Init

Ecosystem Initialization (Init Skill)

NOTE: This is an auto-evolution skill designed to run when Agent Studio is first deployed into a new repository (e.g., when AGENTS.md does not yet exist).

When to Use This Skill

Use this skill when you first enter a new codebase, or when the user explicitly asks to "initialize the repository" or "generate AGENTS.md".

Work Procedure

This skill executes a rigorous 3-stage "Deep Ecosystem Evolution" pipeline.

Stage 1: Context Gathering & Generation

Analyze the repository root to understand the technology stack and architecture. Specifically, look for and read:

• README.md
• .cursorrules or .cursor/rules/
• .github/copilot-instructions.md
• Package manager files (package.json, requirements.txt, Cargo.toml, go.mod, etc.)
• Framework config files (tsconfig.json, next.config.js, vite.config.ts, etc.)

Generate AGENTS.md Create or update a centralized AGENTS.md file in the repository root containing explicit, localized instruction sets for future Agents working in this codebase. Include: exact test/build CLI commands, architecture notes, and environment quirks. Do not hallucinate support links. Do not include generic fluff. Show the proposed content to the user for confirmation if AGENTS.md already existed but is stale.

Stage 2: Capability Gap Analysis

Cross-reference the discovered tech stack and repository complexity against the current available global tools (using the .claude/CLAUDE.md matrix).

• If a required capability is missing (e.g., the repo is heavily dependent on a specific framework like PyO3, but we only have a generic python-engineer):
• Explicitly prompt the user to authorize invoking the agent-creator or skill-creator to generate bespoke, hyper-localized expert components (e.g., tensor-grep-rust-worker).

Stage 3: Targeted Staleness Audit

Identify ONLY the subset of pre-existing agents and skills that are mathematically applicable to this repository's stack (e.g., if it's a TS web app, target typescript, react, jest, frontend). Do not evaluate all 200+ unrelated framework assets.

For each applicable asset:

1. Execute node .claude/tools/cli/skill-freshness-report.cjs --name [asset-name] (or manually inspect its YAML frontmatter / git logs) to check its lastUpdated or createdAt timestamp.
2. If the applicable asset is > 30 days old, immediately alert the user and propose spawning an agent-updater or skill-updater to refresh its instruction context against modern best practices.

Example Handoff

{
  "salientSummary": "Initialized the repository by scanning package.json and README.md. Detected a Next.js frontend with a Go backend. Generated AGENTS.md with explicit pnpm and Go build commands, testing strategies, and a high-level component map.",
  "whatWasImplemented": "Created AGENTS.md in the repository root. Extracted 4 core architecture rules from .cursorrules. Verified that the test commands provided in AGENTS.md actually execute cleanly.",
  "verification": {
    "commandsRun": [
      {
        "command": "cat package.json | grep 'test'",
        "exitCode": 0,
        "observation": "Found vitest testing configuration."
      }
    ]
  }
}