agent-native-hardening

name	agent-native-hardening
description	TypeScript/JavaScript codebase hardening: maintainability, code quality, architecture cleanup, repo scoring, agent-friendly structure, godfiles, feature folders, DRY, type safety, traversability, feedback loops, worktrees, subagents, swarm refactors.

Agent Native Hardening

Use this skill when the user asks to review and improve a codebase for agent-native maintainability, especially requests like:

• "score this repo"
• "make this codebase agent-friendly"
• "improve code quality / maintainability"
• "clean up the architecture"
• "refactor with swarm/worktrees"
• "improve quality loops and structure"

Must-Read References

Read these supporting files before applying the skill, not after improvising the workflow:

• references/scoring-rubric.md before any scorecard, findings list, or severity ranking.
• references/swarm-lanes.md before planning discovery lanes, implementation lanes, worktrees, or subagent splits.

If a reference file is missing or unreadable, say so and continue with the closest fallback, but do not silently skip it.

Core Principles

1. Fail fast. Do not hide errors.
2. Prefer in-code discoverability over markdown sprawl.
3. Keep tests light and deterministic; avoid flaky integration tests unless requested.
4. Use lanes to separate evidence gathering and implementation. A lane can be a read-only exploration task, a direct coding pass, a subagent task, or a worktree branch depending on scope and risk.
5. Keep each lane focused, low-overlap, and easy to verify.
6. Prefer feature folders over catch-all files; godfiles must be extracted into clear feature-owned modules.
7. Push toward DRY and separation of concerns; remove copy-paste and mixed-responsibility modules without replacing them with new junk drawers.

Required Scorecard