A curated catalog of reusable Agent Skills for Agent harness setup, requirements interviews, and Feishu/Lark operations.
Install only what adds value to your host or project. Ordinary investigation, testing, cleanup, review, and delivery use the host/project workflow; there is no mandatory skill chain. See the selection guide for route boundaries and retired installations.
Run from the consumer project, not this catalog checkout. These examples use the audited installer pin, not a claim about the latest version.
# One skill for Claude Code and Codex
npx --yes skills@1.5.17 add sean2077/skills --skill agent-scaffold -a claude-code -a codex
# Complete catalog for just these two targets
npx --yes skills@1.5.17 add sean2077/skills --skill '*' -a claude-code -a codex
# Local catalog source; replace the absolute path
npx --yes skills@1.5.17 add /absolute/path/to/skills --skill agent-scaffold -a codexRepeat --skill and -a for selective installs; quote '*'. Relative local sources need ./ or ../. Installer semantics covers discovery-only checks, global scope, and safe removal. Project-scope removal from this catalog checkout can delete product files.
Installing agent-scaffold makes the skill available; it does not initialize the consumer's harness. Its entry point starts with a read-only plan. Full setup adopts existing project guidance and records one-time convention exclusions in the managed AGENTS.md block; see onboarding selection. Each selected domain's generic guide is installed under .agents/conventions/ with a route from that block, so later Agents find it without the catalog skill. Host trust and hook approval remain separate from installation.
Each skill is independently installable. Its entry point owns the workflow and on-demand references.
| Skill | Use |
|---|---|
| agent-scaffold | Initialize or maintain the harness and selected project conventions; preserve existing layouts and remember one-time exclusions. |
| deep-interview | Resolve requirements into an approved specification; exact-file approval records are optional. |
| lark-cli | Perform selected 飞书/Feishu/Lark CLI operations with identity and side-effect safeguards. |
Project skill skill-eval and subagent skill-verifier support this repository's evaluations. Neither is a catalog install target.
| Need | Start here |
|---|---|
| Choose routes or remove retired installations | Selection and composition |
| Find the source to edit | Architecture and ownership |
| Contribute, verify, regenerate, or release | Development |
| Check installer behavior, host trust, and dated evidence | Compatibility |
| Maintain docs or design a skill | Documentation maintenance · Harness principles |
| Evaluate decisions or actual task outcomes | Routing probes · Task outcomes |
| Work as a repository Agent | AGENTS.md · Terminology |
| Review changes and past decisions | CHANGELOG.md · Historical records |
MIT. Preserve skill-specific attribution notices when redistributing payloads.