@jetrabbits/agentic 0.0.1

package/areas/template/GUIDE.md

@@ -0,0 +1,102 @@
+# Template authoring guide
+
+This guide explains how to use the templates in this directory to create new specializations.
+
+---
+
+## The four-layer structure
+
+Every specialization follows this layout — no exceptions:
+
+```text
+<spec-name>/
+├── AGENTS.md      # Navigation index — always loaded first
+├── rules/         # Constraints — always loaded for this spec
+├── skills/        # Capabilities — loaded on demand
+├── workflows/     # Orchestrated processes — triggered by /command
+└── prompts/       # Human-copy-paste templates (EN + RU)
+```
+
+Each layer has a purpose:
+
+| Layer | Purpose | Load behavior |
+|:---|:---|:---|
+| `AGENTS.md` | Maps the spec's files; sets load order | Always first |
+| `rules/` | Hard constraints the agent cannot override | Always all |
+| `skills/` | Technical patterns loaded for specific tasks | On demand — agent checks "When to load" |
+| `workflows/` | Step-by-step processes triggered by slash commands | On /command trigger |
+| `prompts/` | Copy-paste inputs for humans (bilingual) | Human reference |
+
+---
+
+## Step-by-step: adding a new spec
+
+### 1. Choose the parent area
+
+- `areas/software/` — application development domains
+- `areas/devops/` — infrastructure and platform domains
+
+If neither fits, propose a new area in a GitHub issue before creating files.
+
+### 2. Create the directory structure
+
+```bash
+mkdir -p areas/<domain>/<spec-name>/{rules,skills,workflows,prompts}
+```
+
+### 3. Author `AGENTS.md` using `AGENTS.tmpl.md`
+
+Fill in:
+- **What this area covers** — one paragraph, specific scope
+- **Inherited constraints** — what the parent area already enforces (don't repeat these in rules)
+- **Spec-specific constraints** — imperatives only; no advice
+- **Spec map** — every file with an inline description
+
+### 4. Author rules using `rule.tmpl.md`
+
+Rules are constraints, not guidelines. Each file covers one coherent concern. Ask:
+- What must the agent *never* do?
+- What is *required* before proceeding?
+- What blocks the task (P0) vs flags for review (P1)?
+
+### 5. Author skills using `skill.tmpl.md`
+
+Skills are technical capabilities loaded on demand. Each skill:
+- Has a precise "When to load" section
+- Covers 1–3 related techniques with real code examples
+- Ends with "Common mistakes" (3–5 items)
+
+### 6. Author workflows using `workflow.tmpl.md`
+
+Workflows are orchestrated processes. Each workflow:
+- Maps to a `/command` trigger
+- Has 3–8 steps with explicit roles, inputs, actions, and done-when criteria
+- Includes at least one failure path
+- Has measurable quality gates in the frontmatter
+
+### 7. Author prompts using `prompt.tmpl.md`
+
+Prompts are bilingual (EN + RU) templates for humans to copy-paste. Each prompt:
+- Links to a workflow via frontmatter
+- Has 2–3 examples with realistic values (no placeholders)
+- EN and RU blocks are semantically identical
+
+### 8. Update the parent area's `AGENTS.md`
+
+Add the new spec to the spec map. If this is the first spec in a new area, create the area-level `AGENTS.md` using `AGENTS-area.tmpl.md`.
+
+### 9. Test before submitting
+
+Run at least one workflow in a real agent session. Document the tool, model, and outcome in your PR.
+
+---
+
+## Quality bar
+
+A spec is ready to submit when:
+
+- All `AGENT INSTRUCTIONS` comment blocks are deleted.
+- All `{{PLACEHOLDER}}` values are replaced with real content.
+- All code examples are runnable without modification.
+- File lengths are within guidelines: rules ≤ 150, skills 150–350, workflows 60–200 lines.
+- The spec has been tested in a real agent session.

