X9 Context Files Generator

Repository context files

Create concise, evidence-backed onboarding files for two audiences: agents (AGENTS.md/CLAUDE.md) and humans (README.md). Existing files are user-owned inputs, not blank templates.

Select the action

• Create or update: an explicit request to create, update, fix, or rewrite a context file authorizes the requested in-scope edits; do not ask for the same permission again.
• Audit: an explicit audit/check request is read-only. Report findings and proposed changes, then ask once whether to apply them.
• Bare skill invocation: inspect the current repository read-only by default. If changes would help, show the proposal and ask once before editing.

Infer the action from the whole request, not only trigger words. When an authorized edit reveals an additional normalization outside the requested change, ask once before applying that extra change.

Workflow

1. Read current repository instructions and README completely enough to preserve non-inferable rules. Inspect actual commands, manifests, CI, and representative code before writing.
2. Classify content by audience:

• Agent file: commands, dangerous gotchas, local constraints, non-inferable conventions, and verification contracts.
• README: purpose, setup, normal usage, and links to deeper material.

1. Keep one canonical owner for each fact. A short audience-specific summary is allowed when it changes behavior; link to the owner instead of copying full detail.
2. Make root AGENTS.md the single canonical local instruction file and root CLAUDE.md exactly @AGENTS.md plus a final newline, unless the repository or user declares an exception. Put every new local rule in AGENTS.md. Before normalizing an existing CLAUDE.md, merge its unique rules into AGENTS.md; never discard them silently or preserve a duplicate copy by default. Apply this normalization immediately when it was explicitly requested or already required by repository policy; otherwise propose it and ask once before editing.
3. Treat an existing AGENTS.md, CLAUDE.md, or README.md as merge-only unless the user explicitly requests replacement. Show a diff for substantive rewrites.
4. Include architecture only when the rationale or boundary cannot be recovered cheaply from code and materially affects decisions. Avoid generated file trees and generic overviews.
5. Use English for machine-facing instructions by default; follow repository/user language when human maintenance or domain literals make that clearer.
6. Run commands you present only when they are safe, local, and within the requested repository scope. Read-only inspection and ordinary local validation are allowed by default. Destructive, external, costly, deployment, migration, production-data, account-mutating, or credential-changing commands require explicit user authority; otherwise preserve the command and label it unverified with the reason.

Read [references/agents-md.md](references/agents-md.md) when editing agent files, [references/readme.md](references/readme.md) for README work, and [references/research-basis.md](references/research-basis.md) when judging how much inferred architecture belongs in context.

Boundaries

• Global ~/.claude/CLAUDE.md and ~/.codex/AGENTS.md belong to x9-agent-instructions.
• Skill folders belong to x9-skill-creator.
• Do not put personal absolute paths such as /Users//... or /home//... in committed files. Prefer repository-relative paths or neutral placeholders; keep a machine-bound path only when the repository genuinely depends on that machine and the user explicitly wants it documented.
• Do not overwrite existing instructions by position from a stale read; re-read immediately before applying a patch.

Done

• Existing non-inferable rules are preserved or intentionally changed with user authority.
• When canonical-file normalization was authorized and no exception applies, CLAUDE.md contains only @AGENTS.md and all local instructions are owned by AGENTS.md.
• Commands and paths trace to the current repository.
• Agent and human files contain only behavior-changing audience-specific material.
• A fresh-context reader can start the project without inventing a command.

Source & license

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

• Author: xonika9
• Source: xonika9/agent-skills
• License: MIT
• Homepage: https://skills.sh/xonika9/agent-skills

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