init

name	init
description	Scaffold the full Polisade Orchestrator framework (.state, docs, templates, backlog, CLAUDE.md) into the current project. Use when PM mentions "init project", "initialize", "scaffold project", "set up Polisade", "bootstrap project", "инициализируй проект", "разверни структуру", or any request to bring up the autonomous development structure in a fresh or existing repository.
argument-hint	[ProjectName]
cli_requires	task_tool

/polisade:init — Initialize Polisade Orchestrator project structure

Scaffold the full Polisade Orchestrator autonomous development framework into the current project.

Usage

/polisade:init MyProject
/polisade:init              # uses current directory name

What it creates

.state/
├── PROJECT_STATE.json
├── counters.json
└── knowledge.json

docs/
├── prd/
├── specs/
├── plans/
├── adr/
├── architecture/   # design packages from /polisade:design
├── contracts/
│   ├── README.md       # integration map
│   ├── provided/       # contracts we provide (OpenAPI, AsyncAPI, Protobuf…)
│   └── consumed/       # contracts we consume from external systems (read-only)
└── templates/
    ├── prd-template.md
    ├── spec-template.md
    ├── plan-template.md
    ├── feature-brief-template.md
    ├── task-template.md
    ├── adr-template.md
    ├── chore-template.md
    ├── spike-template.md
    └── design-package-template.md

backlog/
├── features/
├── bugs/
├── tech-debt/
├── chores/
└── spikes/

tasks/

CLAUDE.md
.claude/settings.json
.gitignore (updated)

Algorithm

1. If $ARGUMENTS is provided, use it as project name. Otherwise use the current directory name.

2. Check if Polisade Orchestrator is already initialized:

   • If .state/PROJECT_STATE.json exists → warn and ask to confirm overwrite

3. Create directory structure:

   mkdir -p .state
   mkdir -p docs/prd docs/specs docs/plans docs/architecture docs/architecture/decisions docs/contracts/provided docs/contracts/consumed docs/templates
   mkdir -p backlog/features backlog/bugs backlog/tech-debt backlog/chores backlog/spikes
   mkdir -p tasks
   mkdir -p .claude
   # Git does not track empty directories — without a placeholder, `git add` silently
   # drops this scaffold and downstream skills (defect/feature/spec/...) then fail to
   # place BUG-*/FEAT-*/SPEC-*/... artifacts in their canonical dirs (#122). The
   # .gitkeep is harmless and goes away naturally once a real artifact lands.
   touch docs/prd/.gitkeep docs/specs/.gitkeep docs/plans/.gitkeep docs/architecture/.gitkeep docs/architecture/decisions/.gitkeep
   touch backlog/features/.gitkeep backlog/bugs/.gitkeep backlog/tech-debt/.gitkeep backlog/chores/.gitkeep backlog/spikes/.gitkeep
   touch tasks/.gitkeep
   

4. Copy template files from plugin templates directory.

   ⛔ INVARIANT: target files MUST be byte-identical to the source templates, except for the explicit substitutions in step 6 (project.name in two JSON fields). Do NOT reconstruct, regenerate, paraphrase, summarize, or "improve" template content. If a Read fails, STOP and report — do NOT proceed by reconstructing from memory. Issue #119: under GigaCode Filesystem Guard the install dir is read-protected; the converter substitutes inline canonical content for Qwen/GigaCode builds (see the INLINE TEMPLATES block below) so the Write tool always has byte-identical bytes regardless of Guard state.

   The plugin templates are located at: skills/init/templates/ relative to the plugin root. Use the Read tool to read each template file from the plugin directory, then use Write tool to write it to the target project.

   Files to copy:

   • skills/init/templates/PROJECT_STATE.json → .state/PROJECT_STATE.json
   • skills/init/templates/counters.json → .state/counters.json
   • skills/init/templates/knowledge.json → .state/knowledge.json (update project name)
   • skills/init/templates/settings.json → .claude/settings.json
   • skills/init/templates/CLAUDE.md → CLAUDE.md
   • skills/init/templates/docs/*.md → docs/templates/*.md (all 9 templates)
   • skills/init/templates/docs/contracts-readme-template.md → docs/contracts/README.md
   • Create empty docs/contracts/provided/.gitkeep
   • Create empty docs/contracts/consumed/.gitkeep

5. Update .gitignore — append Polisade Orchestrator-specific entries if not already present:

   # Polisade Orchestrator state (regenerated by Claude)
   .state/
   !.state/knowledge.json
   
   # Polisade Orchestrator worktrees (git worktree isolation)
   .worktrees/
   
   # Polisade intermediate artefacts — project-local temp dir
   # (issue #57 / legacy OPS-009; /tmp is sandboxed in GigaCode CLI).
   .polisade/tmp/
   
   # CLI extensions — service dirs (issue #74 / OPS-027).
   # NB: .claude/ intentionally excluded — settings.json is committed.
   .gigacode/
   .qwen/
   .codex/
   

   Append is idempotent — each line is added only if not already present in .gitignore.