package/areas/template/PROMPTS.tmpl.md

@@ -0,0 +1,29 @@
+# PROMPTS: {{spec-name}}
+
+<!--
+AGENT INSTRUCTIONS:
+This file is a quick-reference index of all slash commands for this spec.
+Keep it as a simple table — no prose.
+Update this file every time a new prompt file is added or a command is renamed.
+Delete all AGENT INSTRUCTIONS comments before finalising.
+-->
+
+| Prompt | Use when |
+|:---|:---|
+| `/{{command-1}}` | {{ONE_LINE_TRIGGER — start with: a verb describing the situation, e.g. "Creating a new...", "Auditing an existing...", "Debugging a..."}} |
+| `/{{command-2}}` | {{ONE_LINE_TRIGGER}} |
+| `/{{command-3}}` | {{ONE_LINE_TRIGGER}} |
+| `/{{command-4}}` | {{ONE_LINE_TRIGGER}} |
+| `/{{command-5}}` | {{ONE_LINE_TRIGGER}} |
+
+<!--
+Example (for a copywriting spec):
+
+| Prompt | Use when |
+|:---|:---|
+| `/write-headline` | Generating A/B headline variants for a landing page or ad |
+| `/write-landing-page` | Creating conversion-focused landing page copy end-to-end |
+| `/rewrite-cta` | Optimising a CTA button, link, or inline prompt for higher conversion |
+| `/review-copy` | Auditing existing copy against brand voice and conversion criteria |
+| `/translate-adapt` | Adapting EN copy for a Russian-language market (not literal translation) |
+-->

package/areas/template/README.md

@@ -0,0 +1,57 @@
+# Template — authoring guide
+
+Use these templates when adding a new specialization area, spec, skill, workflow, rule, or prompt to agent-guides.
+
+## Templates
+
+| File | Use for |
+|:---|:---|
+| `AGENTS-area.tmpl.md` | Root `AGENTS.md` for a new domain area (e.g., `areas/design/`) |
+| `AGENTS.tmpl.md` | Spec-level `AGENTS.md` (e.g., `areas/design/copywriting/AGENTS.md`) |
+| `skill.tmpl.md` | A new `skills/{{name}}/SKILL.md` |
+| `workflow.tmpl.md` | A new `workflows/{{name}}.md` |
+| `rule.tmpl.md` | A new `rules/{{name}}.md` |
+| `prompt.tmpl.md` | A new `prompts/{{name}}.md` |
+| `PROMPTS.tmpl.md` | The `PROMPTS.md` index for a spec |
+
+## Authoring checklist
+
+Before opening a PR with new content:
+
+- [ ] All `{{PLACEHOLDER}}` values replaced with real content.
+- [ ] All `AGENT INSTRUCTIONS` comment blocks deleted.
+- [ ] Frontmatter fields fully populated (no empty values).
+- [ ] Skill "When to load" section is specific and actionable.
+- [ ] Rule constraints use imperative language: "must", "never", "required", "forbidden".
+- [ ] Workflow steps each have: `@role`, Input, Actions, and a checkable "Done when" criterion.
+- [ ] Prompt examples contain no placeholders; both EN and RU blocks present.
+- [ ] File stays within line-count targets (skill: 150–350; workflow: 60–200; rule: < 150).
+- [ ] Spec map in `AGENTS.md` matches the actual files on disk.
+
+## File naming conventions
+
+- Skill directories: `kebab-case/` (e.g., `api-design/`)
+- Workflow files: `kebab-case.md` (e.g., `create-endpoint.md`)
+- Rule files: `kebab-case.md` (e.g., `migration-safety.md`)
+- Prompt files: match workflow name (e.g., `create-endpoint.md`)
+
+## Line count targets
+
+| File type | Target |
+|:---|:---|
+| Area `AGENTS.md` | ≤ 100 lines |
+| Spec `AGENTS.md` | ≤ 80 lines |
+| `SKILL.md` | 150–350 lines |
+| Workflow | 60–200 lines |
+| Rule | ≤ 150 lines |
+| Prompt | no hard limit, but examples must be self-contained |
+
+## Guidance on quality gates
+
+Quality gates in workflows must be **objectively checkable** without human judgment:
+
+❌ Bad: "Output looks professional."
+✅ Good: "All required frontmatter fields are populated; file is under 350 lines; no `{{PLACEHOLDER}}` strings remain."
+
+❌ Bad: "Tests pass."
+✅ Good: "All unit tests in `src/__tests__/` pass with exit code 0 when run with `npm test`."

