code-authoring

name	code-authoring
description	Bounded code authoring and editing with source-first analysis, risk and execution gates, and validation truth. Use when the user asks to write, edit, refactor, or fix code in a repository.

code-authoring

Apply this skill for any explicit code-writing or code-editing request.

Required gate sequence before any file mutation:

1. Risk classification.
2. Intent and stage routing.
3. No-assumption contract.
4. Execution control.
5. Tool permission policy.
6. Human approval when the change is destructive.
7. Secret-aware behavior when credentials may be touched.

Authoring contract:

• Inspect the actual repository, language, framework, conventions, and existing tests before writing.
• Stay inside the requested scope. Do not silently refactor unrelated code, change deps, or touch generated files.
• Prefer minimal diffs unless a refactor is explicitly authorized.
• Write or update tests when behavior changes.
• Cite the source paths you read and the diffs you produced.
• Run validators that exist in the repo (lint, test, build, type-check); report each command and pass or fail.
• Never claim validation that did not actually run.

Diff-mode and patch-mode:

• Surgical edits → minimal-patch mode.
• Large changes → reduce diff first, classify risk, then proceed in slices.

Refactor flows:

• Pair with refactor-strict for planning. Execute only the bounded slice that was approved.

Hard stops:

• Destructive Git or release operation requested → escalate for human approval.
• Apparent secret value in code or input → switch to secret-aware credential handling.
• Implementation cannot be done without source the user did not provide → return a minimum-input request instead of guessing.

Output:

• Exact file paths changed.
• Behavior changed.
• Tests added or updated.
• Validation run (exact commands and exit status).
• Validation not run (with reason).
• Assumptions, blockers, and rollback notes.