feat: skill content injection and smarter matching

[rubenmarcus]

Summary

Two improvements to the skills feature:

1. Skill content injection

Current: Skills are only listed by name/description in the agent prompt — the actual skill file content is never injected. The agent sees:

- **frontend-design**: Create distinctive, production-grade frontend interfaces

But never receives the skill's actual instructions.

Proposed: For auto-applied skills, inject the full skill file content into the agent prompt. The agent actually follows the instructions, not just sees the name.

File: src/loop/skills.ts:315-354 — formatSkillsForPrompt()

2. Smarter skill matching

Current: shouldAutoApplySkill() only matches web/design keywords (lines 293-313). Skills for testing, API, database, DevOps, etc. are never auto-applied.

Proposed: Expand keyword categories:

• Testing: test, tdd, spec, quality, coverage, jest, vitest, pytest
• API: api, rest, graphql, endpoint, route
• Database: database, sql, migration, schema, prisma
• DevOps: deploy, ci, cd, docker, pipeline
• Security: security, auth, csrf, xss, injection

File: src/loop/skills.ts — shouldAutoApplySkill()

Files

• src/loop/skills.ts