package/areas/template/README.tmpl.md

@@ -0,0 +1,51 @@
+# {{SPEC_NAME}}
+
+<!--
+AGENT INSTRUCTIONS:
+This README is for HUMANS, not agents. Write it clearly and concisely.
+Agents use AGENTS.md for navigation. Humans use README.md to understand what this spec covers.
+Keep it under 60 lines.
+Delete all AGENT INSTRUCTIONS comments before finalising.
+-->
+
+> {{ONE_SENTENCE_WHAT_THIS_SPEC_COVERS_AND_FOR_WHOM}}
+
+## Who this is for
+
+{{JOB_TITLE_OR_ROLE_THAT_OWNS_THIS_SPEC}} working on {{DOMAIN_OF_WORK}}.
+
+**Typical tasks:** {{COMMA_SEPARATED_LIST_OF_3_5_TYPICAL_TASKS}}
+
+## What's included
+
+| Type | Files | Covers |
+|:---|:---|:---|
+| Rules | {{N}} | {{WHAT_THE_RULES_GOVERN}} |
+| Skills | {{N}} | {{WHAT_TECHNIQUES_AND_TOOLS}} |
+| Workflows | {{N}} | {{WHAT_PROCEDURES}} |
+| Prompts | {{N}} | {{WHAT_SLASH_COMMANDS}} |
+
+## Slash commands
+
+| Command | Use when |
+|:---|:---|
+| `/{{command-1}}` | {{ONE_LINE_TRIGGER}} |
+| `/{{command-2}}` | {{ONE_LINE_TRIGGER}} |
+| `/{{command-3}}` | {{ONE_LINE_TRIGGER}} |
+
+*Full prompt reference: see [PROMPTS.md](./PROMPTS.md)*
+
+## Out of scope
+
+<!--
+Be explicit about what this spec does NOT cover, especially where adjacent specs exist.
+This prevents overlap and helps users find the right spec.
+-->
+
+- {{OUT_OF_SCOPE_ITEM_1}} → handled by `{{other-spec}}/`
+- {{OUT_OF_SCOPE_ITEM_2}} → handled by `{{other-spec}}/`
+
+## Related specs
+
+- [`../{{adjacent-spec-1}}/`](../{{adjacent-spec-1}}/) — {{one_line_relationship}}
+- [`../{{adjacent-spec-2}}/`](../{{adjacent-spec-2}}/) — {{one_line_relationship}}

package/areas/template/prompt.tmpl.md

@@ -0,0 +1,101 @@
+---
+workflow: {{workflow-name}}
+---
+
+# Prompt: `/{{command-name}}`
+
+<!--
+AGENT INSTRUCTIONS:
+1. "workflow-name" must match the workflow file stem exactly (if a workflow exists).
+2. "Use when:" is ONE sentence describing the exact scenario.
+3. Write 2–3 examples minimum. Example 1 = standard case. Example 2 = complex/edge case.
+4. Every example MUST have both EN and RU blocks, semantically identical.
+5. Examples must contain NO placeholders like [YOUR_PRODUCT] or {{INSERT_URL}}.
+6. Use realistic values: real tool names, realistic metrics, realistic company names.
+7. EN block should be ≥ 200 words total across all fields to give sufficient context.
+8. Delete all AGENT INSTRUCTIONS comments before finalising.
+-->
+
+Use when: {{SPECIFIC_SCENARIO — one sentence, e.g. "setting up CI/CD for a new monorepo that uses pnpm workspaces and deploys to AWS ECS."}}
+
+---
+
+## Example 1 — {{STANDARD_CASE_NAME}}
+
+**EN:**
+```
+/{{command-name}}
+
+{{FIELD_1_LABEL}}: {{REALISTIC_VALUE}}
+{{FIELD_2_LABEL}}: {{REALISTIC_VALUE}}
+{{FIELD_3_LABEL}}: {{REALISTIC_VALUE}}
+
+{{ADDITIONAL_CONTEXT — constraints, existing setup, non-goals}}
+
+{{OUTPUT_SPECIFICATION — what exactly should be delivered, in what format}}
+```
+
+**RU:**
+```
+/{{command-name}}
+
+{{FIELD_1_LABEL_RU}}: {{REALISTIC_VALUE_RU}}
+{{FIELD_2_LABEL_RU}}: {{REALISTIC_VALUE_RU}}
+{{FIELD_3_LABEL_RU}}: {{REALISTIC_VALUE_RU}}
+
+{{ADDITIONAL_CONTEXT_RU}}
+
+{{OUTPUT_SPECIFICATION_RU}}
+```
+
+---
+
+## Example 2 — {{COMPLEX_OR_EDGE_CASE_NAME}}
+
+**EN:**
+```
+/{{command-name}}
+
+{{FIELD_1_LABEL}}: {{DIFFERENT_REALISTIC_VALUE}}
+{{FIELD_2_LABEL}}: {{DIFFERENT_REALISTIC_VALUE}}
+
+{{MORE_SPECIFIC_CONSTRAINTS_OR_CONTEXT}}
+
+{{OUTPUT_SPECIFICATION}}
+```
+
+**RU:**
+```
+/{{command-name}}
+
+{{FIELD_1_LABEL_RU}}: {{DIFFERENT_REALISTIC_VALUE_RU}}
+{{FIELD_2_LABEL_RU}}: {{DIFFERENT_REALISTIC_VALUE_RU}}
+
+{{MORE_SPECIFIC_CONSTRAINTS_RU}}
+
+{{OUTPUT_SPECIFICATION_RU}}
+```
+
+---
+
+## Example 3 — {{QUICK_OR_CONTRASTING_CASE}} *(optional)*
+
+<!--
+Use only when there is a meaningfully different usage mode:
+a minimal/quick version, a contrasting domain context, or a specific tool variant.
+Remove this section entirely if 2 examples are sufficient.
+-->
+
+**EN:**
+```
+/{{command-name}}
+
+{{CONTENT}}
+```
+
+**RU:**
+```
+/{{command-name}}
+
+{{CONTENT_RU}}
+```

