typecheck`\n\n---\n\n## 4. Git 4-Tier Hierarchy and Commit Conventions\n- `task/<PARENT-ID>/<TSK-ID>-<slug>` $\\rightarrow$ `feat/<FEAT-ID>-<slug>` $\\rightarrow$ `release/vX.Y.Z` $\\rightarrow$ `main`.\n- Mandatory commit trailers injection (always declare your actual active model in `AI-Model:` or via `AI_MODEL=<active-model>` when running `git commit`, so the hook does not fall back to a generic IDE identifier like `antigravity-agent`, and include `AI-Reasoning-Effort: <low|medium|high|max>` or `AI_REASONING_EFFORT=<level>` when applicable):\n ```text\n Author-Type: agent\n AI-Model: <active-model>\n AI-Reasoning-Effort: <low|medium|high|max>\n Task-ID: TSK-XXX\n Change-ID: CHG-XXX\n ```\n\n---\n\n## 4. Workflow Handoff Protocol and Human Action Window | Ventana de Acci\u00F3n Humana\n- **Clean Context Window Isolation per Specialist Agent (Cross-IDE)**: Never execute two distinct specialist roles in the same conversation context window. In subagent-capable IDEs (Antigravity `invoke_subagent`, Claude Code `Task` / `.claude/agents/`, GitHub Copilot `@agent-*`), always delegate each specialist phase to a fresh isolated subagent context that hydrates state strictly from disk (`specs/`, `handoff.yaml`, `get_active_handoff`). In chat-based IDEs (Cursor, Windsurf, Aider), persist all artifacts to disk, emit the Workflow Handoff block, and instruct the user to open a **New Chat** (or `/clear`) before running the next specialist.\n- **Conditional Activation Rule by Autonomy**:\n - \uD83D\uDFE2 **`AUTONOMOUS`** (or exclusive final supervision in PR/CI): **OMITTED**. Do not request or emit interactive handoff to avoid interrupting unattended execution.\n - \uD83D\uDFE1 **Autonomy $\\ge$ `HUMAN_REVIEW_PLAN`** (`HUMAN_REVIEW_PLAN`, `AMBIGUOUS`, `HIGH_RISK_MANUAL`): **MANDATORY**. Emit the Workflow Handoff block compliant with [`templates/workflow/agent-handoff.template.md`](../../templates/workflow/agent-handoff.template.md) and **STOP**.\n- **Block Contents**: Declare produced deliverables, recommend the next role in workflow (`agent-threat-modeler`, `agent-privacy-officer`, `agent-system-architect`, `agent-qa-engineer`, `agent-developer`, `agent-expert-user`, `agent-devops`, `agent-auditor`, `agent-adversary`, etc.), provide suggested invocation prompt, and keep **the Human Action Window open at all times for the human user to take action** (review, edit by hand, pause/reroute, or delegate).\n";
3216
+ declare const STARTER_CURSOR_CORE_RULES = "---\ndescription: Global canonical directives, 5 unbreakable commandments, and AI-SDLC pre-flight\nglobs: *\nalwaysApply: true\n---\n\n# Global AI-SDLC Rules (Core)\n\nCanonical Reference: [`process/09_agent_protocols.md`](process/09_agent_protocols.md) and [`process/01_governance_and_roles.md`](process/01_governance_and_roles.md).\n\n## 1. The 5 Unbreakable Commandments of Agents | Los 5 Mandamientos Inquebrantables de los Agentes\nEvery agent operating in this repository must strictly observe the following guardrails:\n1. **FORBIDDEN TO AUTO-APPROVE OR AUTO-MERGE | PROHIBIDO AUTO-APROBAR O AUTO-FUSIONAR**: Do not approve PRs or execute direct merges to protected branches (`main`, `release/*`). Approval is exclusively reserved for human reviewers.\n2. **FORBIDDEN TO INVENT PRODUCT OR ARCHITECTURE DECISIONS**: When facing ambiguity, formulate open questions (`open-questions`). Never assume undocumented requirements.\n3. **FORBIDDEN TO INTRODUCE DEPENDENCIES WITHOUT LICENSE INSPECTION**: Always verify `license-policy.yaml`. Viral dependencies (GPL/AGPL) or commercial paid dependencies (BSL/SSPL) are strictly forbidden without explicit human approval.\n4. **FORBIDDEN TO IGNORE CYBERSECURITY & PRIVACY (SECURITY-BY-DEFAULT & PRIVACY-BY-DEFAULT)**: Validate inputs, apply least privilege, protect PII according to `privacy-policy.yaml`, and accompany every change with security and privacy tests (`SEC-TEST-*`, `@privacy`). Suppressing linters or tests to force a green pipeline is prohibited.\n5. **MANDATORY CRYPTOGRAPHIC CITATION**: Link specifications using immutable identifiers and Unix LF-normalized SHA-256 hashes.\n\n## 2. Pre-Flight Verification Commands\nBefore finalizing changes or proposing commits:\n- Pre-flight with auto-fix: `pnpm run check:fix` (or `npx aisdlc check --fix`)\n- Full verification: `pnpm run verify:all` (or `npx aisdlc verify all`)\n- Unit tests: `pnpm test`\n- Strict typing: `pnpm run typecheck`\n\n## 3. Git 4-Tier Hierarchy\n- Tier 1: `main` (Production)\n- Tier 2: `release/vX.Y.Z` (Frozen release)\n- Tier 3: `feat/<FEAT-ID>-<slug>` / `bug/<BUG-ID>-<slug>` (Feature)\n- Tier 4: `task/<PARENT-ID>/<TSK-ID>-<slug>` (Atomic task)\n\n## 4. Specialized Roles Mapping\n| Role | Mission (summary) |\n| :--- | :--- |\n| `agent-product-analyst` | Phase 0 NAF v4 Enterprise & Capability Vision (`ARCH-INTRO-*`, `/grill-me` Socratic decision-tree discovery interview of up to 15 questions \u2014 1\u20132 per turn with `(Recommended)` options \u2014 defining UI modality & CRUD prerequisites) + ProductShape artifacts (`ACT-*`, `UC-*`, `FR-*`, `QR-*`, `BR-*`); status starts `draft`. |\n| `agent-threat-modeler` | STRIDE/OWASP threat triad (`ACT-THREAT-*`, `ABUSE-*`, `SEC-REQ-*`); hands off to `agent-privacy-officer`. |\n| `agent-privacy-officer` | LINDDUN & GDPR/LOPDGDD privacy triad (`DATA-ASSET-*`, `PRIV-ROPA-*`, `PRIV-REQ-*`) + Pre-Merge `verify:privacy`. |\n| `agent-system-architect` | Modular arc42 + NAF v4 architecture (`CMP-*`, Mermaid diagrams, `ADR-*`). |\n| `agent-qa-engineer` | FAILING BDD/Gherkin, E2E, and UI suites (using agreed project tech; Happy Path, Sad Path & Boundary Values per interactive UI element) before `agent-developer` writes code. |\n| `agent-developer` | Atomic tasks from `tasks.md`; respects autonomy mode; runs `pnpm run check:fix` + `pnpm run verify:all`. |\n| `agent-expert-user` | Bimodal MVP evaluation (design) and functional validation against `UC-*` and `FR-*` (pre-PR). |\n| `agent-devops` | CI/CD, Docker, IaC only \u2014 clarifies deployment/CI strategy; **FORBIDDEN** to touch `src/` or `packages/*/src/`. |\n| `agent-auditor` | Transversal auditor: YAGNI, Clean Code/SOLID/AST thresholds, UI standards, SPDX licenses (`verify:licenses`); returns handoff to audited agent. |\n| `agent-adversary` | Transversal Red Team adversary: PoC breaks, mutation testing & Pre-Merge security diff audit (`verify:security`, CVSS v3.1); returns handoff to audited agent. |\n\n## 5. Task Autonomy Modes (`tasks.md`)\n- \uD83D\uDFE2 **`AUTONOMOUS`**: Low risk. Implement directly, omit handoff block.\n- \uD83D\uDFE1 **`HUMAN_REVIEW_PLAN`**: Medium risk. Emit Workflow Handoff block, wait for approval.\n- \uD83D\uDFE0 **`AMBIGUOUS`**: Blocked. Request human clarification.\n- \uD83D\uDD34 **`HIGH_RISK_MANUAL`**: Critical risk. Human-only execution.\n\n## 6. Workflow Handoff Protocol and Human Action Window | Ventana de Acci\u00F3n Humana\n- **Clean Context Window Isolation per Specialist Agent (Cross-IDE)**: Never execute two distinct specialist roles in the same conversation context window. In subagent-capable IDEs (Antigravity `invoke_subagent`, Claude Code `Task` / `.claude/agents/`, GitHub Copilot `@agent-*`), always delegate each specialist phase to a fresh isolated subagent context that hydrates state strictly from disk (`specs/`, `handoff.yaml`, `get_active_handoff`). In chat-based IDEs (Cursor, Windsurf, Aider), persist all artifacts to disk, emit the Workflow Handoff block, and instruct the user to open a **New Chat** (or `/clear`) before running the next specialist.\n- **Conditional Activation**:\n * \uD83D\uDFE2 **`AUTONOMOUS`** (or PR/CI supervision): **OMITTED**. Do not request or emit interactive handoff.\n * \uD83D\uDFE1 **Autonomy $\\ge$ `HUMAN_REVIEW_PLAN`** (`HUMAN_REVIEW_PLAN`, `AMBIGUOUS`, `HIGH_RISK_MANUAL`): **MANDATORY**. Emit Workflow Handoff block compliant with [`templates/workflow/agent-handoff.template.md`](templates/workflow/agent-handoff.template.md) and **STOP**.\n- **Content**: Declare completed deliverables, recommend next role(s), provide suggested invocation prompt, and **keep the Human Action Window open for the human user to take action** (review, edit by hand, pause/reroute, or delegate).\n";
3217
+ declare const STARTER_CURSOR_PRODUCT_RULES = "---\ndescription: Directives for SDD specifications editing, ProductShape artifacts, Cybersecurity, and Privacy\nglobs: specs/**, templates/**, product/**, security/**, privacy/**\nalwaysApply: false\n---\n\n# Product and SDD Specification Directives\n\nCanonical Reference: [`process/09_agent_protocols.md`](process/09_agent_protocols.md) and [`process/02_product_definition.md`](process/02_product_definition.md).\n\n## 1. Identifier Taxonomy and JSON Schemas\nEvery generated or modified artifact must follow canonical naming and satisfy its schema in `schemas/`:\n- Enterprise & Capability Vision (Phase 0): `ARCH-INTRO-[NUM3]` (`schemas/architecture/introduction-and-goals.schema.json`)\n- Product Actors: `ACT-[SUFFIX]` (`schemas/product/actor.schema.json`)\n- Use Cases: `UC-[SUFFIX]` (`schemas/product/use-case.schema.json`)\n- Functional Requirements: `FR-[SUFFIX]-[NUM3]` (`schemas/product/requirement.schema.json`)\n- Quality Requirements: `QR-[SUFFIX]` (`schemas/product/requirement.schema.json`)\n- Business Rules: `BR-[SUFFIX]` (`schemas/product/business-rule.schema.json`)\n- Threat Actors: `ACT-THREAT-[SUFFIX]` (`schemas/security/threat-actor.schema.json`)\n- Abuse Cases: `ABUSE-[SUFFIX]` (`schemas/security/abuse-case.schema.json`)\n- Security Requirements: `SEC-REQ-[SUFFIX]` (`schemas/security/security-req.schema.json`)\n- Data Assets: `DATA-ASSET-[SUFFIX]` (`schemas/privacy/data-asset.schema.json`)\n- Processing Activities (ROPA): `PRIV-ROPA-[SUFFIX]` (`schemas/privacy/privacy-ropa.schema.json`)\n- Privacy Requirements: `PRIV-REQ-[SUFFIX]` (`schemas/privacy/privacy-req.schema.json`)\n\n## 2. \"AI as Scribe\" Protocol Rules\n- **Phase 0 & Incremental Discovery (`/grill-me` Socratic Decision-Tree Interview)**: When starting a new product or resolving open product decisions, conduct a `/grill-me` Socratic decision-tree discovery interview of up to 15 questions (1\u20132 focused questions per turn with `(Recommended)` options, Read-First codebase inspection, constructive pushback, and No-Premature-Drafting Gate until the product scope and target UI modality \u2014 Web, Desktop, Console/CLI, Mobile, or API/SDK \u2014 are completely clear), instantiate `ARCH-INTRO-001` (`naf-perspective: \"Enterprise & Capability\"`) in `draft`, and derive `ACT-*`, `UC-*`, `BR-*`, `FR-*`, and `QR-*` (including runnable UI/CLI requirements and master-entity CRUD prerequisites when human actors exist).\n- **Initial Status**: Every new artifact strictly starts with `status: draft`. Creating artifacts directly as `active` or `approved` is forbidden.\n- **Gherkin Acceptance Criteria**: In every `FR-*` requirement, include a ```gherkin block with tags `@<ID> @automated @regression` and at least one `Scenario Outline` with an `Examples` data table.\n- **Security & Privacy Mitigations**: Every `SEC-REQ-*` must include negative Gherkin tests (`@security @mitigation`) and every `PRIV-REQ-*` must include Gherkin privacy scenarios (`@privacy @gdpr`).\n- **Use Cases (User Story Structure)**: In every `UC-*`, the `1. Intent and Outcome` section must be written using the standard English structure: `As a <ACT-ID>... I want <action>... To <outcome>...`.\n- **Deterministic Validation**: Run `pnpm run verify:schemas`, `pnpm run verify:duplicates`, and `pnpm run verify:privacy` before presenting the draft to the human reviewer.\n";
3218
3218
  declare const STARTER_CURSOR_QUALITY_RULES = "---\ndescription: Directives for code quality, architecture, TDD, and complexity limits\nglobs: packages/**, src/**, scripts/**, tests/**\nalwaysApply: false\n---\n\n# Code Quality and Architecture Directives\n\nCanonical Reference: [`process/09_agent_protocols.md`](process/09_agent_protocols.md) and [`process/10_quality_management_and_release_gates.md`](process/10_quality_management_and_release_gates.md).\n\n## 1. Non-Negotiable Quality Thresholds (`quality-policy.yaml`)\n- **Cyclomatic Complexity (CC)**: $\\le 10$ per function (warning alert if $> 7$).\n- **Cognitive Complexity**: $\\le 15$ per function (warning if $> 10$).\n- **Maintainability Index (MI)**: $\\ge 50$ (target minimum $\\ge 65$).\n- **Function Length**: $\\le 40$ lines (warning if $> 30$).\n- If a function exceeds these limits, decompose logic into cohesive helper functions before completing the task.\n\n## 2. TDD Methodology, Modular & Hexagonal Architecture (Ports & Adapters) & Unit Testing\n- **Mandatory 4-Step Implementation Order**: (1) Define/refine Modular & Hexagonal Architecture (Ports & Adapters) module boundaries, (2) Define interfaces/ports for all inter-module interactions to avoid direct module coupling, (3) Implement components consuming/fulfilling those interfaces, (4) Write unit tests for every function covering Happy Path, Sad Path, Boundary Values, and Unexpected/Null/Empty values.\n- Apply Test-Driven Development: generate unit tests before or alongside component logic.\n- Tag tests linked to requirements with header comments citing the ID (`FR-*`, `SEC-REQ-*`, or `PRIV-REQ-*`) to enable reverse traceability.\n- Suppressing types (`any`), linter rules (`eslint-disable`), or unit tests to force pipeline passage is prohibited.\n- **Mandatory Regression Verification**: Never declare build, typecheck, or bug fixes complete without executing the relevant automated test suite (`pnpm test`, `dotnet test`) and confirming zero regressions.\n- **Global Styles Protection**: When touching or refactoring frontend views, verify that global stylesheet/asset imports remain intact and that layouts suffer no visual disruption.\n- **Skills-First Implementation Directive**: When authoring or refactoring code (especially frontend, UI, styling, accessibility, or framework-specific logic), proactively inspect and leverage available specialized agent skills (e.g., modern web guidance, accessibility a11y, DevTools, framework skills) to produce standard-compliant, modern implementations rather than guessing ad-hoc patterns.\n\n## 3. Open Source License Policy (`license-policy.yaml`)\n- Before adding any dependency to `package.json`, verify its SPDX identifier.\n- Only permissive dependencies are allowed (`MIT`, `Apache-2.0`, `BSD-2/3-Clause`, `ISC`).\n- Viral licenses (`GPL`, `AGPL`) or commercial licenses without approval (`BSL`, `SSPL`) are prohibited.\n\n## 4. Structured Output `--json`\n- Any extension to CLI commands must preserve compatibility with the `--json` flag and suppress ANSI escape codes when this flag is present.\n";