6. Update knowledge.json with the project name from arguments.

6.5. Update PROJECT_STATE.json — set project.name from the project name argument.

6.6. Tech stack autodetect — scan project root for stack markers and pre-fill knowledge.json:

Markers to check:

• package.json → Node.js/TypeScript (check dependencies/devDependencies)
• tsconfig.json → TypeScript (even without package.json, e.g. Deno)
• go.mod → Go
• pyproject.toml / requirements.txt / setup.py → Python
• Cargo.toml → Rust
• pom.xml / build.gradle / build.gradle.kts → Java/Kotlin
• build.sbt / .scalafmt.conf → Scala/sbt
• gradlew / mvnw → JVM wrapper scripts (Gradle/Maven)
• application.yml / application.properties → Spring Boot
• *.csproj / *.sln → C# / .NET
• docker-compose.yml → infrastructure hints (DB, cache, queue, Kafka)
• .env.example → environment variables
• Makefile / Justfile → build/test commands
• jest.config.* / vitest.config.* / pytest.ini / .rspec → test framework
• playwright.config.* → Playwright (E2E)
• cucumber.yml / features/*.feature → Cucumber (BDD)

For each detected marker, update knowledge.json:

• projectContext.techStack — append detected languages, frameworks, databases, infra
• testing.testCommand — infer test command (e.g. npm test, ./gradlew test, sbt test, npx playwright test, npx cucumber-js)
• testing.lintCommand — infer lint command if detectable (e.g. npx eslint ., ./gradlew check, sbt scalafmtCheck)

This is best-effort: if nothing is detected, leave fields empty — /polisade:spec and /polisade:design will ask during their interview.

6.7. VCS provider setup — ask the PM which VCS hosts the project:

Options:

• github — GitHub cloud via gh CLI (default).
• bitbucket-server — self-hosted Bitbucket Server (corporate, via REST API).

Record the choice in .state/PROJECT_STATE.json → settings.vcsProvider.

If the PM selects bitbucket-server:

a) Copy skills/init/templates/env.example to the project root as the single target .env.example — a committable reference. Do NOT create .env itself: under GigaCode Filesystem Guard the tool-layer write of a target-project .env is hard-denied (issue #128). The PM creates .env manually (sub-step c).

b) Append an uncommented .env line to .gitignore if not already present. Match regex ^\s*\.env(\s|$) on each non-# line to avoid false positives from pre-existing # .env template comments. If no match, append: # VCS provider (.env contains Bitbucket tokens) .env

c) Remind the PM (the agent must NOT attempt the copy itself — Filesystem Guard blocks autowrite of .env): ⚠️ Создай .env вручную (Filesystem Guard блокирует автозапись): cp .env.example .env затем заполни BITBUCKET_DOMAIN1_URL / BITBUCKET_DOMAIN1_TOKEN (и/или DOMAIN2). Проверь через /polisade:doctor.

This logic is duplicated in scripts/polisade_migrate.py (compute_vcs_bootstrap_migrations) so that existing projects can switch to Bitbucket later via /polisade:migrate --apply.

6.8. Verify the generated structure in a clean context. After all files are created, invoke polisade-init-verify in a clean-context subagent so the checks run without this command's reconstruction-prone context. Under GigaCode the subagent_type: line is stripped by the converter and routing happens by the skill name polisade-init-verify:

Task tool:
  subagent_type: "general-purpose"
  description: "Verify polisade:init output"
  prompt: "Run the polisade-init-verify checks against the current project; report PASS/FAIL."

Contract — fail loud: if verify reports FAIL, STOP. Print the FAIL reasons verbatim and do NOT print the INITIALIZED banner. A FAIL means the generated files were reconstructed from memory rather than written from the canonical bytes (issue #128 / #119) — re-run /polisade:init, do not report success. Only proceed to step 7 when verify reports PASS.

7. Output confirmation:

═══════════════════════════════════════════
Polisade Orchestrator INITIALIZED
═══════════════════════════════════════════

Project: {ProjectName}
Directory: {cwd}

Created:
  ✓ .state/ — project state files
  ✓ docs/templates/ — 9 document templates
  ✓ docs/architecture/ — design packages (created on demand by /polisade:design)
  ✓ docs/contracts/ — integration contracts (provided/consumed)
  ✓ backlog/ — features, bugs, tech-debt, chores, spikes
  ✓ tasks/ — work items
  ✓ CLAUDE.md — framework instructions
  ✓ .claude/settings.json — permissions

Detected stack:                        ← only if autodetect found anything
  • TypeScript 5, Express 4
  • PostgreSQL 16, Redis 7
  • Jest (unit), Playwright (E2E)
  (saved to .state/knowledge.json)

═══════════════════════════════════════════
NEXT STEPS:
   → /polisade:feature "Description" — add a feature
   → /polisade:prd "Name" — create PRD for large initiative
   → /polisade:state — view project status
   → /polisade:continue — start autonomous work
═══════════════════════════════════════════

If running under GigaCode / Qwen CLI, also print this reminder (#121):

⚠️  Do NOT commit `.gigacode/settings.json` / `.qwen/settings.json` — the CLI's
    auto-permission-prompt writes absolute paths containing your corp user-id.
    These files are already in `.gitignore`; never `git add -f` them.
    Run `/polisade:doctor` to flag any polluted permission entries.