Agent Ready Init

Agent-Ready Init — Greenfield Baseline (v2)

Give a new project an agent-ready foundation from day one: portable instructions, secret-hygiene defaults, and a documented safe-execution posture — so it starts ready instead of accumulating debt that /agent-ready fix later has to remediate. Minimal and opinionated, not a stack wizard.

Canonical rubric: .claude/skills/agent-ready/references/scoring.md — the 7 dimensions, sub-criterion ids, layer tags, score math. Do not re-derive numbers here.

Remediation registry: .claude/skills/agent-ready/references/remediation.md — keyed by the same ids. Init reuses fix's generators conceptually but emits a fixed opinionated baseline rather than gap-driven output.

Routed as /agent-ready init [path] [--agents ...].

Phase 0: ARGUMENTS

Parse $ARGUMENTS (order-independent):

• path — first non-flag token: target dir (default cwd). mkdir -p it if absent.
• --agents — target list (claude,codex,opencode,pi); drives bridges only. If absent, default portable (AGENTS.md only).

Phase 1: LIGHTLY DETECT & GUARD

1. Ecosystem (light): presence of pyproject.toml/setup.py (Python), package.json (Node), go.mod (Go), Cargo.toml (Rust), pom.xml/build.gradle (JVM). Unknown → generic placeholders + advice.
2. Env-var refs (for .env.example): python3 .claude/skills/agent-ready/scripts/secret_hygiene.py when available; else grep os.environ/os.getenv/process.env/env::var. None found → empty template.
3. Populated-project guard: if the dir already has source code, a build manifest with real deps, or existing instruction files (AGENTS.md/CLAUDE.md), it is not greenfield — stop and defer:

This directory looks like an existing project (found: pyproject.toml with deps, src/).
`init` scaffolds greenfield baselines and never overwrites. For an existing repo,
run `/agent-ready scan` then `/agent-ready fix` to remediate by impact.

A truly empty or near-empty dir (no source, no real deps) proceeds.

Phase 2: BASELINE FILE SET

Generate only missing files. Reuse fix's generators conceptually; keep content minimal but real.

| Artifact | Content | |---|---| | AGENTS.md (canonical, ** [--agents ...]` to compute and show the starting score:

## 🌱 Agent-Ready Init — Baseline Established

Scaffolded: AGENTS.md, CLAUDE.md (symlink), .env.example, docs/agent-execution.md,
.github/workflows/ci.yml, .pre-commit-config.yaml, specs/TEMPLATE.md
+ .gitignore secret coverage

Baseline score: 41/100  🟡 Partially Ready

Next: add deps + commit the lockfile, write tests, then `/agent-ready scan` to track progress.

Init vs Fix

| | init (greenfield) | fix (brownfield) | |---|---|---| | Trigger | New/empty project | Existing repo, after a scan | | Input | Light detection | agent-ready-scores.json gap analysis | | Scope | Fixed opinionated baseline | Highest-impact gaps first | | Output | Templated placeholders OK | Detected, project-specific content | | Overwrite | Never (mostly empty repo) | Never; append where sensible |

Critical Rules

• NEVER overwrite existing files — create only missing ones; append to .gitignore.
• Defer to fix on a populated existing project (Phase 1 guard).
• AGENTS.md is canonical & < 200 lines — bridge to claude via symlink; codex/opencode/pi read it natively.
• Safe by default — secret-hygiene .gitignore + documented execution policy (LINCE among options) ship from day one.
• Minimal opinionated — fixed baseline, at most confirm detected stack + targets; do NOT become a stack wizard.
• Confirmation gate — list files and wait for approval before writing.
• End with a quick scan showing the baseline score.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: RisorseArtificiali
• Source: RisorseArtificiali/agent-ready-skill
• License: MIT

Install and usage instructions live in the source repository linked above.