inits
npx machina-cli add skill OutlineDriven/odin-claude-plugin/init --openclawInit - AGENTS.md Generator
Analyze this codebase and create or improve an AGENTS.md file for future ODIN Code Agent instances.
Core principle
Only encode knowledge that is expensive to rediscover. An agent can fd, rg, ast-grep, and read files in seconds. If the information is one search away, omit it.
Objective guard
AGENTS.md is for priming and guiding future agents with conventions, constraints, and rationale. It is not a repository summary.
What to include
- Cross-cutting conventions and boundaries that span multiple files.
- Implicit contracts and operational constraints that are easy to miss in quick code exploration.
- Non-obvious command guidance only when hidden constraints matter (required ordering, flags, env vars, caveats).
- For each included item, state:
- the rule or constraint
- why it exists (rationale)
What to omit (the agent can discover these)
- File trees, per-directory descriptions, or component inventories.
- Dependency/version tables.
- Generic development best practices.
- Information that is easily discoverable via search/read.
- Fabricated filler sections such as "Common Development Tasks", "Tips", or "Support".
- Any summary-style prose that restates code organization without conventions or rationale.
Workflow
- If
AGENTS.mdalready exists, suggest targeted improvements instead of rewriting. - Do not repeat yourself. Each fact appears once.
- Every statement must be grounded in files you actually read. Never invent.
- If uncertain, omit the claim rather than speculate.
Source
git clone https://github.com/OutlineDriven/odin-claude-plugin/blob/main/skills/init/SKILL.mdView on GitHub Overview
Analyzes a codebase to generate or enhance AGENTS.md for future ODIN Code Agent instances. It records cross-cutting conventions, implicit contracts, and operational constraints that are hard to rediscover, aiding onboarding and repeatability.
How This Skill Works
The tool scans the repository to surface non-obvious rules and constraints that justify their inclusion with rationale. It avoids duplicating easily discoverable information, and if AGENTS.md already exists, it suggests targeted improvements rather than a full rewrite, grounding every item in observed files.
When to Use It
- Onboarding a new repository to capture hard-to-rediscover conventions and constraints.
- After a codebase evolution to preserve tacit rules and rationale.
- When AGENTS.md exists but needs targeted improvements instead of a full rewrite.
- During ODIN Code Agent onboarding to ensure consistent agent behavior across instances.
- To prevent loss of non-obvious guidance when exploring codebases quickly.
Quick Start
- Step 1: Run the generator on the target repository to produce a draft AGENTS.md.
- Step 2: Review items for discoverability and remove anything easily found by search/read.
- Step 3: Commit AGENTS.md and reference it in onboarding materials or agent initialization.
Best Practices
- Ground every item in actual files read; do not invent or guess.
- For each item, include both the rule/constraint and its rationale.
- Omit easily discoverable information and avoid restating code structure as summary.
- If AGENTS.md exists, propose targeted improvements rather than rewriting it wholesale.
- Maintain concise, non-duplicated statements; each fact appears only once.
Example Use Cases
- Rule: Conventions for cross-tool invocation order (fd, rg, ast-grep) across files | Rationale: Ensures consistent agent behavior when re-running analyses.
- Rule: Implicit contracts require specific environment variables for agent operations | Rationale: Prevents silent failures due to missing context.
- Rule: Non-obvious command constraints (required flags, ordering) | Rationale: Avoids violating hidden constraints that affect reproducibility.
- Rule: Do not include file-tree or generic development best practices | Rationale: Keeps AGENTS.md focused on conventions and rationale.
- Rule: If AGENTS.md exists, propose improvements rather than rewriting | Rationale: Preserves history and minimizes noise.