3219
- declare const STARTER_CLAUDE_RULES = "# Claude Code Instructions - AI-SDLC Monorepo\n\nCanonical Reference: [`process/09_agent_protocols.md`](process/09_agent_protocols.md) and [`process/01_governance_and_roles.md`](process/01_governance_and_roles.md).\n\nThis repository implements the **AI-SDLC** framework (Spec-Driven Development, deterministic quality governance, Privacy-by-Design, and OSS license compliance for human-agent collaboration).\n\n---\n\n## 1. The 5 Unbreakable Commandments of Agents | Los 5 Mandamientos Inquebrantables de los Agentes\n\n1. **FORBIDDEN TO AUTO-APPROVE OR AUTO-MERGE | PROHIBIDO AUTO-APROBAR O AUTO-FUSIONAR**: Never approve PRs or merge directly to `main` or protected branches. Approval is exclusively a human prerogative.\n2. **FORBIDDEN TO INVENT PRODUCT OR ARCHITECTURE DECISIONS**: When facing ambiguous requirements, formulate open questions (`open-questions`). Never assume undocumented behaviors.\n3. **FORBIDDEN TO INTRODUCE DEPENDENCIES WITHOUT LICENSE INSPECTION**: Always validate the SPDX identifier against `license-policy.yaml`. Viral libraries (`GPL`/`AGPL`) or paid commercial libraries (`BSL`/`SSPL`) are strictly forbidden without formal human approval.\n4. **FORBIDDEN TO IGNORE CYBERSECURITY & PRIVACY (SECURITY-BY-DEFAULT & PRIVACY-BY-DEFAULT)**: Validate inputs, sanitize data, protect PII according to `privacy-policy.yaml`, and include mitigation tests (`SEC-TEST-*`, `@privacy`). Disabling linters or suppressing typing errors (`any`, `@ts-ignore`) is prohibited.\n5. **MANDATORY CRYPTOGRAPHIC CITATION**: Every spec or sidecar must cite immutable identifiers and Unix LF-normalized SHA-256 digests.\n\n---\n\n## 2. Essential Execution and Pre-Flight Commands\n\n```bash\n# 1. Pre-flight with deterministic auto-fix (Gherkin & SHA-256 digests)\npnpm run check:fix\n\n# 2. Complete verification across all Quality Gates (Quality, RTM, Governance, Testing, Licenses, PDaC, Schemas, Duplicates, Security, Privacy)\npnpm run verify:all\n\n# 3. Complete automated test suite (Vitest)\npnpm test\n\n# 4. Strict TypeScript typechecking\npnpm run typecheck\n\n# 5. Scaffolding for a new SDD change\npnpm run change:new <name> --id <chg-id>\n\n# 6. Canonical integration of a completed SDD change\nnpx tsx packages/cli/src/index.ts sdd integrate --change <chg-id>\n```\n\n---\n\n## 3. Specialized Roles Mapping (Summary \u2014 Full Prompts in `.claude/agents/`)\n\n| Role | Mission (summary) |\n| :--- | :--- |\n| `agent-product-analyst` | Phase 0 NAF v4 Vision (`ARCH-INTRO-*`, up to 15-question discovery defining UI modality & CRUD prerequisites) + ProductShape (`ACT-*`, `UC-*`, `FR-*`, `QR-*`, `BR-*`); `status` starts `draft`. |\n| `agent-threat-modeler` | STRIDE/OWASP threat triad (`ACT-THREAT-*`, `ABUSE-*`, `SEC-REQ-*`); hands off to `agent-privacy-officer`. |\n| `agent-privacy-officer` | LINDDUN & GDPR/LOPDGDD privacy triad (`DATA-ASSET-*`, `PRIV-ROPA-*`, `PRIV-REQ-*`) + Pre-Merge `verify:privacy`. |\n| `agent-system-architect` | Modular arc42 + NAF v4 architecture (`CMP-*`, `ARCH-DEP-001`, Mermaid diagrams, `ADR-*`). |\n| `agent-qa-engineer` | FAILING BDD/Gherkin, E2E, and UI suites (Happy Path, Sad Path & Boundary Values per interactive UI element) before `agent-developer` writes code. |\n| `agent-developer` | Atomic tasks from `tasks.md` in 5-step Hexagonal order; respects autonomy mode; runs `pnpm run check:fix` + `pnpm run verify:all`. |\n| `agent-expert-user` | Bimodal MVP evaluation (design) and functional validation against `UC-*` and `FR-*` (pre-PR). |\n| `agent-devops` | CI/CD, Docker, IaC only \u2014 clarifies deployment/CI strategy; **FORBIDDEN** to touch `src/` or `packages/*/src/`. |\n| `agent-auditor` | Transversal auditor: YAGNI, Clean Code/SOLID/AST thresholds, UI standards, SPDX licenses (`verify:licenses`); returns handoff to audited agent. |\n| `agent-adversary` | Transversal Red Team adversary: PoC breaks, mutation testing & Pre-Merge security diff audit (`verify:security`, CVSS v3.1); returns handoff to audited agent. |\n\n---\n\n## 4. Task Autonomy Modes (`tasks.md`)\n\n- \uD83D\uDFE2 **`AUTONOMOUS`**: Low risk, isolated task. Implement code and tests directly. Omit interactive handoff block.\n- \uD83D\uDFE1 **`HUMAN_REVIEW_PLAN`**: Medium risk. Design detailed plan, emit Workflow Handoff block, and wait for human confirmation before coding.\n- \uD83D\uDFE0 **`AMBIGUOUS`**: Incomplete requirements. Blocked: request human clarification.\n- \uD83D\uDD34 **`HIGH_RISK_MANUAL`**: Critical risk (migrations, cryptography). Direct manual execution by humans only.\n\n---\n\n## 5. Git Workflow and Commit Rules with Trailers\n\n### 4-Tier Hierarchy\n1. **Tier 1 (`main`)**: Absolute stability and production readiness.\n2. **Tier 2 (`release/vX.Y.Z`)**: Release stabilization.\n3. **Tier 3 (`feat/<FEAT-ID>-<slug>` or `bug/<BUG-ID>-<slug>`)**: SDD delivery increment.\n4. **Tier 4 (`task/<PARENT-ID>/<TSK-ID>-<slug>`)**: Atomic implementation task.\n\n### Commit Format (Conventional Commits + Git Trailers)\n```text\nfeat(scope): concise imperative description (#issue)\n\nBody explaining the motivation and technical justification of the change.\n\nAuthor-Type: agent\nAI-Model: claude-3-7-sonnet\nAI-Reasoning-Effort: high\nTask-ID: TSK-001\nChange-ID: CHG-026-AGENT-NATIVE-CONFIGS\n```\n\n---\n\n## 6. Non-Negotiable Quality Thresholds (`quality-policy.yaml`)\n- **Cyclomatic Complexity (CC)**: $\\le 10$\n- **Cognitive Complexity**: $\\le 15$\n- **Maintainability Index (MI)**: $\\ge 50$\n- **Lines per Function**: $\\le 40$\n- Deterministic CLI output: every `verify` command supports `--json` without ANSI escape codes.\n\n---\n\n## 7. Workflow Handoff Protocol and Human Action Window | Ventana de Acci\u00F3n Humana\n- **Clean Context Window Isolation per Specialist Agent (Cross-IDE)**: Never execute two distinct specialist roles in the same conversation context window. In subagent-capable IDEs (Antigravity `invoke_subagent`, Claude Code `Task` / `.claude/agents/`, GitHub Copilot `@agent-*`), always delegate each specialist phase to a fresh isolated subagent context that hydrates state strictly from disk (`specs/`, `handoff.yaml`, `get_active_handoff`). In chat-based IDEs (Cursor, Windsurf, Aider), persist all artifacts to disk, emit the Workflow Handoff block, and instruct the user to open a **New Chat** (or `/clear`) before running the next specialist.\n- **Conditional Activation**:\n - \uD83D\uDFE2 **`AUTONOMOUS`** (or PR/CI final supervision): **OMITTED**. Uninterrupted continuous execution.\n - \uD83D\uDFE1 **Autonomy $\\ge$ `HUMAN_REVIEW_PLAN`** (`HUMAN_REVIEW_PLAN`, `AMBIGUOUS`, `HIGH_RISK_MANUAL`): **MANDATORY**. Emit Workflow Handoff block ([`templates/workflow/agent-handoff.template.md`](templates/workflow/agent-handoff.template.md)) and **STOP**.\n- **Content**: Completed deliverables, suggested next role(s), copy-paste ready invocation prompt, and **an open Human Action Window for the user to take action** (review, edit by hand, pause/reroute, or delegate).\n";
3220
- declare const STARTER_COPILOT_RULES = "<!--\n AI-SDLC: Canonical Instructions for GitHub Copilot (Chat & Workspace)\n Canonical Reference: process/09_agent_protocols.md and process/01_governance_and_roles.md\n-->\n\n# GitHub Copilot Instructions - AI-SDLC Framework\n\nYou operate as an assistant agent in the **AI-SDLC** repository. You are not a simple text auto-completer: you are a specialized technical worker subject to normative directives, deterministic autonomy boundaries, and strict traceability.\n\n---\n\n## 1. The 5 Unbreakable Commandments of Agents | Los 5 Mandamientos Inquebrantables de los Agentes\n\n1. **FORBIDDEN TO AUTO-APPROVE OR AUTO-MERGE | PROHIBIDO AUTO-APROBAR O AUTO-FUSIONAR**:\n - Never execute PR approvals or direct merges to stable branches (`main`, `release/*`). Approval is an exclusive human prerogative.\n2. **FORBIDDEN TO INVENT PRODUCT OR ARCHITECTURE DECISIONS**:\n - If a requirement is ambiguous or incomplete, formulate open questions (`open-questions`) to the human user. Guessing undocumented intentions is forbidden.\n3. **FORBIDDEN TO INTRODUCE DEPENDENCIES WITHOUT LICENSE INSPECTION**:\n - Before suggesting or adding dependencies to manifests (`package.json`, etc.), validate their SPDX identifier against `license-policy.yaml`. Adding viral (`GPL`, `AGPL`) or commercial paid (`BSL`, `SSPL`) licenses without human authorization is prohibited.\n4. **FORBIDDEN TO IGNORE CYBERSECURITY & PRIVACY (SECURITY-BY-DEFAULT & PRIVACY-BY-DEFAULT)**:\n - All code must enforce least privilege, sanitize external inputs, protect PII according to `privacy-policy.yaml`, and include mitigation tests (`SEC-TEST-*`, `@privacy`). Suppressing types (`any`), linters, or tests to force a green pipeline is prohibited.\n5. **MANDATORY CRYPTOGRAPHIC CITATION**:\n - Every specification, design, or sidecar must reference canonical identifiers and their Unix LF-normalized SHA-256 digests.\n\n---\n\n## 2. Git 4-Tier Branch Hierarchy\n\nWhen proposing or creating branches, respect the strict hierarchical structure:\n- **Tier 1**: `main` (Protected production branch with maximum stability).\n- **Tier 2**: `release/vX.Y.Z` (Release consolidation and scope freeze branch).\n- **Tier 3**: `feat/<FEAT-ID>-<slug>` or `bug/<BUG-ID>-<slug>` (SDD increment or feature branch).\n- **Tier 4**: `task/<PARENT-ID>/<TSK-ID>-<slug>` (Atomic development task branch).\n\n---\n\n## 3. Pre-Flight Commands and Quality Gates\n\nBefore concluding any intervention or suggesting a commit:\n- **Unified pre-flight with auto-fix**: `pnpm run check:fix` (or `npx aisdlc check --fix`).\n- **Complete Quality Gates verification**: `pnpm run verify:all` (or `npx aisdlc verify all`).\n- **Unit test suite**: `pnpm test`.\n- **Strict typing**: `pnpm run typecheck`.\n\nNon-Negotiable Quality Thresholds (`quality-policy.yaml`):\n- Cyclomatic Complexity (CC) $\\le 10$.\n- Cognitive Complexity $\\le 15$.\n- Maintainability Index (MI) $\\ge 50$.\n- Maximum lines per function $\\le 40$.\n\n---\n\n## 4. Task Autonomy Modes (`tasks.md`)\n\n- \uD83D\uDFE2 **`AUTONOMOUS`**: Low risk, isolated task. Implement code and tests directly. Omit interactive handoff block.\n- \uD83D\uDFE1 **`HUMAN_REVIEW_PLAN`**: Medium risk. Design detailed plan, emit Workflow Handoff block, and wait for human confirmation before coding.\n- \uD83D\uDFE0 **`AMBIGUOUS`**: Incomplete requirements. Blocked: request human clarification.\n- \uD83D\uDD34 **`HIGH_RISK_MANUAL`**: Critical risk (migrations, cryptography). Direct manual execution by humans only.\n\n---\n\n## 5. Workflow Handoff Protocol and Human Action Window | Ventana de Acci\u00F3n Humana\n- **Clean Context Window Isolation per Specialist Agent (Cross-IDE)**: Never execute two distinct specialist roles in the same conversation context window. In subagent-capable IDEs (Antigravity `invoke_subagent`, Claude Code `Task` / `.claude/agents/`, GitHub Copilot `@agent-*`), always delegate each specialist phase to a fresh isolated subagent context that hydrates state strictly from disk (`specs/`, `handoff.yaml`, `get_active_handoff`). In chat-based IDEs (Cursor, Windsurf, Aider), persist all artifacts to disk, emit the Workflow Handoff block, and instruct the user to open a **New Chat** (or `/clear`) before running the next specialist.\n- **Conditional Activation**:\n * \uD83D\uDFE2 **`AUTONOMOUS` Mode** (or PR/CI final supervision): **OMITTED**. Do not interrupt unattended execution.\n * \uD83D\uDFE1 **Autonomy $\\ge$ `HUMAN_REVIEW_PLAN`** (`HUMAN_REVIEW_PLAN`, `AMBIGUOUS`, `HIGH_RISK_MANUAL`): **MANDATORY**. Emit Workflow Handoff block compliant with [`templates/workflow/agent-handoff.template.md`](../templates/workflow/agent-handoff.template.md) and **STOP**.\n- **Components**: Declare completed deliverables, recommend next roles (`agent-threat-modeler`, `agent-privacy-officer`, `agent-system-architect`, `agent-developer`, `agent-devops`, `agent-auditor`, `agent-adversary`), provide suggested invocation prompt, and **keep the Human Action Window open at all times for the user to take action** (review, edit by hand, pause/reroute, or delegate).