ai-skills-init

name	ai-skills-init
description	Use this skill when bootstrapping a target repository to be ai-skills-aware — on the first run of any ai-skills workflow in a fresh repo, when adopting the ai-skills plugin in an existing repo, or after upgrading to a plugin version that adds new memory paths or templates, including when the user does not say "init" but asks to "set up" or "onboard" the repo — to detect codebase type, create CLAUDE.md + AGENTS.md scaffolding, initialize the .ai-skills-memory/ directory tree from L1 templates, and configure .gitignore. Idempotent — safe to re-run. Accepts `--codebase-type <type>` and `--overwrite`. Not for re-initializing only memory — use `/memory-init` instead.
context	fork
argument-hint	[--codebase-type <type>] [--overwrite]

/ai-skills-init — Bootstrap a Target Repository

One-time (or re-runnable) setup for a target repo. Detects codebase type, scaffolds CLAUDE.md + AGENTS.md, creates .ai-skills-memory/ tree from L1 templates, appends .gitignore rules. Idempotent.

When to use

• First run of any ai-skills workflow on a fresh repo
• Adopting the ai-skills plugin in an existing repo (new for the team)
• After upgrading the plugin to a version that adds new memory paths or templates

Not for

• Re-initializing only memory (use /memory-init for that)
• Modifying an existing CLAUDE.md (manual edit — this skill never overwrites without --overwrite)

Invocation

/ai-skills-init
/ai-skills-init --codebase-type python-flask
/ai-skills-init --overwrite                  # rare; only when CLAUDE.md is empty/stale

Arguments

Flag	Default	Effect
--codebase-type	auto-detect	python-flask, python-fastapi, nodejs-express, nodejs-nextjs, astro, sveltekit, remix, nodejs-bun, deno, java-spring, kotlin-spring, kotlin-ktor, elixir-phoenix, go, ruby-rails, rust, dotnet, mixed, generic
--overwrite	false	If CLAUDE.md already exists, OVERWRITE with fresh scaffold. Default: skip if exists

Output

• <repo>/CLAUDE.md — scaffolded with codebase type, empty sections for user
• <repo>/AGENTS.md — empty template (optional, for team customization of agents per-repo)
• <repo>/.ai-skills-memory/ directory tree (per /memory-init spec)
• <repo>/.gitignore — appended .ai-skills-memory/ exclusion rule (if not present)

Agent roster

Agent	Model	Effort	Tools	Role
scaffolder (internal — resolves to software-engineer with restricted tools)	haiku	low	Write, Read, Glob	Detects codebase type, generates scaffolds

Pipeline

┌─ Auto-detect codebase type:
│  └─ Check for: Pipfile, pyproject.toml, package.json, pom.xml,
│     build.gradle.kts, go.mod, Cargo.toml, Gemfile, mix.exs,
│     *.csproj, astro.config.mjs, svelte.config.js, remix.config.js,
│     bun.lockb / bunfig.toml, deno.json / deno.jsonc / deps.ts
│     Multiple matches → mixed; none → generic
│
├─ Generate CLAUDE.md scaffold:
│  └─ Sections: Overview, Tech Stack, Directory Layout, Key Decisions,
│     Constraints, Getting Started
│     Pre-fill: tech stack (detected), codebase type
│     Leave blank for user: business context, architectural decisions
│
├─ Generate (optional) AGENTS.md:
│  └─ List all plugin agents enumerated from `plugin/agents/`
│     (count auto-tracks future additions) with brief role descriptions
│     User can override / add per-repo customization
│
├─ Create .ai-skills-memory/ tree (delegate to /memory-init logic):
│  ├─ .gitignore (from plugin/memory/templates/ai-skills-memory.gitignore)
│  ├─ .committed/ subdir with README + allowlist-extensions
│  ├─ config.json (per-repo override stub for token caps + RALF caps)
│  ├─ learnings.md (empty template)
│  ├─ runs.jsonl, errors.log, redactions.log (touch-create)
│  └─ workflow subdirs: designs/, develop/, bugfix/, refactor/,
│     migrate/, spikes/, security-audits/, env-reports/, docs/
│
├─ Update root .gitignore:
│  └─ Add .ai-skills-memory/ rule (if not already present);
│     also add `.committed/` is tracked exception (negation rule)
│
└─ Print scaffold creation summary + next steps
   ("Fill business context in CLAUDE.md, run /feature-design or /develop")

No RALF — scaffolding is one-pass; idempotent so safe to re-run.