package/areas/template/rule.tmpl.md

@@ -0,0 +1,71 @@
+# Rule: {{CONSTRAINT_DOMAIN_NAME}}
+
+<!--
+AGENT INSTRUCTIONS:
+1. Replace {{CONSTRAINT_DOMAIN_NAME}} with a noun phrase: "Brand Voice Standards", "Migration Safety Requirements".
+2. Set Priority to P0 or P1 (see definitions below).
+3. Write every constraint as imperative: "must", "never", "required", "forbidden".
+4. Never use: "consider", "try", "ideally", "should" — these are advice, not rules.
+5. Include at least 1 compliant and 1 non-compliant example per section.
+6. Target: under 150 lines.
+7. Delete all AGENT INSTRUCTIONS comments before finalising.
+-->
+
+**Priority**: {{P0_OR_P1}} — {{ONE_SENTENCE: what non-compliance causes}}
+
+<!--
+P0 = Non-compliance blocks the task. Agent must refuse to proceed until resolved.
+     Use for: legal risk, security violation, data loss, brand damage requiring immediate fix.
+P1 = Non-compliance triggers a review flag. Agent continues but marks the item for human review.
+     Use for: quality risk, process deviation, inconsistency that humans should assess.
+-->
+
+---
+
+## {{SECTION_1 — group related constraints under one coherent concern}}
+
+<!--
+Example section names: "Tone and Voice", "Legal Disclaimers", "Secret Handling", "Migration Safety"
+-->
+
+1. **{{CONSTRAINT_NAME}}**
+   - {{SPECIFIC_REQUIREMENT — imperative statement}}
+   - {{SUB_REQUIREMENT_OR_EXCEPTION_IF_ANY}}
+
+2. **{{CONSTRAINT_NAME}}**
+   - {{REQUIREMENT}}
+   - Exception: {{EXCEPTION_CONDITION_IF_ANY}}
+
+## {{SECTION_2}}
+
+3. **{{CONSTRAINT_NAME}}**
+   - {{REQUIREMENT}}
+
+4. **{{CONSTRAINT_NAME}}**
+   - {{REQUIREMENT}}
+
+---
+
+## Compliant examples
+
+✅ {{REAL_EXAMPLE_OF_CORRECT_OUTPUT_OR_BEHAVIOR}}
+
+✅ {{ANOTHER_COMPLIANT_EXAMPLE}}
+
+## Non-compliant examples
+
+❌ {{REAL_EXAMPLE_OF_VIOLATION}} — {{BRIEF_REASON}}
+
+❌ {{ANOTHER_VIOLATION}} — {{REASON}}
+
+---
+
+<details>
+<summary>Rationale</summary>
+
+{{WHY_THIS_RULE_EXISTS — 2 to 4 sentences}}
+
+Regulatory / legal basis (if applicable): {{REGULATION_OR_POLICY_REFERENCE}}
+
+Last reviewed: {{DATE_OR_LEAVE_BLANK}}
+</details>