\n\n---\n\n## 6. Specialized Roles Mapping\n\n| Role | Mission (summary) |\n| :--- | :--- |\n| `agent-product-analyst` | Phase 0 NAF v4 Enterprise & Capability Vision (`ARCH-INTRO-*`, up to 15-question discovery interview defining UI modality, delivery model & CRUD prerequisites) + ProductShape artifacts (`ACT-*`, `UC-*`, `FR-*`, `QR-*`, `BR-*`); `status` starts `draft`. |\n| `agent-threat-modeler` | STRIDE/OWASP threat triad (`ACT-THREAT-*`, `ABUSE-*`, `SEC-REQ-*`); hands off to `agent-privacy-officer`. |\n| `agent-privacy-officer` | LINDDUN & GDPR/LOPDGDD privacy triad (`DATA-ASSET-*`, `PRIV-ROPA-*`, `PRIV-REQ-*`) + Pre-Merge `verify:privacy`. |\n| `agent-system-architect` | Modular arc42 + NAF v4 architecture (`CMP-*`, `ARCH-DEP-001`, Mermaid diagrams, `ADR-*`). |\n| `agent-qa-engineer` | FAILING BDD/Gherkin, E2E, and UI suites (using agreed project tech; Happy Path, Sad Path & Boundary Values per interactive UI element) before `agent-developer` writes production code. |\n| `agent-developer` | Atomic tasks from `tasks.md`; autonomy mode must be respected; runs `pnpm run check:fix` + `pnpm run verify:all`. |\n| `agent-expert-user` | Bimodal MVP evaluation (design) and functional validation against `UC-*` and `FR-*` (pre-PR). |\n| `agent-devops` | CI/CD, Docker, IaC only \u2014 clarifies deployment/CI strategy; **FORBIDDEN** to touch `src/` or `packages/*/src/`. |\n| `agent-auditor` | Transversal auditor: YAGNI, Clean Code/SOLID/AST thresholds, UI standards, SPDX licenses (`verify:licenses`); returns handoff to audited agent. |\n| `agent-adversary` | Transversal Red Team adversary: PoC breaks, mutation testing & Pre-Merge security diff audit (`verify:security`, CVSS v3.1); returns handoff to audited agent. |\n\n---\n\n## 7. Canonical Reference\nTo consult full protocols, specialized prompts, and interface contracts:\n- [`process/09_agent_protocols.md`](../process/09_agent_protocols.md)\n- [`process/01_governance_and_roles.md`](../process/01_governance_and_roles.md)\n";
3219
+ declare const STARTER_CLAUDE_RULES = "# Claude Code Instructions - AI-SDLC Monorepo\n\nCanonical Reference: [`process/09_agent_protocols.md`](process/09_agent_protocols.md) and [`process/01_governance_and_roles.md`](process/01_governance_and_roles.md).\n\nThis repository implements the **AI-SDLC** framework (Spec-Driven Development, deterministic quality governance, Privacy-by-Design, and OSS license compliance for human-agent collaboration).\n\n---\n\n## 1. The 5 Unbreakable Commandments of Agents | Los 5 Mandamientos Inquebrantables de los Agentes\n\n1. **FORBIDDEN TO AUTO-APPROVE OR AUTO-MERGE | PROHIBIDO AUTO-APROBAR O AUTO-FUSIONAR**: Never approve PRs or merge directly to `main` or protected branches. Approval is exclusively a human prerogative.\n2. **FORBIDDEN TO INVENT PRODUCT OR ARCHITECTURE DECISIONS**: When facing ambiguous requirements, formulate open questions (`open-questions`). Never assume undocumented behaviors.\n3. **FORBIDDEN TO INTRODUCE DEPENDENCIES WITHOUT LICENSE INSPECTION**: Always validate the SPDX identifier against `license-policy.yaml`. Viral libraries (`GPL`/`AGPL`) or paid commercial libraries (`BSL`/`SSPL`) are strictly forbidden without formal human approval.\n4. **FORBIDDEN TO IGNORE CYBERSECURITY & PRIVACY (SECURITY-BY-DEFAULT & PRIVACY-BY-DEFAULT)**: Validate inputs, sanitize data, protect PII according to `privacy-policy.yaml`, and include mitigation tests (`SEC-TEST-*`, `@privacy`). Disabling linters or suppressing typing errors (`any`, `@ts-ignore`) is prohibited.\n5. **MANDATORY CRYPTOGRAPHIC CITATION**: Every spec or sidecar must cite immutable identifiers and Unix LF-normalized SHA-256 digests.\n\n---\n\n## 2. Essential Execution and Pre-Flight Commands\n\n```bash\n# 1. Pre-flight with deterministic auto-fix (Gherkin & SHA-256 digests)\npnpm run check:fix\n\n# 2. Complete verification across all Quality Gates (Quality, RTM, Governance, Testing, Licenses, PDaC, Schemas, Duplicates, Security, Privacy)\npnpm run verify:all\n\n# 3. Complete automated test suite (Vitest)\npnpm test\n\n# 4. Strict TypeScript typechecking\npnpm run typecheck\n\n# 5. Scaffolding for a new SDD change\npnpm run change:new <name> --id <chg-id>\n\n# 6. Canonical integration of a completed SDD change\nnpx tsx packages/cli/src/index.ts sdd integrate --change <chg-id>\n```\n\n---\n\n## 3. Specialized Roles Mapping (Summary \u2014 Full Prompts in `.claude/agents/`)\n\n| Role | Mission (summary) |\n| :--- | :--- |\n| `agent-product-analyst` | Phase 0 NAF v4 Vision (`ARCH-INTRO-*`, `/grill-me` Socratic decision-tree discovery interview of up to 15 questions \u2014 1\u20132 per turn with `(Recommended)` options \u2014 defining UI modality & CRUD prerequisites) + ProductShape (`ACT-*`, `UC-*`, `FR-*`, `QR-*`, `BR-*`); `status` starts `draft`. |\n| `agent-threat-modeler` | STRIDE/OWASP threat triad (`ACT-THREAT-*`, `ABUSE-*`, `SEC-REQ-*`); hands off to `agent-privacy-officer`. |\n| `agent-privacy-officer` | LINDDUN & GDPR/LOPDGDD privacy triad (`DATA-ASSET-*`, `PRIV-ROPA-*`, `PRIV-REQ-*`) + Pre-Merge `verify:privacy`. |\n| `agent-system-architect` | Modular arc42 + NAF v4 architecture (`CMP-*`, `ARCH-DEP-001`, Mermaid diagrams, `ADR-*`). |\n| `agent-qa-engineer` | FAILING BDD/Gherkin, E2E, and UI suites (Happy Path, Sad Path & Boundary Values per interactive UI element) before `agent-developer` writes code. |\n| `agent-developer` | Atomic tasks from `tasks.md` in 5-step Hexagonal order; respects autonomy mode; runs `pnpm run check:fix` + `pnpm run verify:all`. |\n| `agent-expert-user` | Bimodal MVP evaluation (design) and functional validation against `UC-*` and `FR-*` (pre-PR). |\n| `agent-devops` | CI/CD, Docker, IaC only \u2014 clarifies deployment/CI strategy; **FORBIDDEN** to touch `src/` or `packages/*/src/`. |\n| `agent-auditor` | Transversal auditor: YAGNI, Clean Code/SOLID/AST thresholds, UI standards, SPDX licenses (`verify:licenses`); returns handoff to audited agent. |\n| `agent-adversary` | Transversal Red Team adversary: PoC breaks, mutation testing & Pre-Merge security diff audit (`verify:security`, CVSS v3.1); returns handoff to audited agent. |\n\n---\n\n## 4. Task Autonomy Modes (`tasks.md`)\n\n- \uD83D\uDFE2 **`AUTONOMOUS`**: Low risk, isolated task. Implement code and tests directly. Omit interactive handoff block.\n- \uD83D\uDFE1 **`HUMAN_REVIEW_PLAN`**: Medium risk. Design detailed plan, emit Workflow Handoff block, and wait for human confirmation before coding.\n- \uD83D\uDFE0 **`AMBIGUOUS`**: Incomplete requirements. Blocked: request human clarification.\n- \uD83D\uDD34 **`HIGH_RISK_MANUAL`**: Critical risk (migrations, cryptography). Direct manual execution by humans only.\n\n---\n\n## 5. Git Workflow and Commit Rules with Trailers\n\n### 4-Tier Hierarchy\n1. **Tier 1 (`main`)**: Absolute stability and production readiness.\n2. **Tier 2 (`release/vX.Y.Z`)**: Release stabilization.\n3. **Tier 3 (`feat/<FEAT-ID>-<slug>` or `bug/<BUG-ID>-<slug>`)**: SDD delivery increment.\n4. **Tier 4 (`task/<PARENT-ID>/<TSK-ID>-<slug>`)**: Atomic implementation task.\n\n### Commit Format (Conventional Commits + Git Trailers)\n```text\nfeat(scope): concise imperative description (#issue)\n\nBody explaining the motivation and technical justification of the change.\n\nAuthor-Type: agent\nAI-Model: claude-3-7-sonnet\nAI-Reasoning-Effort: high\nTask-ID: TSK-001\nChange-ID: CHG-026-AGENT-NATIVE-CONFIGS\n```\n\n---\n\n## 6. Non-Negotiable Quality Thresholds (`quality-policy.yaml`)\n- **Cyclomatic Complexity (CC)**: $\\le 10$\n- **Cognitive Complexity**: $\\le 15$\n- **Maintainability Index (MI)**: $\\ge 50$\n- **Lines per Function**: $\\le 40$\n- Deterministic CLI output: every `verify` command supports `--json` without ANSI escape codes.\n\n---\n\n## 7. Workflow Handoff Protocol and Human Action Window | Ventana de Acci\u00F3n Humana\n- **Clean Context Window Isolation per Specialist Agent (Cross-IDE)**: Never execute two distinct specialist roles in the same conversation context window. In subagent-capable IDEs (Antigravity `invoke_subagent`, Claude Code `Task` / `.claude/agents/`, GitHub Copilot `@agent-*`), always delegate each specialist phase to a fresh isolated subagent context that hydrates state strictly from disk (`specs/`, `handoff.yaml`, `get_active_handoff`). In chat-based IDEs (Cursor, Windsurf, Aider), persist all artifacts to disk, emit the Workflow Handoff block, and instruct the user to open a **New Chat** (or `/clear`) before running the next specialist.\n- **Conditional Activation**:\n - \uD83D\uDFE2 **`AUTONOMOUS`** (or PR/CI final supervision): **OMITTED**. Uninterrupted continuous execution.\n - \uD83D\uDFE1 **Autonomy $\\ge$ `HUMAN_REVIEW_PLAN`** (`HUMAN_REVIEW_PLAN`, `AMBIGUOUS`, `HIGH_RISK_MANUAL`): **MANDATORY**. Emit Workflow Handoff block ([`templates/workflow/agent-handoff.template.md`](templates/workflow/agent-handoff.template.md)) and **STOP**.\n- **Content**: Completed deliverables, suggested next role(s), copy-paste ready invocation prompt, and **an open Human Action Window for the user to take action** (review, edit by hand, pause/reroute, or delegate).\n";
3220
+ declare const STARTER_COPILOT_RULES = "<!--\n AI-SDLC: Canonical Instructions for GitHub Copilot (Chat & Workspace)\n Canonical Reference: process/09_agent_protocols.md and process/01_governance_and_roles.md\n-->\n\n# GitHub Copilot Instructions - AI-SDLC Framework\n\nYou operate as an assistant agent in the **AI-SDLC** repository. You are not a simple text auto-completer: you are a specialized technical worker subject to normative directives, deterministic autonomy boundaries, and strict traceability.\n\n---\n\n## 1. The 5 Unbreakable Commandments of Agents | Los 5 Mandamientos Inquebrantables de los Agentes\n\n1. **FORBIDDEN TO AUTO-APPROVE OR AUTO-MERGE | PROHIBIDO AUTO-APROBAR O AUTO-FUSIONAR**:\n - Never execute PR approvals or direct merges to stable branches (`main`, `release/*`). Approval is an exclusive human prerogative.\n2. **FORBIDDEN TO INVENT PRODUCT OR ARCHITECTURE DECISIONS**:\n - If a requirement is ambiguous or incomplete, formulate open questions (`open-questions`) to the human user. Guessing undocumented intentions is forbidden.\n3. **FORBIDDEN TO INTRODUCE DEPENDENCIES WITHOUT LICENSE INSPECTION**:\n - Before suggesting or adding dependencies to manifests (`package.json`, etc.), validate their SPDX identifier against `license-policy.yaml`. Adding viral (`GPL`, `AGPL`) or commercial paid (`BSL`, `SSPL`) licenses without human authorization is prohibited.\n4. **FORBIDDEN TO IGNORE CYBERSECURITY & PRIVACY (SECURITY-BY-DEFAULT & PRIVACY-BY-DEFAULT)**:\n - All code must enforce least privilege, sanitize external inputs, protect PII according to `privacy-policy.yaml`, and include mitigation tests (`SEC-TEST-*`, `@privacy`). Suppressing types (`any`), linters, or tests to force a green pipeline is prohibited.\n5. **MANDATORY CRYPTOGRAPHIC CITATION**:\n - Every specification, design, or sidecar must reference canonical identifiers and their Unix LF-normalized SHA-256 digests.\n\n---\n\n## 2. Git 4-Tier Branch Hierarchy\n\nWhen proposing or creating branches, respect the strict hierarchical structure:\n- **Tier 1**: `main` (Protected production branch with maximum stability).\n- **Tier 2**: `release/vX.Y.Z` (Release consolidation and scope freeze branch).\n- **Tier 3**: `feat/<FEAT-ID>-<slug>` or `bug/<BUG-ID>-<slug>` (SDD increment or feature branch).\n- **Tier 4**: `task/<PARENT-ID>/<TSK-ID>-<slug>` (Atomic development task branch).\n\n---\n\n## 3. Pre-Flight Commands and Quality Gates\n\nBefore concluding any intervention or suggesting a commit:\n- **Unified pre-flight with auto-fix**: `pnpm run check:fix` (or `npx aisdlc check --fix`).\n- **Complete Quality Gates verification**: `pnpm run verify:all` (or `npx aisdlc verify all`).\n- **Unit test suite**: `pnpm test`.\n- **Strict typing**: `pnpm run typecheck`.\n\nNon-Negotiable Quality Thresholds (`quality-policy.yaml`):\n- Cyclomatic Complexity (CC) $\\le 10$.\n- Cognitive Complexity $\\le 15$.\n- Maintainability Index (MI) $\\ge 50$.\n- Maximum lines per function $\\le 40$.\n\n---\n\n## 4. Task Autonomy Modes (`tasks.md`)\n\n- \uD83D\uDFE2 **`AUTONOMOUS`**: Low risk, isolated task. Implement code and tests directly. Omit interactive handoff block.\n- \uD83D\uDFE1 **`HUMAN_REVIEW_PLAN`**: Medium risk. Design detailed plan, emit Workflow Handoff block, and wait for human confirmation before coding.\n- \uD83D\uDFE0 **`AMBIGUOUS`**: Incomplete requirements. Blocked: request human clarification.\n- \uD83D\uDD34 **`HIGH_RISK_MANUAL`**: Critical risk (migrations, cryptography). Direct manual execution by humans only.\n\n---\n\n## 5. Workflow Handoff Protocol and Human Action Window | Ventana de Acci\u00F3n Humana\n- **Clean Context Window Isolation per Specialist Agent (Cross-IDE)**: Never execute two distinct specialist roles in the same conversation context window. In subagent-capable IDEs (Antigravity `invoke_subagent`, Claude Code `Task` / `.claude/agents/`, GitHub Copilot `@agent-*`), always delegate each specialist phase to a fresh isolated subagent context that hydrates state strictly from disk (`specs/`, `handoff.yaml`, `get_active_handoff`). In chat-based IDEs (Cursor, Windsurf, Aider), persist all artifacts to disk, emit the Workflow Handoff block, and instruct the user to open a **New Chat** (or `/clear`) before running the next specialist.