Codebase-type detection markers + template deltas

Type	Markers	CLAUDE.md template deltas
astro	astro.config.mjs; package.json has "astro" dep	Primary role: frontend-engineer. Note SSG/SSR/hybrid mode. Likely Tailwind / shadcn-ui. Playwright for E2E
sveltekit	svelte.config.js; package.json has "@sveltejs/kit" dep	Primary role: frontend-engineer. Note SSR/SSG modes. Tailwind / shadcn-ui likely. Playwright for E2E
remix	remix.config.js; package.json has "@remix-run/*" deps	Primary role: frontend-engineer. Note SSR-first full-stack. Tailwind / shadcn-ui likely. Playwright for E2E
nodejs-bun	bun.lockb or bunfig.toml	Use bunx over npx. Native TypeScript (no tsc step). Native test runner (bun test)
deno	deno.json / deno.jsonc; deps.ts	Use deno run / deno test. Built-in TS. Document the permissions model (--allow-net, --allow-read, …)
elixir-phoenix	mix.exs with :phoenix dep	Primary role: elixir-engineer (NOT software-engineer for stack-specific work). ExUnit for tests. mix release for deploy
kotlin-spring	build.gradle.kts + kotlin plugin + Spring deps	Primary role: java-engineer. Gradle Kotlin DSL. JUnit + Kotlin idioms
kotlin-ktor	build.gradle.kts + io.ktor deps	Primary role: java-engineer. Ktor coroutines model. H2 + Exposed common stack

Hard rules

• Never overwrite CLAUDE.md without --overwrite — refuse with: "CLAUDE.md exists; pass --overwrite to replace"
• pre-tool-use-committed-write.py hook applies to any .committed/ writes (committed-allowlist enforcement)
• Idempotent — re-running on a fully-set-up repo is a no-op + report

Eval rubric

Pointer: plugin/eval/judge-rubrics/ai-skills-init.md (B10).

Dimensions:

1. Correctness — scaffold matches detected codebase type
2. Completeness — all expected directories and files present
3. Clarity — placeholder comments are helpful
4. No conflicts — respects existing CLAUDE.md if --overwrite not set
5. Gitignore safety — no important files accidentally ignored

Pass: avg ≥ 4.0, no dimension < 3.

Memory writes

Layer	When	Shape
L4	After complete	.ai-skills-memory/init-summary.md — timestamp, plugin version, codebase type detected, files created/skipped

Failure modes

• Codebase type ambiguous (multiple stacks detected): scaffolder defaults to mixed; user can specify with --codebase-type
• CLAUDE.md exists and --overwrite not set: skip CLAUDE.md creation; report exists; continue with /memory-init portion
• Write permission denied: escalate to user with clear path; suggest checking repo writeability or running with elevated permissions
• Plugin not installed properly (${CLAUDE_PLUGIN_ROOT} empty): error with: "Plugin templates not found. Verify plugin install with /plugin status ai-skills"

Observability events

• workflow_start — ai-skills-init
• codebase_type_detected — detected type
• scaffold_created — files/dirs created (vs skipped)
• workflow_end — COMPLETE

Integration

• Reads templates from: ${CLAUDE_PLUGIN_ROOT}/memory/templates/ (7 files for memory skeleton). PII pattern file is at ${CLAUDE_PLUGIN_ROOT}/hooks/scripts/pii-patterns.txt (loaded by hooks, not copied during init); project extension at .ai-skills-memory/.committed/pii-patterns.txt (created on demand).
• Schemas: plugin/schemas/spawn-payload.schema.json (G7 — even though scaffolder is internal, payload still embedded for consistency)
• Sub-workflow: /memory-init (the memory portion of the bootstrap)
• Companions: /plugin-doctor (verify install before init), /context-load (after init, validate context loads correctly)
• Rules: memory-discipline (write rules per layer), untrusted-content-wrapping (G1 wrap on existing CLAUDE.md if --overwrite reads it for backup)
• Hooks: pre-tool-use-committed-write.py (committed-allowlist enforcement on .committed/ writes), session-start-context.py (will read newly-created CLAUDE.md on next session)