package/areas/template/skill.tmpl.md

@@ -0,0 +1,108 @@
+---
+name: {{kebab-case-skill-name}}
+type: skill
+description: {{ONE_SENTENCE_STARTING_WITH_VERB — e.g. "Design and execute conversion-focused landing page copy."}}
+related-rules:
+  - {{rule-filename-1.md}}
+  - {{rule-filename-2.md}}
+allowed-tools: {{COMMA_SEPARATED — Read, Write, Edit, Bash, WebSearch}}
+---
+
+<!--
+AGENT INSTRUCTIONS:
+1. Fill all frontmatter fields before writing the body.
+2. "name" must match the directory name exactly.
+3. "description" starts with a verb; states what the agent CAN DO after loading this skill.
+4. "related-rules" lists only the rule files that constrain how this skill is applied.
+5. "allowed-tools" lists only tools actually needed — don't include all five by default.
+6. Target: 150–350 lines total. Shorter = stub. Longer = split into two skills.
+7. All examples must be REAL and runnable — no placeholders like [YOUR_VALUE].
+8. Every skill needs at least 2 substantive sections and a "Common mistakes" section.
+9. Delete all AGENT INSTRUCTIONS comments before finalising.
+-->
+
+# Skill: {{DISPLAY_NAME}}
+
+> **Expertise:** {{COMMA_SEPARATED_LIST_OF_SPECIFIC_TECHNIQUES_TOOLS_OR_CONCEPTS}}
+
+## When to load
+
+<!--
+Critical section — the agent reads this to decide whether to load the skill.
+Write precise triggers. Vague triggers waste context.
+
+Bad:  "When working in this spec."
+Good: "When writing a new dbt model that joins more than two source tables and requires incremental materialization."
+-->
+
+Load this skill when:
+- {{SPECIFIC_TRIGGER_1 — describe the exact task condition}}
+- {{SPECIFIC_TRIGGER_2}}
+- {{SPECIFIC_TRIGGER_3}}
+
+Do NOT load for: {{WHAT_LOOKS_SIMILAR_BUT_DOES_NOT_NEED_THIS_SKILL}}
+
+---
+
+## {{MAIN_CONCEPT_OR_FRAMEWORK_SECTION}}
+
+<!--
+Explain the core technique, framework, or pattern set this skill covers.
+For technical domains: use runnable code or commands.
+For process domains: use named methodologies with real document snippets.
+For creative domains: use named frameworks (AIDA, PAS) with real examples.
+-->
+
+{{EXPLANATION_IN_2_TO_4_SENTENCES}}
+
+### {{PATTERN_OR_TECHNIQUE_1}}
+
+{{DESCRIPTION}}
+
+```{{language}}
+{{REAL_WORKING_EXAMPLE — not a placeholder}}
+```
+
+### {{PATTERN_OR_TECHNIQUE_2}}
+
+{{DESCRIPTION}}
+
+```{{language}}
+{{REAL_WORKING_EXAMPLE}}
+```
+
+### {{PATTERN_OR_TECHNIQUE_3}}
+
+{{DESCRIPTION}}
+
+---
+
+## {{SECOND_TECHNIQUE_OR_TOOL_SECTION}}
+
+<!--
+Add a second section for a related technique, tool operations, quality criteria, or decision framework.
+Every skill must have at least 2 substantive sections.
+-->
+
+{{CONTENT}}
+
+---
+
+## {{OPTIONAL_THIRD_SECTION — e.g. "Quality Checklist", "Decision Framework", "Common Scenarios"}}
+
+{{CONTENT}}
+
+---
+
+## Common mistakes
+
+<!--
+3–5 items. State the mistake, then the correction. Keep each under 2 sentences.
+Write from experience — what do agents (or humans) actually get wrong here?
+-->
+
+1. **{{MISTAKE_1}}** — {{CORRECTION}}
+2. **{{MISTAKE_2}}** — {{CORRECTION}}
+3. **{{MISTAKE_3}}** — {{CORRECTION}}
+4. **{{MISTAKE_4_OPTIONAL}}** — {{CORRECTION}}
+5. **{{MISTAKE_5_OPTIONAL}}** — {{CORRECTION}}