\n- **Conditional Activation**:\n * \uD83D\uDFE2 **`AUTONOMOUS` Mode** (or PR/CI final supervision): **OMITTED**. Do not interrupt unattended execution.\n * \uD83D\uDFE1 **Autonomy $\\ge$ `HUMAN_REVIEW_PLAN`** (`HUMAN_REVIEW_PLAN`, `AMBIGUOUS`, `HIGH_RISK_MANUAL`): **MANDATORY**. Emit Workflow Handoff block compliant with [`templates/workflow/agent-handoff.template.md`](../templates/workflow/agent-handoff.template.md) and **STOP**.\n- **Components**: Declare completed deliverables, recommend next roles (`agent-threat-modeler`, `agent-privacy-officer`, `agent-system-architect`, `agent-developer`, `agent-devops`, `agent-auditor`, `agent-adversary`), provide suggested invocation prompt, and **keep the Human Action Window open at all times for the user to take action** (review, edit by hand, pause/reroute, or delegate).\n\n---\n\n## 6. Specialized Roles Mapping\n\n| Role | Mission (summary) |\n| :--- | :--- |\n| `agent-product-analyst` | Phase 0 NAF v4 Enterprise & Capability Vision (`ARCH-INTRO-*`, `/grill-me` Socratic decision-tree discovery interview of up to 15 questions \u2014 1\u20132 per turn with `(Recommended)` options \u2014 defining UI modality, delivery model & CRUD prerequisites) + ProductShape artifacts (`ACT-*`, `UC-*`, `FR-*`, `QR-*`, `BR-*`); `status` starts `draft`. |\n| `agent-threat-modeler` | STRIDE/OWASP threat triad (`ACT-THREAT-*`, `ABUSE-*`, `SEC-REQ-*`); hands off to `agent-privacy-officer`. |\n| `agent-privacy-officer` | LINDDUN & GDPR/LOPDGDD privacy triad (`DATA-ASSET-*`, `PRIV-ROPA-*`, `PRIV-REQ-*`) + Pre-Merge `verify:privacy`. |\n| `agent-system-architect` | Modular arc42 + NAF v4 architecture (`CMP-*`, `ARCH-DEP-001`, Mermaid diagrams, `ADR-*`). |\n| `agent-qa-engineer` | FAILING BDD/Gherkin, E2E, and UI suites (using agreed project tech; Happy Path, Sad Path & Boundary Values per interactive UI element) before `agent-developer` writes production code. |\n| `agent-developer` | Atomic tasks from `tasks.md`; autonomy mode must be respected; runs `pnpm run check:fix` + `pnpm run verify:all`. |\n| `agent-expert-user` | Bimodal MVP evaluation (design) and functional validation against `UC-*` and `FR-*` (pre-PR). |\n| `agent-devops` | CI/CD, Docker, IaC only \u2014 clarifies deployment/CI strategy; **FORBIDDEN** to touch `src/` or `packages/*/src/`. |\n| `agent-auditor` | Transversal auditor: YAGNI, Clean Code/SOLID/AST thresholds, UI standards, SPDX licenses (`verify:licenses`); returns handoff to audited agent. |\n| `agent-adversary` | Transversal Red Team adversary: PoC breaks, mutation testing & Pre-Merge security diff audit (`verify:security`, CVSS v3.1); returns handoff to audited agent. |\n\n---\n\n## 7. Canonical Reference\nTo consult full protocols, specialized prompts, and interface contracts:\n- [`process/09_agent_protocols.md`](../process/09_agent_protocols.md)\n- [`process/01_governance_and_roles.md`](../process/01_governance_and_roles.md)\n";
3221
3221
  declare const STARTER_CURSOR_MCP = "{\n \"mcpServers\": {\n \"ai-sdlc\": {\n \"command\": \"npx\",\n \"args\": [\n \"@aisdlc/mcp\"\n ],\n \"env\": {\n \"AISDLC_MCP_ROLE\": \"developer\"\n }\n }\n }\n}\n";
3222
3222
  declare const STARTER_ANTIGRAVITY_MCP = "{\n \"mcpServers\": {\n \"ai-sdlc\": {\n \"command\": \"npx\",\n \"args\": [\n \"@aisdlc/mcp\"\n ],\n \"env\": {\n \"AISDLC_MCP_ROLE\": \"developer\"\n }\n }\n }\n}\n";
3223
3223
  declare const STARTER_VSCODE_MCP = "{\n \"servers\": {\n \"ai-sdlc\": {\n \"command\": \"npx\",\n \"args\": [\n \"@aisdlc/mcp\"\n ],\n \"env\": {\n \"AISDLC_MCP_ROLE\": \"developer\"\n }\n }\n }\n}\n";
package/dist/index.d.ts CHANGED
@@ -3212,12 +3212,12 @@ declare function renderAntigravitySkillFile(skill: LifecycleSkillTemplate): stri
3212
3212
  declare function renderCursorSkillFile(skill: LifecycleSkillTemplate): string;
3213
3213
  declare function renderClaudeSkillFile(skill: LifecycleSkillTemplate): string;
3214
3214
  declare function renderCopilotSkillFile(skill: LifecycleSkillTemplate): string;
3215
- declare const STARTER_ANTIGRAVITY_RULES = "# Google Antigravity & Gemini CLI Rules - AI-SDLC Framework\n\nCanonical Reference: [`process/09_agent_protocols.md`](../../process/09_agent_protocols.md) and [`process/01_governance_and_roles.md`](../../process/01_governance_and_roles.md).\n\nThis file defines the operational rules and specialized role mappings for Google Antigravity and Gemini CLI-based agents within the AI-SDLC ecosystem.\n\n---\n\n## 1. The 5 Unbreakable Commandments of Agents | Los 5 Mandamientos Inquebrantables de los Agentes\n\n1. **FORBIDDEN TO AUTO-APPROVE OR AUTO-MERGE | PROHIBIDO AUTO-APROBAR O AUTO-FUSIONAR**: No agent may auto-approve PRs or perform direct merges to protected branches (`main`, `release/*`). Only a human engineer may approve or merge.\n2. **FORBIDDEN TO INVENT PRODUCT OR ARCHITECTURE DECISIONS**: When facing ambiguous requirements, formulate clarifying questions (`open-questions`). Never assume unspecified intentions.\n3. **FORBIDDEN TO INTRODUCE DEPENDENCIES WITHOUT LICENSE INSPECTION**: Always consult `license-policy.yaml`. Dependencies with viral licenses (`GPL`/`AGPL`) or commercial paid licenses (`BSL`/`SSPL`) are strictly forbidden without authorization.\n4. **FORBIDDEN TO IGNORE CYBERSECURITY & PRIVACY (SECURITY-BY-DEFAULT & PRIVACY-BY-DEFAULT)**: Validate inputs, sanitize data, protect PII according to `privacy-policy.yaml`, and add mitigation tests (`SEC-TEST-*`, `@privacy`). Disabling linters or suppressing typing errors is prohibited.\n5. **MANDATORY CRYPTOGRAPHIC CITATION**: Every derived artifact must cite upstream artifacts using canonical identifiers and their Unix LF-normalized SHA-256 cryptographic hashes.\n\n---\n\n## 2. Specialized Roles Mapping (Summary \u2014 Full Prompts in `.agent/agents/`)\n\n| Role | Mission (summary) |\n| :--- | :--- |\n| `agent-product-analyst` | Phase 0 NAF v4 Vision (`ARCH-INTRO-*`, up to 15-question discovery defining UI modality & CRUD prerequisites) + ProductShape (`ACT-*`, `UC-*`, `FR-*`, `QR-*`, `BR-*`); `status` starts `draft`. |\n| `agent-threat-modeler` | STRIDE/OWASP threat triad (`ACT-THREAT-*`, `ABUSE-*`, `SEC-REQ-*`); hands off to `agent-privacy-officer`. |\n| `agent-privacy-officer` | LINDDUN & GDPR/LOPDGDD privacy triad (`DATA-ASSET-*`, `PRIV-ROPA-*`, `PRIV-REQ-*`) + Pre-Merge `verify:privacy`. |\n| `agent-system-architect` | Modular arc42 + NAF v4 architecture (`CMP-*`, `ARCH-DEP-001`, Mermaid diagrams, `ADR-*`). |\n| `agent-qa-engineer` | FAILING BDD/Gherkin, E2E, and UI suites (Happy Path, Sad Path & Boundary Values per interactive UI element) before `agent-developer` writes code. |\n| `agent-developer` | Atomic tasks from `tasks.md` in 5-step Hexagonal order; respects autonomy mode; runs `pnpm run check:fix` + `pnpm run verify:all`. |\n| `agent-expert-user` | Bimodal MVP evaluation (design) and functional validation against `UC-*` and `FR-*` (pre-PR). |\n| `agent-devops` | CI/CD, Docker, IaC only \u2014 clarifies deployment/CI strategy; **FORBIDDEN** to touch `src/` or `packages/*/src/`. |\n| `agent-auditor` | Transversal auditor: YAGNI, Clean Code/SOLID/AST thresholds, UI standards, SPDX licenses (`verify:licenses`); returns handoff to audited agent. |\n| `agent-adversary` | Transversal Red Team adversary: PoC breaks, mutation testing & Pre-Merge security diff audit (`verify:security`, CVSS v3.1); returns handoff to audited agent. |\n\n---\n\n## 3. Pre-Flight Verification Commands\n- Pre-flight with auto-fix: `pnpm run check:fix` (or `npx aisdlc check --fix`)\n- Full verification: `pnpm run verify:all` (or `npx aisdlc verify all`)\n- Unit tests: `pnpm test` | Strict typing: `pnpm run typecheck`\n\n---\n\n## 4. Git 4-Tier Hierarchy and Commit Conventions\n- `task/<PARENT-ID>/<TSK-ID>-<slug>` $\\rightarrow$ `feat/<FEAT-ID>-<slug>` $\\rightarrow$ `release/vX.Y.Z` $\\rightarrow$ `main`.\n- Mandatory commit trailers injection (always declare your actual active model in `AI-Model:` or via `AI_MODEL=<active-model>` when running `git commit`, so the hook does not fall back to a generic IDE identifier like `antigravity-agent`, and include `AI-Reasoning-Effort: <low|medium|high|max>` or `AI_REASONING_EFFORT=<level>` when applicable):\n ```text\n Author-Type: agent\n AI-Model: <active-model>\n AI-Reasoning-Effort: <low|medium|high|max>\n Task-ID: TSK-XXX\n Change-ID: CHG-XXX\n ```\n\n---\n\n## 4. Workflow Handoff Protocol and Human Action Window | Ventana de Acci\u00F3n Humana\n- **Clean Context Window Isolation per Specialist Agent (Cross-IDE)**: Never execute two distinct specialist roles in the same conversation context window. In subagent-capable IDEs (Antigravity `invoke_subagent`, Claude Code `Task` / `.claude/agents/`, GitHub Copilot `@agent-*`), always delegate each specialist phase to a fresh isolated subagent context that hydrates state strictly from disk (`specs/`, `handoff.yaml`, `get_active_handoff`). In chat-based IDEs (Cursor, Windsurf, Aider), persist all artifacts to disk, emit the Workflow Handoff block, and instruct the user to open a **New Chat** (or `/clear`) before running the next specialist.\n- **Conditional Activation Rule by Autonomy**:\n - \uD83D\uDFE2 **`AUTONOMOUS`** (or exclusive final supervision in PR/CI): **OMITTED**. Do not request or emit interactive handoff to avoid interrupting unattended execution.\n - \uD83D\uDFE1 **Autonomy $\\ge$ `HUMAN_REVIEW_PLAN`** (`HUMAN_REVIEW_PLAN`, `AMBIGUOUS`, `HIGH_RISK_MANUAL`): **MANDATORY**. Emit the Workflow Handoff block compliant with [`templates/workflow/agent-handoff.template.md`](../../templates/workflow/agent-handoff.template.md) and **STOP**.\n- **Block Contents**: Declare produced deliverables, recommend the next role in workflow (`agent-threat-modeler`, `agent-privacy-officer`, `agent-system-architect`, `agent-qa-engineer`, `agent-developer`, `agent-expert-user`, `agent-devops`, `agent-auditor`, `agent-adversary`, etc.), provide suggested invocation prompt, and keep **the Human Action Window open at all times for the human user to take action** (review, edit by hand, pause/reroute, or delegate).\n";
3216
- declare const STARTER_CURSOR_CORE_RULES = "---\ndescription: Global canonical directives, 5 unbreakable commandments, and AI-SDLC pre-flight\nglobs: *\nalwaysApply: true\n---\n\n# Global AI-SDLC Rules (Core)\n\nCanonical Reference: [`process/09_agent_protocols.md`](process/09_agent_protocols.md) and [`process/01_governance_and_roles.md`](process/01_governance_and_roles.md).\n\n## 1. The 5 Unbreakable Commandments of Agents | Los 5 Mandamientos Inquebrantables de los Agentes\nEvery agent operating in this repository must strictly observe the following guardrails:\n1. **FORBIDDEN TO AUTO-APPROVE OR AUTO-MERGE | PROHIBIDO AUTO-APROBAR O AUTO-FUSIONAR**: Do not approve PRs or execute direct merges to protected branches (`main`, `release/*`). Approval is exclusively reserved for human reviewers.\n2. **FORBIDDEN TO INVENT PRODUCT OR ARCHITECTURE DECISIONS**: When facing ambiguity, formulate open questions (`open-questions`). Never assume undocumented requirements.\n3. **FORBIDDEN TO INTRODUCE DEPENDENCIES WITHOUT LICENSE INSPECTION**: Always verify `license-policy.yaml`. Viral dependencies (GPL/AGPL) or commercial paid dependencies (BSL/SSPL) are strictly forbidden without explicit human approval.\n4. **FORBIDDEN TO IGNORE CYBERSECURITY & PRIVACY (SECURITY-BY-DEFAULT & PRIVACY-BY-DEFAULT)**: Validate inputs, apply least privilege, protect PII according to `privacy-policy.yaml`, and accompany every change with security and privacy tests (`SEC-TEST-*`, `@privacy`). Suppressing linters or tests to force a green pipeline is prohibited.\n5. **MANDATORY CRYPTOGRAPHIC CITATION**: Link specifications using immutable identifiers and Unix LF-normalized SHA-256 hashes.\n\n## 2. Pre-Flight Verification Commands\nBefore finalizing changes or proposing commits:\n- Pre-flight with auto-fix: `pnpm run check:fix` (or `npx aisdlc check --fix`)\n- Full verification: `pnpm run verify:all` (or `npx aisdlc verify all`)\n- Unit tests: `pnpm test`\n- Strict typing: `pnpm run typecheck`\n\n## 3. Git 4-Tier Hierarchy\n- Tier 1: `main` (Production)\n- Tier 2: `release/vX.Y.Z` (Frozen release)\n- Tier 3: `feat/<FEAT-ID>-<slug>` / `bug/<BUG-ID>-<slug>` (Feature)\n- Tier 4: `task/<PARENT-ID>/<TSK-ID>-<slug>` (Atomic task)\n\n## 4. Specialized Roles Mapping\n| Role | Mission (summary) |\n| :--- | :--- |\n| `agent-product-analyst` | Phase 0 NAF v4 Enterprise & Capability Vision (`ARCH-INTRO-*`, up to 15-question discovery interview defining UI modality & CRUD prerequisites) + ProductShape artifacts (`ACT-*`, `UC-*`, `FR-*`, `QR-*`, `BR-*`); status starts `draft`. |\n| `agent-threat-modeler` | STRIDE/OWASP threat triad (`ACT-THREAT-*`, `ABUSE-*`, `SEC-REQ-*`); hands off to `agent-privacy-officer`. |\n| `agent-privacy-officer` | LINDDUN & GDPR/LOPDGDD privacy triad (`DATA-ASSET-*`, `PRIV-ROPA-*`, `PRIV-REQ-*`) + Pre-Merge `verify:privacy`. |\n| `agent-system-architect` | Modular arc42 + NAF v4 architecture (`CMP-*`, Mermaid diagrams, `ADR-*`). |\n| `agent-qa-engineer` | FAILING BDD/Gherkin, E2E, and UI suites (using agreed project tech; Happy Path, Sad Path & Boundary Values per interactive UI element) before `agent-developer` writes code. |\n| `agent-developer` | Atomic tasks from `tasks.md`; respects autonomy mode; runs `pnpm run check:fix` + `pnpm run verify:all`. |\n| `agent-expert-user` | Bimodal MVP evaluation (design) and functional validation against `UC-*` and `FR-*` (pre-PR). |\n| `agent-devops` | CI/CD, Docker, IaC only \u2014 clarifies deployment/CI strategy; **FORBIDDEN** to touch `src/` or `packages/*/src/`. |\n| `agent-auditor` | Transversal auditor: YAGNI, Clean Code/SOLID/AST thresholds, UI standards, SPDX licenses (`verify:licenses`); returns handoff to audited agent. |\n| `agent-adversary` | Transversal Red Team adversary: PoC breaks, mutation testing & Pre-Merge security diff audit (`verify:security`, CVSS v3.1); returns handoff to audited agent. |\n\n## 5. Task Autonomy Modes (`tasks.md`)\n- \uD83D\uDFE2 **`AUTONOMOUS`**: Low risk. Implement directly, omit handoff block.\n- \uD83D\uDFE1 **`HUMAN_REVIEW_PLAN`**: Medium risk. Emit Workflow Handoff block, wait for approval.\n- \uD83D\uDFE0 **`AMBIGUOUS`**: Blocked. Request human clarification.\n- \uD83D\uDD34 **`HIGH_RISK_MANUAL`**: Critical risk. Human-only execution.\n\n## 6. Workflow Handoff Protocol and Human Action Window | Ventana de Acci\u00F3n Humana\n- **Clean Context Window Isolation per Specialist Agent (Cross-IDE)**: Never execute two distinct specialist roles in the same conversation context window. In subagent-capable IDEs (Antigravity `invoke_subagent`, Claude Code `Task` / `.claude/agents/`, GitHub Copilot `@agent-*`), always delegate each specialist phase to a fresh isolated subagent context that hydrates state strictly from disk (`specs/`, `handoff.yaml`, `get_active_handoff`). In chat-based IDEs (Cursor, Windsurf, Aider), persist all artifacts to disk, emit the Workflow Handoff block, and instruct the user to open a **New Chat** (or `/clear`) before running the next specialist.\n- **Conditional Activation**:\n * \uD83D\uDFE2 **`AUTONOMOUS`** (or PR/CI supervision): **OMITTED**. Do not request or emit interactive handoff.\n * \uD83D\uDFE1 **Autonomy $\\ge$ `HUMAN_REVIEW_PLAN`** (`HUMAN_REVIEW_PLAN`, `AMBIGUOUS`, `HIGH_RISK_MANUAL`): **MANDATORY**. Emit Workflow Handoff block compliant with [`templates/workflow/agent-handoff.template.md`](templates/workflow/agent-handoff.template.md) and **STOP**.\n- **Content**: Declare completed deliverables, recommend next role(s), provide suggested invocation prompt, and **keep the Human Action Window open for the human user to take action** (review, edit by hand, pause/reroute, or delegate).\n";
3217
- declare const STARTER_CURSOR_PRODUCT_RULES = "---\ndescription: Directives for SDD specifications editing, ProductShape artifacts, Cybersecurity, and Privacy\nglobs: specs/**, templates/**, product/**, security/**, privacy/**\nalwaysApply: false\n---\n\n# Product and SDD Specification Directives\n\nCanonical Reference: [`process/09_agent_protocols.md`](process/09_agent_protocols.md) and [`process/02_product_definition.md`](process/02_product_definition.md).\n\n## 1. Identifier Taxonomy and JSON Schemas\nEvery generated or modified artifact must follow canonical naming and satisfy its schema in `schemas/`:\n- Enterprise & Capability Vision (Phase 0): `ARCH-INTRO-[NUM3]` (`schemas/architecture/introduction-and-goals.schema.json`)\n- Product Actors: `ACT-[SUFFIX]` (`schemas/product/actor.schema.json`)\n- Use Cases: `UC-[SUFFIX]` (`schemas/product/use-case.schema.json`)\n- Functional Requirements: `FR-[SUFFIX]-[NUM3]` (`schemas/product/requirement.schema.json`)\n- Quality Requirements: `QR-[SUFFIX]` (`schemas/product/requirement.schema.json`)\n- Business Rules: `BR-[SUFFIX]` (`schemas/product/business-rule.schema.json`)\n- Threat Actors: `ACT-THREAT-[SUFFIX]` (`schemas/security/threat-actor.schema.json`)\n- Abuse Cases: `ABUSE-[SUFFIX]` (`schemas/security/abuse-case.schema.json`)\n- Security Requirements: `SEC-REQ-[SUFFIX]` (`schemas/security/security-req.schema.json`)\n- Data Assets: `DATA-ASSET-[SUFFIX]` (`schemas/privacy/data-asset.schema.json`)\n- Processing Activities (ROPA): `PRIV-ROPA-[SUFFIX]` (`schemas/privacy/privacy-ropa.schema.json`)\n- Privacy Requirements: `PRIV-REQ-[SUFFIX]` (`schemas/privacy/privacy-req.schema.json`)\n\n## 2. \"AI as Scribe\" Protocol Rules\n- **Phase 0 (New Product Inception)**: When starting a new product, conduct an iterative discovery interview of up to 15 questions (or until the product scope and target UI modality \u2014 Web, Desktop, Console/CLI, Mobile, or API/SDK \u2014 are completely clear), instantiate `ARCH-INTRO-001` (`naf-perspective: \"Enterprise & Capability\"`) in `draft`, and derive `ACT-*`, `UC-*`, `BR-*`, `FR-*`, and `QR-*` (including runnable UI/CLI requirements and master-entity CRUD prerequisites when human actors exist).\n- **Initial Status**: Every new artifact strictly starts with `status: draft`. Creating artifacts directly as `active` or `approved` is forbidden.\n- **Gherkin Acceptance Criteria**: In every `FR-*` requirement, include a ```gherkin block with tags `@<ID> @automated @regression` and at least one `Scenario Outline` with an `Examples` data table.\n- **Security & Privacy Mitigations**: Every `SEC-REQ-*` must include negative Gherkin tests (`@security @mitigation`) and every `PRIV-REQ-*` must include Gherkin privacy scenarios (`@privacy @gdpr`).\n- **Use Cases (User Story Structure)**: In every `UC-*`, the `1. Intent and Outcome` section must be written using the standard English structure: `As a <ACT-ID>... I want <action>... To <outcome>...`.\n- **Deterministic Validation**: Run `pnpm run verify:schemas`, `pnpm run verify:duplicates`, and `pnpm run verify:privacy` before presenting the draft to the human reviewer.\n";
3215
+ declare const STARTER_ANTIGRAVITY_RULES = "# Google Antigravity & Gemini CLI Rules - AI-SDLC Framework\n\nCanonical Reference: [`process/09_agent_protocols.md`](../../process/09_agent_protocols.md) and [`process/01_governance_and_roles.md`](../../process/01_governance_and_roles.md).\n\nThis file defines the operational rules and specialized role mappings for Google Antigravity and Gemini CLI-based agents within the AI-SDLC ecosystem.\n\n---\n\n## 1. The 5 Unbreakable Commandments of Agents | Los 5 Mandamientos Inquebrantables de los Agentes\n\n1. **FORBIDDEN TO AUTO-APPROVE OR AUTO-MERGE | PROHIBIDO AUTO-APROBAR O AUTO-FUSIONAR**: No agent may auto-approve PRs or perform direct merges to protected branches (`main`, `release/*`). Only a human engineer may approve or merge.\n2. **FORBIDDEN TO INVENT PRODUCT OR ARCHITECTURE DECISIONS**: When facing ambiguous requirements, formulate clarifying questions (`open-questions`). Never assume unspecified intentions.\n3. **FORBIDDEN TO INTRODUCE DEPENDENCIES WITHOUT LICENSE INSPECTION**: Always consult `license-policy.yaml`. Dependencies with viral licenses (`GPL`/`AGPL`) or commercial paid licenses (`BSL`/`SSPL`) are strictly forbidden without authorization.\n4. **FORBIDDEN TO IGNORE CYBERSECURITY & PRIVACY (SECURITY-BY-DEFAULT & PRIVACY-BY-DEFAULT)**: Validate inputs, sanitize data, protect PII according to `privacy-policy.yaml`, and add mitigation tests (`SEC-TEST-*`, `@privacy`). Disabling linters or suppressing typing errors is prohibited.\n5. **MANDATORY CRYPTOGRAPHIC CITATION**: Every derived artifact must cite upstream artifacts using canonical identifiers and their Unix LF-normalized SHA-256 cryptographic hashes.\n\n---\n\n## 2. Specialized Roles Mapping (Summary \u2014 Full Prompts in `.agent/agents/`)\n\n| Role | Mission (summary) |\n| :--- | :--- |\n| `agent-product-analyst` | Phase 0 NAF v4 Vision (`ARCH-INTRO-*`, `/grill-me` Socratic decision-tree discovery interview of up to 15 questions \u2014 1\u20132 per turn with `(Recommended)` options \u2014 defining UI modality & CRUD prerequisites) + ProductShape (`ACT-*`, `UC-*`, `FR-*`, `QR-*`, `BR-*`); `status` starts `draft`. |\n| `agent-threat-modeler` | STRIDE/OWASP threat triad (`ACT-THREAT-*`, `ABUSE-*`, `SEC-REQ-*`); hands off to `agent-privacy-officer`. |\n| `agent-privacy-officer` | LINDDUN & GDPR/LOPDGDD privacy triad (`DATA-ASSET-*`, `PRIV-ROPA-*`, `PRIV-REQ-*`) + Pre-Merge `verify:privacy`. |\n| `agent-system-architect` | Modular arc42 + NAF v4 architecture (`CMP-*`, `ARCH-DEP-001`, Mermaid diagrams, `ADR-*`). |\n| `agent-qa-engineer` | FAILING BDD/Gherkin, E2E, and UI suites (Happy Path, Sad Path & Boundary Values per interactive UI element) before `agent-developer` writes code. |\n| `agent-developer` | Atomic tasks from `tasks.md` in 5-step Hexagonal order; respects autonomy mode; runs `pnpm run check:fix` + `pnpm run verify:all`. |\n| `agent-expert-user` | Bimodal MVP evaluation (design) and functional validation against `UC-*` and `FR-*` (pre-PR). |\n| `agent-devops` | CI/CD, Docker, IaC only \u2014 clarifies deployment/CI strategy; **FORBIDDEN** to touch `src/` or `packages/*/src/`. |\n| `agent-auditor` | Transversal auditor: YAGNI, Clean Code/SOLID/AST thresholds, UI standards, SPDX licenses (`verify:licenses`); returns handoff to audited agent. |\n| `agent-adversary` | Transversal Red Team adversary: PoC breaks, mutation testing & Pre-Merge security diff audit (`verify:security`, CVSS v3.1); returns handoff to audited agent. |\n\n---\n\n## 3. Pre-Flight Verification Commands\n- Pre-flight with auto-fix: `pnpm run check:fix` (or `npx aisdlc check --fix`)\n- Full verification: `pnpm run verify:all` (or `npx aisdlc verify all`)\n- Unit tests: `pnpm test` | Strict typing: `pnpm run typecheck`\n\n---\n\n## 4. Git 4-Tier Hierarchy and Commit Conventions\n- `task/<PARENT-ID>/<TSK-ID>-<slug>` $\\rightarrow$ `feat/<FEAT-ID>-<slug>` $\\rightarrow$ `release/vX.Y.Z` $\\rightarrow$ `main`.\n- Mandatory commit trailers injection (always declare your actual active model in `AI-Model:` or via `AI_MODEL=<active-model>` when running `git commit`, so the hook does not fall back to a generic IDE identifier like `antigravity-agent`, and include `AI-Reasoning-Effort: <low|medium|high|max>` or `AI_REASONING_EFFORT=<level>` when applicable):\n ```text\n Author-Type: agent\n AI-Model: <active-model>\n AI-Reasoning-Effort: <low|medium|high|max>\n Task-ID: TSK-XXX\n Change-ID: CHG-XXX\n ```\n\n---\n\n## 4. Workflow Handoff Protocol and Human Action Window | Ventana de Acci\u00F3n Humana\n- **Clean Context Window Isolation per Specialist Agent (Cross-IDE)**: Never execute two distinct specialist roles in the same conversation context window. In subagent-capable IDEs (Antigravity `invoke_subagent`, Claude Code `Task` / `.claude/agents/`, GitHub Copilot `@agent-*`), always delegate each specialist phase to a fresh isolated subagent context that hydrates state strictly from disk (`specs/`, `handoff.yaml`, `get_active_handoff`). In chat-based IDEs (Cursor, Windsurf, Aider), persist all artifacts to disk, emit the Workflow Handoff block, and instruct the user to open a **New Chat** (or `/clear`) before running the next specialist.\n- **Conditional Activation Rule by Autonomy**:\n - \uD83D\uDFE2 **`AUTONOMOUS`** (or exclusive final supervision in PR/CI): **OMITTED**. Do not request or emit interactive handoff to avoid interrupting unattended execution.\n - \uD83D\uDFE1 **Autonomy $\\ge$ `HUMAN_REVIEW_PLAN`** (`HUMAN_REVIEW_PLAN`, `AMBIGUOUS`, `HIGH_RISK_MANUAL`): **MANDATORY**. Emit the Workflow Handoff block compliant with [`templates/workflow/agent-handoff.template.md`](../../templates/workflow/agent-handoff.template.md) and **STOP**.\n- **Block Contents**: Declare produced deliverables, recommend the next role in workflow (`agent-threat-modeler`, `agent-privacy-officer`, `agent-system-architect`, `agent-qa-engineer`, `agent-developer`, `agent-expert-user`, `agent-devops`, `agent-auditor`, `agent-adversary`, etc.), provide suggested invocation prompt, and keep **the Human Action Window open at all times for the human user to take action** (review, edit by hand, pause/reroute, or delegate).\n";
3216