package/areas/template/workflow.tmpl.md

@@ -0,0 +1,104 @@
+---
+name: {{kebab-case-workflow-name}}
+type: workflow
+trigger: /{{workflow-name}}
+description: {{ONE_SENTENCE — what completing this workflow produces}}
+inputs:
+  - {{input_1 — name and description of required input}}
+  - {{input_2}}
+outputs:
+  - {{output_1 — concrete deliverable, e.g. "signed-off implementation_plan.md"}}
+  - {{output_2}}
+roles:
+  - {{role-1 — function name, e.g. "copywriter", "campaign-manager"}}
+  - {{role-2}}
+execution:
+  initiator: {{role — must be one of: product-owner | pm | team-lead | developer | qa | designer}}
+related-rules:
+  - {{rule-filename.md}}
+uses-skills:
+  - {{skill-directory-name}}
+quality-gates:
+  - {{MEASURABLE_CRITERION_1 — checkable without human judgment}}
+  - {{MEASURABLE_CRITERION_2}}
+---
+
+<!--
+AGENT INSTRUCTIONS:
+1. Fill all frontmatter fields. Missing fields cause the workflow to be skipped.
+2. "trigger" must start with / and match the prompt filename exactly.
+3. "roles" are who performs steps — use function names, not seniority titles.
+4. "execution.initiator" must be one of the six standard roles.
+5. "quality-gates" must be objectively checkable. "Looks good" is NOT a gate.
+   Good gate: "All copy variants pass brand-voice rule 2 and score ≥ 70 on Flesch-Kincaid."
+6. Every step MUST have: @role, Input, Actions (specific), Done when (checkable criterion).
+7. Use imperative voice: "Create", "Check", "Verify", "Ask" — not "You should" or "Consider".
+8. Include a failure path for at least one step.
+9. Target: 60–200 lines total. Over 200 lines = split into two workflows.
+10. Delete all AGENT INSTRUCTIONS comments before finalising.
+-->
+
+## Steps
+
+### 1. {{STEP_NAME}} — `@{{role-1}}`
+
+- **Input:** {{what arrives at this step — be specific}}
+- **Actions:**
+
+  {{SPECIFIC_ACTION_1 — include the actual command, query, template, or decision the agent should use}}
+
+  {{SPECIFIC_ACTION_2}}
+
+  > **If {{FAILURE_CONDITION}}:** {{EXPLICIT_FAILURE_PATH — what to do instead, not just "handle the error"}}
+
+- **Done when:** {{CHECKABLE_CRITERION — e.g. "brief confirmed in writing", "all required fields in the template are non-empty"}}
+
+---
+
+### 2. {{STEP_NAME}} — `@{{role-1}}`
+
+- **Input:** output of step 1
+- **Actions:**
+
+  {{ACTIONS}}
+
+- **Done when:** {{CRITERION}}
+
+---
+
+### 3. {{STEP_NAME}} — `@{{role-2}}`
+
+<!--
+Steps involving a different role must state the handoff explicitly.
+Example: "Receive draft from @developer. Review against architecture.md constraints."
+-->
+
+- **Input:** output of step 2
+- **Actions:**
+
+  {{ACTIONS}}
+
+- **Done when:** {{CRITERION}}
+
+---
+
+### 4. {{STEP_NAME}} — `@{{role-1}}`
+
+<!--
+Add or remove steps as needed. Most workflows are 3–7 steps.
+Fewer than 3 steps = write a skill instead.
+More than 8 steps = split into two workflows.
+-->
+
+- **Input:** output of step 3
+- **Actions:**
+
+  {{ACTIONS}}
+
+- **Done when:** {{CRITERION}}
+
+---
+
+## Exit
+
+{{ONE_SENTENCE: when the workflow is complete and what was produced.}}

package/bin/agentic.js

@@ -0,0 +1,24 @@
+#!/usr/bin/env node
+
+const { spawnSync } = require('node:child_process');
+const path = require('node:path');
+
+const scriptPath = path.resolve(__dirname, '..', 'agentic');
+const args = process.argv.slice(2);
+
+const result = spawnSync('bash', [scriptPath, ...args], {
+  stdio: 'inherit',
+  env: process.env,
+});
+
+if (result.error) {
+  if (result.error.code === 'ENOENT') {
+    console.error('[agentic][error] bash is required to run the agentic CLI from npx.');
+    process.exit(127);
+  }
+
+  console.error(`[agentic][error] failed to launch CLI: ${result.error.message}`);
+  process.exit(1);
+}
+
+process.exit(result.status ?? 1);

package/extensions/antigravity/GEMINI.md

@@ -0,0 +1,10 @@
+# AI Instructions
+
+Before doing any work in this repository, read the file **AGENTS.md**.
+
+AGENTS.md defines:
+- agent roles
+- development workflow
+- rules for modifying the repository
+
+Always follow the instructions from AGENTS.md.

package/extensions/claude/CLAUDE.md

@@ -0,0 +1,10 @@
+# AI Instructions
+
+Before doing any work in this repository, read the file **AGENTS.md**.
+
+AGENTS.md defines:
+- agent roles
+- development workflow
+- rules for modifying the repository
+
+Always follow the instructions from AGENTS.md.