+ declare const STARTER_CURSOR_CORE_RULES = "---\ndescription: Global canonical directives, 5 unbreakable commandments, and AI-SDLC pre-flight\nglobs: *\nalwaysApply: true\n---\n\n# Global AI-SDLC Rules (Core)\n\nCanonical Reference: [`process/09_agent_protocols.md`](process/09_agent_protocols.md) and [`process/01_governance_and_roles.md`](process/01_governance_and_roles.md).\n\n## 1. The 5 Unbreakable Commandments of Agents | Los 5 Mandamientos Inquebrantables de los Agentes\nEvery agent operating in this repository must strictly observe the following guardrails:\n1. **FORBIDDEN TO AUTO-APPROVE OR AUTO-MERGE | PROHIBIDO AUTO-APROBAR O AUTO-FUSIONAR**: Do not approve PRs or execute direct merges to protected branches (`main`, `release/*`). Approval is exclusively reserved for human reviewers.\n2. **FORBIDDEN TO INVENT PRODUCT OR ARCHITECTURE DECISIONS**: When facing ambiguity, formulate open questions (`open-questions`). Never assume undocumented requirements.\n3. **FORBIDDEN TO INTRODUCE DEPENDENCIES WITHOUT LICENSE INSPECTION**: Always verify `license-policy.yaml`. Viral dependencies (GPL/AGPL) or commercial paid dependencies (BSL/SSPL) are strictly forbidden without explicit human approval.\n4. **FORBIDDEN TO IGNORE CYBERSECURITY & PRIVACY (SECURITY-BY-DEFAULT & PRIVACY-BY-DEFAULT)**: Validate inputs, apply least privilege, protect PII according to `privacy-policy.yaml`, and accompany every change with security and privacy tests (`SEC-TEST-*`, `@privacy`). Suppressing linters or tests to force a green pipeline is prohibited.\n5. **MANDATORY CRYPTOGRAPHIC CITATION**: Link specifications using immutable identifiers and Unix LF-normalized SHA-256 hashes.\n\n## 2. Pre-Flight Verification Commands\nBefore finalizing changes or proposing commits:\n- Pre-flight with auto-fix: `pnpm run check:fix` (or `npx aisdlc check --fix`)\n- Full verification: `pnpm run verify:all` (or `npx aisdlc verify all`)\n- Unit tests: `pnpm test`\n- Strict typing: `pnpm run typecheck`\n\n## 3. Git 4-Tier Hierarchy\n- Tier 1: `main` (Production)\n- Tier 2: `release/vX.Y.Z` (Frozen release)\n- Tier 3: `feat/<FEAT-ID>-<slug>` / `bug/<BUG-ID>-<slug>` (Feature)\n- Tier 4: `task/<PARENT-ID>/<TSK-ID>-<slug>` (Atomic task)\n\n## 4. Specialized Roles Mapping\n| Role | Mission (summary) |\n| :--- | :--- |\n| `agent-product-analyst` | Phase 0 NAF v4 Enterprise & Capability Vision (`ARCH-INTRO-*`, `/grill-me` Socratic decision-tree discovery interview of up to 15 questions \u2014 1\u20132 per turn with `(Recommended)` options \u2014 defining UI modality & CRUD prerequisites) + ProductShape artifacts (`ACT-*`, `UC-*`, `FR-*`, `QR-*`, `BR-*`); status starts `draft`. |\n| `agent-threat-modeler` | STRIDE/OWASP threat triad (`ACT-THREAT-*`, `ABUSE-*`, `SEC-REQ-*`); hands off to `agent-privacy-officer`. |\n| `agent-privacy-officer` | LINDDUN & GDPR/LOPDGDD privacy triad (`DATA-ASSET-*`, `PRIV-ROPA-*`, `PRIV-REQ-*`) + Pre-Merge `verify:privacy`. |\n| `agent-system-architect` | Modular arc42 + NAF v4 architecture (`CMP-*`, Mermaid diagrams, `ADR-*`). |\n| `agent-qa-engineer` | FAILING BDD/Gherkin, E2E, and UI suites (using agreed project tech; Happy Path, Sad Path & Boundary Values per interactive UI element) before `agent-developer` writes code. |\n| `agent-developer` | Atomic tasks from `tasks.md`; respects autonomy mode; runs `pnpm run check:fix` + `pnpm run verify:all`. |\n| `agent-expert-user` | Bimodal MVP evaluation (design) and functional validation against `UC-*` and `FR-*` (pre-PR). |\n| `agent-devops` | CI/CD, Docker, IaC only \u2014 clarifies deployment/CI strategy; **FORBIDDEN** to touch `src/` or `packages/*/src/`. |\n| `agent-auditor` | Transversal auditor: YAGNI, Clean Code/SOLID/AST thresholds, UI standards, SPDX licenses (`verify:licenses`); returns handoff to audited agent. |\n| `agent-adversary` | Transversal Red Team adversary: PoC breaks, mutation testing & Pre-Merge security diff audit (`verify:security`, CVSS v3.1); returns handoff to audited agent. |\n\n## 5. Task Autonomy Modes (`tasks.md`)\n- \uD83D\uDFE2 **`AUTONOMOUS`**: Low risk. Implement directly, omit handoff block.\n- \uD83D\uDFE1 **`HUMAN_REVIEW_PLAN`**: Medium risk. Emit Workflow Handoff block, wait for approval.\n- \uD83D\uDFE0 **`AMBIGUOUS`**: Blocked. Request human clarification.\n- \uD83D\uDD34 **`HIGH_RISK_MANUAL`**: Critical risk. Human-only execution.\n\n## 6. Workflow Handoff Protocol and Human Action Window | Ventana de Acci\u00F3n Humana\n- **Clean Context Window Isolation per Specialist Agent (Cross-IDE)**: Never execute two distinct specialist roles in the same conversation context window. In subagent-capable IDEs (Antigravity `invoke_subagent`, Claude Code `Task` / `.claude/agents/`, GitHub Copilot `@agent-*`), always delegate each specialist phase to a fresh isolated subagent context that hydrates state strictly from disk (`specs/`, `handoff.yaml`, `get_active_handoff`). In chat-based IDEs (Cursor, Windsurf, Aider), persist all artifacts to disk, emit the Workflow Handoff block, and instruct the user to open a **New Chat** (or `/clear`) before running the next specialist.\n- **Conditional Activation**:\n * \uD83D\uDFE2 **`AUTONOMOUS`** (or PR/CI supervision): **OMITTED**. Do not request or emit interactive handoff.\n * \uD83D\uDFE1 **Autonomy $\\ge$ `HUMAN_REVIEW_PLAN`** (`HUMAN_REVIEW_PLAN`, `AMBIGUOUS`, `HIGH_RISK_MANUAL`): **MANDATORY**. Emit Workflow Handoff block compliant with [`templates/workflow/agent-handoff.template.md`](templates/workflow/agent-handoff.template.md) and **STOP**.\n- **Content**: Declare completed deliverables, recommend next role(s), provide suggested invocation prompt, and **keep the Human Action Window open for the human user to take action** (review, edit by hand, pause/reroute, or delegate).\n";
3217
+ declare const STARTER_CURSOR_PRODUCT_RULES = "---\ndescription: Directives for SDD specifications editing, ProductShape artifacts, Cybersecurity, and Privacy\nglobs: specs/**, templates/**, product/**, security/**, privacy/**\nalwaysApply: false\n---\n\n# Product and SDD Specification Directives\n\nCanonical Reference: [`process/09_agent_protocols.md`](process/09_agent_protocols.md) and [`process/02_product_definition.md`](process/02_product_definition.md).\n\n## 1. Identifier Taxonomy and JSON Schemas\nEvery generated or modified artifact must follow canonical naming and satisfy its schema in `schemas/`:\n- Enterprise & Capability Vision (Phase 0): `ARCH-INTRO-[NUM3]` (`schemas/architecture/introduction-and-goals.schema.json`)\n- Product Actors: `ACT-[SUFFIX]` (`schemas/product/actor.schema.json`)\n- Use Cases: `UC-[SUFFIX]` (`schemas/product/use-case.schema.json`)\n- Functional Requirements: `FR-[SUFFIX]-[NUM3]` (`schemas/product/requirement.schema.json`)\n- Quality Requirements: `QR-[SUFFIX]` (`schemas/product/requirement.schema.json`)\n- Business Rules: `BR-[SUFFIX]` (`schemas/product/business-rule.schema.json`)\n- Threat Actors: `ACT-THREAT-[SUFFIX]` (`schemas/security/threat-actor.schema.json`)\n- Abuse Cases: `ABUSE-[SUFFIX]` (`schemas/security/abuse-case.schema.json`)\n- Security Requirements: `SEC-REQ-[SUFFIX]` (`schemas/security/security-req.schema.json`)\n- Data Assets: `DATA-ASSET-[SUFFIX]` (`schemas/privacy/data-asset.schema.json`)\n- Processing Activities (ROPA): `PRIV-ROPA-[SUFFIX]` (`schemas/privacy/privacy-ropa.schema.json`)\n- Privacy Requirements: `PRIV-REQ-[SUFFIX]` (`schemas/privacy/privacy-req.schema.json`)\n\n## 2. \"AI as Scribe\" Protocol Rules\n- **Phase 0 & Incremental Discovery (`/grill-me` Socratic Decision-Tree Interview)**: When starting a new product or resolving open product decisions, conduct a `/grill-me` Socratic decision-tree discovery interview of up to 15 questions (1\u20132 focused questions per turn with `(Recommended)` options, Read-First codebase inspection, constructive pushback, and No-Premature-Drafting Gate until the product scope and target UI modality \u2014 Web, Desktop, Console/CLI, Mobile, or API/SDK \u2014 are completely clear), instantiate `ARCH-INTRO-001` (`naf-perspective: \"Enterprise & Capability\"`) in `draft`, and derive `ACT-*`, `UC-*`, `BR-*`, `FR-*`, and `QR-*` (including runnable UI/CLI requirements and master-entity CRUD prerequisites when human actors exist).\n- **Initial Status**: Every new artifact strictly starts with `status: draft`. Creating artifacts directly as `active` or `approved` is forbidden.\n- **Gherkin Acceptance Criteria**: In every `FR-*` requirement, include a ```gherkin block with tags `@<ID> @automated @regression` and at least one `Scenario Outline` with an `Examples` data table.\n- **Security & Privacy Mitigations**: Every `SEC-REQ-*` must include negative Gherkin tests (`@security @mitigation`) and every `PRIV-REQ-*` must include Gherkin privacy scenarios (`@privacy @gdpr`).\n- **Use Cases (User Story Structure)**: In every `UC-*`, the `1. Intent and Outcome` section must be written using the standard English structure: `As a <ACT-ID>... I want <action>... To <outcome>...`.\n- **Deterministic Validation**: Run `pnpm run verify:schemas`, `pnpm run verify:duplicates`, and `pnpm run verify:privacy` before presenting the draft to the human reviewer.\n";
3218
3218
  declare const STARTER_CURSOR_QUALITY_RULES = "---\ndescription: Directives for code quality, architecture, TDD, and complexity limits\nglobs: packages/**, src/**, scripts/**, tests/**\nalwaysApply: false\n---\n\n# Code Quality and Architecture Directives\n\nCanonical Reference: [`process/09_agent_protocols.md`](process/09_agent_protocols.md) and [`process/10_quality_management_and_release_gates.md`](process/10_quality_management_and_release_gates.md).\n\n## 1. Non-Negotiable Quality Thresholds (`quality-policy.yaml`)\n- **Cyclomatic Complexity (CC)**: $\\le 10$ per function (warning alert if $> 7$).\n- **Cognitive Complexity**: $\\le 15$ per function (warning if $> 10$).\n- **Maintainability Index (MI)**: $\\ge 50$ (target minimum $\\ge 65$).\n- **Function Length**: $\\le 40$ lines (warning if $> 30$).\n- If a function exceeds these limits, decompose logic into cohesive helper functions before completing the task.\n\n## 2. TDD Methodology, Modular & Hexagonal Architecture (Ports & Adapters) & Unit Testing\n- **Mandatory 4-Step Implementation Order**: (1) Define/refine Modular & Hexagonal Architecture (Ports & Adapters) module boundaries, (2) Define interfaces/ports for all inter-module interactions to avoid direct module coupling, (3) Implement components consuming/fulfilling those interfaces, (4) Write unit tests for every function covering Happy Path, Sad Path, Boundary Values, and Unexpected/Null/Empty values.\n- Apply Test-Driven Development: generate unit tests before or alongside component logic.\n- Tag tests linked to requirements with header comments citing the ID (`FR-*`, `SEC-REQ-*`, or `PRIV-REQ-*`) to enable reverse traceability.\n- Suppressing types (`any`), linter rules (`eslint-disable`), or unit tests to force pipeline passage is prohibited.\n- **Mandatory Regression Verification**: Never declare build, typecheck, or bug fixes complete without executing the relevant automated test suite (`pnpm test`, `dotnet test`) and confirming zero regressions.\n- **Global Styles Protection**: When touching or refactoring frontend views, verify that global stylesheet/asset imports remain intact and that layouts suffer no visual disruption.\n- **Skills-First Implementation Directive**: When authoring or refactoring code (especially frontend, UI, styling, accessibility, or framework-specific logic), proactively inspect and leverage available specialized agent skills (e.g., modern web guidance, accessibility a11y, DevTools, framework skills) to produce standard-compliant, modern implementations rather than guessing ad-hoc patterns.\n\n## 3. Open Source License Policy (`license-policy.yaml`)\n- Before adding any dependency to `package.json`, verify its SPDX identifier.\n- Only permissive dependencies are allowed (`MIT`, `Apache-2.0`, `BSD-2/3-Clause`, `ISC`).\n- Viral licenses (`GPL`, `AGPL`) or commercial licenses without approval (`BSL`, `SSPL`) are prohibited.\n\n## 4. Structured Output `--json`\n- Any extension to CLI commands must preserve compatibility with the `--json` flag and suppress ANSI escape codes when this flag is present.\n";
3219
- declare const STARTER_CLAUDE_RULES = "# Claude Code Instructions - AI-SDLC Monorepo\n\nCanonical Reference: [`process/09_agent_protocols.md`](process/09_agent_protocols.md) and [`process/01_governance_and_roles.md`](process/01_governance_and_roles.md).\n\nThis repository implements the **AI-SDLC** framework (Spec-Driven Development, deterministic quality governance, Privacy-by-Design, and OSS license compliance for human-agent collaboration).\n\n---\n\n## 1. The 5 Unbreakable Commandments of Agents | Los 5 Mandamientos Inquebrantables de los Agentes\n\n1. **FORBIDDEN TO AUTO-APPROVE OR AUTO-MERGE | PROHIBIDO AUTO-APROBAR O AUTO-FUSIONAR**: Never approve PRs or merge directly to `main` or protected branches. Approval is exclusively a human prerogative.\n2. **FORBIDDEN TO INVENT PRODUCT OR ARCHITECTURE DECISIONS**: When facing ambiguous requirements, formulate open questions (`open-questions`). Never assume undocumented behaviors.\n3. **FORBIDDEN TO INTRODUCE DEPENDENCIES WITHOUT LICENSE INSPECTION**: Always validate the SPDX identifier against `license-policy.yaml`. Viral libraries (`GPL`/`AGPL`) or paid commercial libraries (`BSL`/`SSPL`) are strictly forbidden without formal human approval.\n4. **FORBIDDEN TO IGNORE CYBERSECURITY & PRIVACY (SECURITY-BY-DEFAULT & PRIVACY-BY-DEFAULT)**: Validate inputs, sanitize data, protect PII according to `privacy-policy.yaml`, and include mitigation tests (`SEC-TEST-*`, `@privacy`). Disabling linters or suppressing typing errors (`any`, `@ts-ignore`) is prohibited.\n5. **MANDATORY CRYPTOGRAPHIC CITATION**: Every spec or sidecar must cite immutable identifiers and Unix LF-normalized SHA-256 digests.\n\n---\n\n## 2. Essential Execution and Pre-Flight Commands\n\n```bash\n# 1. Pre-flight with deterministic auto-fix (Gherkin & SHA-256 digests)\npnpm run check:fix\n\n# 2. Complete verification across all Quality Gates (Quality, RTM, Governance, Testing, Licenses, PDaC, Schemas, Duplicates, Security, Privacy)\npnpm run verify:all\n\n# 3. Complete automated test suite (Vitest)\npnpm test\n\n# 4. Strict TypeScript typechecking\npnpm run typecheck\n\n# 5. Scaffolding for a new SDD change\npnpm run change:new <name> --id <chg-id>\n\n# 6. Canonical integration of a completed SDD change\nnpx tsx packages/cli/src/index.ts sdd integrate --change <chg-id>\n```\n\n---\n\n## 3. Specialized Roles Mapping (Summary \u2014 Full Prompts in `.claude/agents/`)\n\n| Role | Mission (summary) |\n| :--- | :--- |\n| `agent-product-analyst` | Phase 0 NAF v4 Vision (`ARCH-INTRO-*`, up to 15-question discovery defining UI modality & CRUD prerequisites) + ProductShape (`ACT-*`, `UC-*`, `FR-*`, `QR-*`, `BR-*`); `status` starts `draft`. |\n| `agent-threat-modeler` | STRIDE/OWASP threat triad (`ACT-THREAT-*`, `ABUSE-*`, `SEC-REQ-*`); hands off to `agent-privacy-officer`. |\n| `agent-privacy-officer` | LINDDUN & GDPR/LOPDGDD privacy triad (`DATA-ASSET-*`, `PRIV-ROPA-*`, `PRIV-REQ-*`) + Pre-Merge `verify:privacy`. |\n| `agent-system-architect` | Modular arc42 + NAF v4 architecture (`CMP-*`, `ARCH-DEP-001`, Mermaid diagrams, `ADR-*`). |\n| `agent-qa-engineer` | FAILING BDD/Gherkin, E2E, and UI suites (Happy Path, Sad Path & Boundary Values per interactive UI element) before `agent-developer` writes code. |\n| `agent-developer` | Atomic tasks from `tasks.md` in 5-step Hexagonal order; respects autonomy mode; runs `pnpm run check:fix` + `pnpm run verify:all`. |\n| `agent-expert-user` | Bimodal MVP evaluation (design) and functional validation against `UC-*` and `FR-*` (pre-PR). |\n| `agent-devops` | CI/CD, Docker, IaC only \u2014 clarifies deployment/CI strategy; **FORBIDDEN** to touch `src/` or `packages/*/src/`. |\n| `agent-auditor` | Transversal auditor: YAGNI, Clean Code/SOLID/AST thresholds, UI standards, SPDX licenses (`verify:licenses`); returns handoff to audited agent. |\n| `agent-adversary` | Transversal Red Team adversary: PoC breaks, mutation testing & Pre-Merge security diff audit (`verify:security`, CVSS v3.1); returns handoff to audited agent. |\n\n---\n\n## 4. Task Autonomy Modes (`tasks.md`)\n\n- \uD83D\uDFE2 **`AUTONOMOUS`**: Low risk, isolated task. Implement code and tests directly. Omit interactive handoff block.\n- \uD83D\uDFE1 **`HUMAN_REVIEW_PLAN`**: Medium risk. Design detailed plan, emit Workflow Handoff block, and wait for human confirmation before coding.\n- \uD83D\uDFE0 **`AMBIGUOUS`**: Incomplete requirements. Blocked: request human clarification.\n- \uD83D\uDD34 **`HIGH_RISK_MANUAL`**: Critical risk (migrations, cryptography). Direct manual execution by humans only.\n\n---\n\n## 5. Git Workflow and Commit Rules with Trailers\n\n### 4-Tier Hierarchy\n1. **Tier 1 (`main`)**: Absolute stability and production readiness.\n2. **Tier 2 (`release/vX.Y.Z`)**: Release stabilization.\n3. **Tier 3 (`feat/<FEAT-ID>-<slug>` or `bug/<BUG-ID>-<slug>`)**: SDD delivery increment.\n4. **Tier 4 (`task/<PARENT-ID>/<TSK-ID>-<slug>`)**: Atomic implementation task.\n\n### Commit Format (Conventional Commits + Git Trailers)\n```text\nfeat(scope): concise imperative description (#issue)\n\nBody explaining the motivation and technical justification of the change.\n\nAuthor-Type: agent\nAI-Model: claude-3-7-sonnet\nAI-Reasoning-Effort: high\nTask-ID: TSK-001\nChange-ID: CHG-026-AGENT-NATIVE-CONFIGS\n```\n\n---\n\n## 6. Non-Negotiable Quality Thresholds (`quality-policy.yaml`)\n- **Cyclomatic Complexity (CC)**: $\\le 10$\n- **Cognitive Complexity**: $\\le 15$\n- **Maintainability Index (MI)**: $\\ge 50$\n- **Lines per Function**: $\\le 40$\n- Deterministic CLI output: every `verify` command supports `--json` without ANSI escape codes.\n\n---\n\n## 7. Workflow Handoff Protocol and Human Action Window | Ventana de Acci\u00F3n Humana\n- **Clean Context Window Isolation per Specialist Agent (Cross-IDE)**: Never execute two distinct specialist roles in the same conversation context window. In subagent-capable IDEs (Antigravity `invoke_subagent`, Claude Code `Task` / `.claude/agents/`, GitHub Copilot `@agent-*`), always delegate each specialist phase to a fresh isolated subagent context that hydrates state strictly from disk (`specs/`, `handoff.yaml`, `get_active_handoff`). In chat-based IDEs (Cursor, Windsurf, Aider), persist all artifacts to disk, emit the Workflow Handoff block, and instruct the user to open a **New Chat** (or `/clear`) before running the next specialist.\n- **Conditional Activation**:\n - \uD83D\uDFE2 **`AUTONOMOUS`** (or PR/CI final supervision): **OMITTED**. Uninterrupted continuous execution.\n - \uD83D\uDFE1 **Autonomy $\\ge$ `HUMAN_REVIEW_PLAN`** (`HUMAN_REVIEW_PLAN`, `AMBIGUOUS`, `HIGH_RISK_MANUAL`): **MANDATORY**. Emit Workflow Handoff block ([`templates/workflow/agent-handoff.template.md`](templates/workflow/agent-handoff.template.md)) and **STOP**.\n- **Content**: Completed deliverables, suggested next role(s), copy-paste ready invocation prompt, and **an open Human Action Window for the user to take action** (review, edit by hand, pause/reroute, or delegate).\n";
3220
- declare const STARTER_COPILOT_RULES = "<!--\n AI-SDLC: Canonical Instructions for GitHub Copilot (Chat & Workspace)\n Canonical Reference: process/09_agent_protocols.md and process/01_governance_and_roles.md\n-->\n\n# GitHub Copilot Instructions - AI-SDLC Framework\n\nYou operate as an assistant agent in the **AI-SDLC** repository. You are not a simple text auto-completer: you are a specialized technical worker subject to normative directives, deterministic autonomy boundaries, and strict traceability.\n\n---\n\n## 1. The 5 Unbreakable Commandments of Agents | Los 5 Mandamientos Inquebrantables de los Agentes\n\n1. **FORBIDDEN TO AUTO-APPROVE OR AUTO-MERGE | PROHIBIDO AUTO-APROBAR O AUTO-FUSIONAR**:\n - Never execute PR approvals or direct merges to stable branches (`main`, `release/*`). Approval is an exclusive human prerogative.\n2. **FORBIDDEN TO INVENT PRODUCT OR ARCHITECTURE DECISIONS**:\n - If a requirement is ambiguous or incomplete, formulate open questions (`open-questions`) to the human user. Guessing undocumented intentions is forbidden.\n3. **FORBIDDEN TO INTRODUCE DEPENDENCIES WITHOUT LICENSE INSPECTION**:\n - Before suggesting or adding dependencies to manifests (`package.json`, etc.), validate their SPDX identifier against `license-policy.yaml`. Adding viral (`GPL`, `AGPL`) or commercial paid (`BSL`, `SSPL`) licenses without human authorization is prohibited.\n4. **FORBIDDEN TO IGNORE CYBERSECURITY & PRIVACY (SECURITY-BY-DEFAULT & PRIVACY-BY-DEFAULT)**:\n - All code must enforce least privilege, sanitize external inputs, protect PII according to `privacy-policy.yaml`, and include mitigation tests (`SEC-TEST-*`, `@privacy`). Suppressing types (`any`), linters, or tests to force a green pipeline is prohibited.\n5. **MANDATORY CRYPTOGRAPHIC CITATION**:\n - Every specification, design, or sidecar must reference canonical identifiers and their Unix LF-normalized SHA-256 digests.\n\n---\n\n## 2. Git 4-Tier Branch Hierarchy\n\nWhen proposing or creating branches, respect the strict hierarchical structure:\n- **Tier 1**: `main` (Protected production branch with maximum stability).\n- **Tier 2**: `release/vX.Y.Z` (Release consolidation and scope freeze branch).\n- **Tier 3**: `feat/<FEAT-ID>-<slug>` or `bug/<BUG-ID>-<slug>` (SDD increment or feature branch).\n- **Tier 4**: `task/<PARENT-ID>/<TSK-ID>-<slug>` (Atomic development task branch).\n\n---\n\n## 3. Pre-Flight Commands and Quality Gates\n\nBefore concluding any intervention or suggesting a commit:\n- **Unified pre-flight with auto-fix**: `pnpm run check:fix` (or `npx aisdlc check --fix`).\n- **Complete Quality Gates verification**: `pnpm run verify:all` (or `npx aisdlc verify all`).\n- **Unit test suite**: `pnpm test`.\n- **Strict typing**: `pnpm run typecheck`.\n\nNon-Negotiable Quality Thresholds (`quality-policy.yaml`):\n- Cyclomatic Complexity (CC) $\\le 10$.\n- Cognitive Complexity $\\le 15$.\n- Maintainability Index (MI) $\\ge 50$.\n- Maximum lines per function $\\le 40$.\n\n---\n\n## 4. Task Autonomy Modes (`tasks.md`)\n\n- \uD83D\uDFE2 **`AUTONOMOUS`**: Low risk, isolated task. Implement code and tests directly. Omit interactive handoff block.\n- \uD83D\uDFE1 **`HUMAN_REVIEW_PLAN`**: Medium risk. Design detailed plan, emit Workflow Handoff block, and wait for human confirmation before coding.\n- \uD83D\uDFE0 **`AMBIGUOUS`**: Incomplete requirements. Blocked: request human clarification.\n- \uD83D\uDD34 **`HIGH_RISK_MANUAL`**: Critical risk (migrations, cryptography). Direct manual execution by humans only.\n\n---\n\n## 5. Workflow Handoff Protocol and Human Action Window | Ventana de Acci\u00F3n Humana\n- **Clean Context Window Isolation per Specialist Agent (Cross-IDE)**: Never execute two distinct specialist roles in the same conversation context window. In subagent-capable IDEs (Antigravity `invoke_subagent`, Claude Code `Task` / `.claude/agents/`, GitHub Copilot `@agent-*`), always delegate each specialist phase to a fresh isolated subagent context that hydrates state strictly from disk (`specs/`, `handoff.yaml`, `get_active_handoff`). In chat-based IDEs (Cursor, Windsurf, Aider), persist all artifacts to disk, emit the Workflow Handoff block, and instruct the user to open a **New Chat** (or `/clear`) before running the next specialist.\n- **Conditional Activation**:\n * \uD83D\uDFE2 **`AUTONOMOUS` Mode** (or PR/CI final supervision): **OMITTED**. Do not interrupt unattended execution.\n * \uD83D\uDFE1 **Autonomy $\\ge$ `HUMAN_REVIEW_PLAN`** (`HUMAN_REVIEW_PLAN`, `AMBIGUOUS`, `HIGH_RISK_MANUAL`): **MANDATORY**. Emit Workflow Handoff block compliant with [`templates/workflow/agent-handoff.template.md`](../templates/workflow/agent-handoff.template.md) and **STOP**.\n- **Components**: Declare completed deliverables, recommend next roles (`agent-threat-modeler`, `agent-privacy-officer`, `agent-system-architect`, `agent-developer`, `agent-devops`, `agent-auditor`, `agent-adversary`), provide suggested invocation prompt, and **keep the Human Action Window open at all times for the user to take action** (review, edit by hand, pause/reroute, or delegate).\n\n---\n\n## 6. Specialized Roles Mapping\n\n| Role | Mission (summary) |\n| :--- | :--- |\n| `agent-product-analyst` | Phase 0 NAF v4 Enterprise & Capability Vision (`ARCH-INTRO-*`, up to 15-question discovery interview defining UI modality, delivery model & CRUD prerequisites) + ProductShape artifacts (`ACT-*`, `UC-*`, `FR-*`, `QR-*`, `BR-*`); `status` starts `draft`. |\n| `agent-threat-modeler` | STRIDE/OWASP threat triad (`ACT-THREAT-*`, `ABUSE-*`, `SEC-REQ-*`); hands off to `agent-privacy-officer`. |\n| `agent-privacy-officer` | LINDDUN & GDPR/LOPDGDD privacy triad (`DATA-ASSET-*`, `PRIV-ROPA-*`, `PRIV-REQ-*`) + Pre-Merge `verify:privacy`. |\n| `agent-system-architect` | Modular arc42 + NAF v4 architecture (`CMP-*`, `ARCH-DEP-001`, Mermaid diagrams, `ADR-*`). |\n| `agent-qa-engineer` | FAILING BDD/Gherkin, E2E, and UI suites (using agreed project tech; Happy Path, Sad Path & Boundary Values per interactive UI element) before `agent-developer` writes production code. |\n| `agent-developer` | Atomic tasks from `tasks.md`; autonomy mode must be respected; runs `pnpm run check:fix` + `pnpm run verify:all`. |\n| `agent-expert-user` | Bimodal MVP evaluation (design) and functional validation against `UC-*` and `FR-*` (pre-PR). |\n| `agent-devops` | CI/CD, Docker, IaC only \u2014 clarifies deployment/CI strategy; **FORBIDDEN** to touch `src/` or `packages/*/src/`. |\n| `agent-auditor` | Transversal auditor: YAGNI, Clean Code/SOLID/AST thresholds, UI standards, SPDX licenses (`verify:licenses`); returns handoff to audited agent. |\n| `agent-adversary` | Transversal Red Team adversary: PoC breaks, mutation testing & Pre-Merge security diff audit (`verify:security`, CVSS v3.1); returns handoff to audited agent. |\n\n---\n\n## 7. Canonical Reference\nTo consult full protocols, specialized prompts, and interface contracts:\n- [`process/09_agent_protocols.md`](../process/09_agent_protocols.md)\n- [`process/01_governance_and_roles.md`](../process/01_governance_and_roles.md)\n";
3219
+ declare const STARTER_CLAUDE_RULES = "# Claude Code Instructions - AI-SDLC Monorepo\n\nCanonical Reference: [`process/09_agent_protocols.md`](process/09_agent_protocols.md) and [`process/01_governance_and_roles.md`](process/01_governance_and_roles.md).\n\nThis repository implements the **AI-SDLC** framework (Spec-Driven Development, deterministic quality governance, Privacy-by-Design, and OSS license compliance for human-agent collaboration).\n\n---\n\n## 1. The 5 Unbreakable Commandments of Agents | Los 5 Mandamientos Inquebrantables de los Agentes\n\n1. **FORBIDDEN TO AUTO-APPROVE OR AUTO-MERGE | PROHIBIDO AUTO-APROBAR O AUTO-FUSIONAR**: Never approve PRs or merge directly to `main` or protected branches. Approval is exclusively a human prerogative.\n2. **FORBIDDEN TO INVENT PRODUCT OR ARCHITECTURE DECISIONS**: When facing ambiguous requirements, formulate open questions (`open-questions`). Never assume undocumented behaviors.\n3. **FORBIDDEN TO INTRODUCE DEPENDENCIES WITHOUT LICENSE INSPECTION**: Always validate the SPDX identifier against `license-policy.yaml`. Viral libraries (`GPL`/`AGPL`) or paid commercial libraries (`BSL`/`SSPL`) are strictly forbidden without formal human approval.\n4. **FORBIDDEN TO IGNORE CYBERSECURITY & PRIVACY (SECURITY-BY-DEFAULT & PRIVACY-BY-DEFAULT)**: Validate inputs, sanitize data, protect PII according to `privacy-policy.yaml`, and include mitigation tests (`SEC-TEST-*`, `@privacy`). Disabling linters or suppressing typing errors (`any`, `@ts-ignore`) is prohibited.\n5. **MANDATORY CRYPTOGRAPHIC CITATION**: Every spec or sidecar must cite immutable identifiers and Unix LF-normalized SHA-256 digests.\n\n---\n\n## 2. Essential Execution and Pre-Flight Commands\n\n```bash\n# 1. Pre-flight with deterministic auto-fix (Gherkin & SHA-256 digests)\npnpm run check:fix\n\n# 2. Complete verification across all Quality Gates (Quality, RTM, Governance, Testing, Licenses, PDaC, Schemas, Duplicates, Security, Privacy)\npnpm run verify:all\n\n# 3. Complete automated test suite (Vitest)\npnpm test\n\n# 4. Strict TypeScript typechecking\npnpm run typecheck\n\n# 5. Scaffolding for a new SDD change\npnpm run change:new <name> --id <chg-id>\n\n# 6. Canonical integration of a completed SDD change\nnpx tsx packages/cli/src/index.ts sdd integrate --change <chg-id>\n```\n\n---\n\n## 3. Specialized Roles Mapping (Summary \u2014 Full Prompts in `.claude/agents/`)\n\n| Role | Mission (summary) |\n| :--- | :--- |\n| `agent-product-analyst` | Phase 0 NAF v4 Vision (`ARCH-INTRO-*`, `/grill-me` Socratic decision-tree discovery interview of up to 15 questions \u2014 1\u20132 per turn with `(Recommended)` options \u2014 defining UI modality & CRUD prerequisites) + ProductShape (`ACT-*`, `UC-*`, `FR-*`, `QR-*`, `BR-*`); `status` starts `draft`. |\n| `agent-threat-modeler` | STRIDE/OWASP threat triad (`ACT-THREAT-*`, `ABUSE-*`, `SEC-REQ-*`); hands off to `agent-privacy-officer`. |\n| `agent-privacy-officer` | LINDDUN & GDPR/LOPDGDD privacy triad (`DATA-ASSET-*`, `PRIV-ROPA-*`, `PRIV-REQ-*`) + Pre-Merge `verify:privacy`. |\n| `agent-system-architect` | Modular arc42 + NAF v4 architecture (`CMP-*`, `ARCH-DEP-001`, Mermaid diagrams, `ADR-*`). |\n| `agent-qa-engineer` | FAILING BDD/Gherkin, E2E, and UI suites (Happy Path, Sad Path & Boundary Values per interactive UI element) before `agent-developer` writes code. |\n| `agent-developer` | Atomic tasks from `tasks.md` in 5-step Hexagonal order; respects autonomy mode; runs `pnpm run check:fix` + `pnpm run verify:all`. |\n| `agent-expert-user` | Bimodal MVP evaluation (design) and functional validation against `UC-*` and `FR-*` (pre-PR). |\n| `agent-devops` | CI/CD, Docker, IaC only \u2014 clarifies deployment/CI strategy; **FORBIDDEN** to touch `src/` or `packages/*/src/`. |\n| `agent-auditor` | Transversal auditor: YAGNI, Clean Code/SOLID/AST thresholds, UI standards, SPDX licenses (`verify:licenses`); returns handoff to audited agent. |\n| `agent-adversary` | Transversal Red Team adversary: PoC breaks, mutation testing & Pre-Merge security diff audit (`verify:security`, CVSS v3.1); returns handoff to audited agent. |\n\n---\n\n## 4. Task Autonomy Modes (`tasks.md`)\n\n- \uD83D\uDFE2 **`AUTONOMOUS`**: Low risk, isolated task. Implement code and tests directly. Omit interactive handoff block.\n- \uD83D\uDFE1 **`HUMAN_REVIEW_PLAN`**: Medium risk. Design detailed plan, emit Workflow Handoff block, and wait for human confirmation before coding.\n- \uD83D\uDFE0 **`AMBIGUOUS`**: Incomplete requirements. Blocked: request human clarification.\n- \uD83D\uDD34 **`HIGH_RISK_MANUAL`**: Critical risk (migrations, cryptography). Direct manual execution by humans only.\n\n---\n\n## 5. Git Workflow and Commit Rules with Trailers\n\n### 4-Tier Hierarchy\n1. **Tier 1 (`main`)**: Absolute stability and production readiness.\n2. **Tier 2 (`release/vX.Y.Z`)**: Release stabilization.\n3. **Tier 3 (`feat/<FEAT-ID>-<slug>` or `bug/<BUG-ID>-<slug>`)**: SDD delivery increment.\n4. **Tier 4 (`task/<PARENT-ID>/<TSK-ID>-<slug>`)**: Atomic implementation task.\n\n### Commit Format (Conventional Commits + Git Trailers)\n```text\nfeat(scope): concise imperative description (#issue)\n\nBody explaining the motivation and technical justification of the change.\n\nAuthor-Type: agent\nAI-Model: claude-3-7-sonnet\nAI-Reasoning-Effort: high\nTask-ID: TSK-001\nChange-ID: CHG-026-AGENT-NATIVE-CONFIGS\n```\n\n---\n\n## 6. Non-Negotiable Quality Thresholds (`quality-policy.yaml`)\n- **Cyclomatic Complexity (CC)**: $\\le 10$\n- **Cognitive Complexity**: $\\le 15$\n- **Maintainability Index (MI)**: $\\ge 50$\n- **Lines per Function**: $\\le 40$\n- Deterministic CLI output: every `verify` command supports `--json` without ANSI escape codes.\n\n---\n\n## 7. Workflow Handoff Protocol and Human Action Window | Ventana de Acci\u00F3n Humana\n- **Clean Context Window Isolation per Specialist Agent (Cross-IDE)**: Never execute two distinct specialist roles in the same conversation context window. In subagent-capable IDEs (Antigravity `invoke_subagent`, Claude Code `Task` / `.claude/agents/`, GitHub Copilot `@agent-*`), always delegate each specialist phase to a fresh isolated subagent context that hydrates state strictly from disk (`specs/`, `handoff.yaml`, `get_active_handoff`). In chat-based IDEs (Cursor, Windsurf, Aider), persist all artifacts to disk, emit the Workflow Handoff block, and instruct the user to open a **New Chat** (or `/clear`) before running the next specialist.\n- **Conditional Activation**:\n - \uD83D\uDFE2 **`AUTONOMOUS`** (or PR/CI final supervision): **OMITTED**. Uninterrupted continuous execution.\n - \uD83D\uDFE1 **Autonomy $\\ge$ `HUMAN_REVIEW_PLAN`** (`HUMAN_REVIEW_PLAN`, `AMBIGUOUS`, `HIGH_RISK_MANUAL`): **MANDATORY**. Emit Workflow Handoff block ([`templates/workflow/agent-handoff.template.md`](templates/workflow/agent-handoff.template.md)) and **STOP**.\n- **Content**: Completed deliverables, suggested next role(s), copy-paste ready invocation prompt, and **an open Human Action Window for the user to take action** (review, edit by hand, pause/reroute, or delegate).\n";
3220
+ declare const STARTER_COPILOT_RULES = "<!--\n AI-SDLC: Canonical Instructions for GitHub Copilot (Chat & Workspace)\n Canonical Reference: process/09_agent_protocols.md and process/01_governance_and_roles.md\n-->\n\n# GitHub Copilot Instructions - AI-SDLC Framework\n\nYou operate as an assistant agent in the **AI-SDLC** repository. You are not a simple text auto-completer: you are a specialized technical worker subject to normative directives, deterministic autonomy boundaries, and strict traceability.\n\n---\n\n## 1. The 5 Unbreakable Commandments of Agents | Los 5 Mandamientos Inquebrantables de los Agentes\n\n1. **FORBIDDEN TO AUTO-APPROVE OR AUTO-MERGE | PROHIBIDO AUTO-APROBAR O AUTO-FUSIONAR**:\n - Never execute PR approvals or direct merges to stable branches (`main`, `release/*`). Approval is an exclusive human prerogative.\n2. **FORBIDDEN TO INVENT PRODUCT OR ARCHITECTURE DECISIONS**:\n - If a requirement is ambiguous or incomplete, formulate open questions (`open-questions`) to the human user. Guessing undocumented intentions is forbidden.\n3. **FORBIDDEN TO INTRODUCE DEPENDENCIES WITHOUT LICENSE INSPECTION**:\n - Before suggesting or adding dependencies to manifests (`package.json`, etc.), validate their SPDX identifier against `license-policy.yaml`. Adding viral (`GPL`, `AGPL`) or commercial paid (`BSL`, `SSPL`) licenses without human authorization is prohibited.\n4. **FORBIDDEN TO IGNORE CYBERSECURITY & PRIVACY (SECURITY-BY-DEFAULT & PRIVACY-BY-DEFAULT)**:\n - All code must enforce least privilege, sanitize external inputs, protect PII according to `privacy-policy.yaml`, and include mitigation tests (`SEC-TEST-*`, `@privacy`). Suppressing types (`any`), linters, or tests to force a green pipeline is prohibited.\n5. **MANDATORY CRYPTOGRAPHIC CITATION**:\n - Every specification, design, or sidecar must reference canonical identifiers and their Unix LF-normalized SHA-256 digests.\n\n---\n\n## 2. Git 4-Tier Branch Hierarchy\n\nWhen proposing or creating branches, respect the strict hierarchical structure:\n- **Tier 1**: `main` (Protected production branch with maximum stability).\n- **Tier 2**: `release/vX.Y.Z` (Release consolidation and scope freeze branch).\n- **Tier 3**: `feat/<FEAT-ID>-<slug>` or `bug/<BUG-ID>-<slug>` (SDD increment or feature branch).\n- **Tier 4**: `task/<PARENT-ID>/<TSK-ID>-<slug>` (Atomic development task branch).\n\n---\n\n## 3. Pre-Flight Commands and Quality Gates\n\nBefore concluding any intervention or suggesting a commit:\n- **Unified pre-flight with auto-fix**: `pnpm run check:fix` (or `npx aisdlc check --fix`).\n- **Complete Quality Gates verification**: `pnpm run verify:all` (or `npx aisdlc verify all`).\n- **Unit test suite**: `pnpm test`.\n- **Strict typing**: `pnpm run typecheck`.\n\nNon-Negotiable Quality Thresholds (`quality-policy.yaml`):\n- Cyclomatic Complexity (CC) $\\le 10$.\n- Cognitive Complexity $\\le 15$.\n- Maintainability Index (MI) $\\ge 50$.\n- Maximum lines per function $\\le 40$.\n\n---\n\n## 4. Task Autonomy Modes (`tasks.md`)\n\n- \uD83D\uDFE2 **`AUTONOMOUS`**: Low risk, isolated task. Implement code and tests directly. Omit interactive handoff block.\n- \uD83D\uDFE1 **`HUMAN_REVIEW_PLAN`**: Medium risk. Design detailed plan, emit Workflow Handoff block, and wait for human confirmation before coding.\n- \uD83D\uDFE0 **`AMBIGUOUS`**: Incomplete requirements. Blocked: request human clarification.\n- \uD83D\uDD34 **`HIGH_RISK_MANUAL`**: Critical risk (migrations, cryptography). Direct manual execution by humans only.\n\n---\n\n## 5. Workflow Handoff Protocol and Human Action Window | Ventana de Acci\u00F3n Humana\n- **Clean Context Window Isolation per Specialist Agent (Cross-IDE)**: Never execute two distinct specialist roles in the same conversation context window. In subagent-capable IDEs (Antigravity `invoke_subagent`, Claude Code `Task` / `.claude/agents/`, GitHub Copilot `@agent-*`), always delegate each specialist phase to a fresh isolated subagent context that hydrates state strictly from disk (`specs/`, `handoff.yaml`, `get_active_handoff`). In chat-based IDEs (Cursor, Windsurf, Aider), persist all artifacts to disk, emit the Workflow Handoff block, and instruct the user to open a **New Chat** (or `/clear`) before running the next specialist.\n- **Conditional Activation**:\n * \uD83D\uDFE2 **`AUTONOMOUS` Mode** (or PR/CI final supervision): **OMITTED**. Do not interrupt unattended execution.\n * \uD83D\uDFE1 **Autonomy $\\ge$ `HUMAN_REVIEW_PLAN`** (`HUMAN_REVIEW_PLAN`, `AMBIGUOUS`, `HIGH_RISK_MANUAL`): **MANDATORY**. Emit Workflow Handoff block compliant with [`templates/workflow/agent-handoff.template.md`](../templates/workflow/agent-handoff.template.md) and **STOP**.\n- **Components**: Declare completed deliverables, recommend next roles (`agent-threat-modeler`, `agent-privacy-officer`, `agent-system-architect`, `agent-developer`, `agent-devops`, `agent-auditor`, `agent-adversary`), provide suggested invocation prompt, and **keep the Human Action Window open at all times for the user to take action** (review, edit by hand, pause/reroute, or delegate).\n\n---\n\n## 6. Specialized Roles Mapping\n\n| Role | Mission (summary) |\n| :--- | :--- |\n| `agent-product-analyst` | Phase 0 NAF v4 Enterprise & Capability Vision (`ARCH-INTRO-*`, `/grill-me` Socratic decision-tree discovery interview of up to 15 questions \u2014 1\u20132 per turn with `(Recommended)` options \u2014 defining UI modality, delivery model & CRUD prerequisites) + ProductShape artifacts (`ACT-*`, `UC-*`, `FR-*`, `QR-*`, `BR-*`); `status` starts `draft`. |\n| `agent-threat-modeler` | STRIDE/OWASP threat triad (`ACT-THREAT-*`, `ABUSE-*`, `SEC-REQ-*`); hands off to `agent-privacy-officer`. |\n| `agent-privacy-officer` | LINDDUN & GDPR/LOPDGDD privacy triad (`DATA-ASSET-*`, `PRIV-ROPA-*`, `PRIV-REQ-*`) + Pre-Merge `verify:privacy`. |\n| `agent-system-architect` | Modular arc42 + NAF v4 architecture (`CMP-*`, `ARCH-DEP-001`, Mermaid diagrams, `ADR-*`). |\n| `agent-qa-engineer` | FAILING BDD/Gherkin, E2E, and UI suites (using agreed project tech; Happy Path, Sad Path & Boundary Values per interactive UI element) before `agent-developer` writes production code. |\n| `agent-developer` | Atomic tasks from `tasks.md`; autonomy mode must be respected; runs `pnpm run check:fix` + `pnpm run verify:all`. |\n| `agent-expert-user` | Bimodal MVP evaluation (design) and functional validation against `UC-*` and `FR-*` (pre-PR). |\n| `agent-devops` | CI/CD, Docker, IaC only \u2014 clarifies deployment/CI strategy; **FORBIDDEN** to touch `src/` or `packages/*/src/`. |\n| `agent-auditor` | Transversal auditor: YAGNI, Clean Code/SOLID/AST thresholds, UI standards, SPDX licenses (`verify:licenses`); returns handoff to audited agent. |\n| `agent-adversary` | Transversal Red Team adversary: PoC breaks, mutation testing & Pre-Merge security diff audit (`verify:security`, CVSS v3.1); returns handoff to audited agent. |\n\n---\n\n## 7. Canonical Reference\nTo consult full protocols, specialized prompts, and interface contracts:\n- [`process/09_agent_protocols.md`](../process/09_agent_protocols.md)\n- [`process/01_governance_and_roles.md`](../process/01_governance_and_roles.md)\n";
3221
3221
  declare const STARTER_CURSOR_MCP = "{\n \"mcpServers\": {\n \"ai-sdlc\": {\n \"command\": \"npx\",\n \"args\": [\n \"@aisdlc/mcp\"\n ],\n \"env\": {\n \"AISDLC_MCP_ROLE\": \"developer\"\n }\n }\n }\n}\n";
3222
3222
  declare const STARTER_ANTIGRAVITY_MCP = "{\n \"mcpServers\": {\n \"ai-sdlc\": {\n \"command\": \"npx\",\n \"args\": [\n \"@aisdlc/mcp\"\n ],\n \"env\": {\n \"AISDLC_MCP_ROLE\": \"developer\"\n }\n }\n }\n}\n";
3223
3223
  declare const STARTER_VSCODE_MCP = "{\n \"servers\": {\n \"ai-sdlc\": {\n \"command\": \"npx\",\n \"args\": [\n \"@aisdlc/mcp\"\n ],\n \"env\": {\n \"AISDLC_MCP_ROLE\": \"developer\"\n }\n }\n }\n}\n";