tkhwang's personal agent skills.
Each top-level folder is one skill (<name>/SKILL.md), usable by Claude Code, Codex, and other skill-aware agents.
| Skill | Description |
|---|---|
plan-decision-grill |
Stress-test a plan/design/PRD before coding by surfacing only the important decision gates. |
plan-execute |
Execute an approved plan step by step in auto or checkpoint mode, verifying build/type/lint/test results at each step. |
plan-decision-grill works standalone. Optionally, installing mattpocock/skills gives it grill-me and grill-with-docs to lean on for a sharper one-question-at-a-time interrogation:
npx skills add mattpocock/skills -s grill-me,grill-with-docs -gUse the skills CLI:
# all skills, global (user-level), all agents
npx skills add tkhwang/skills -g
# a specific skill only
npx skills add tkhwang/skills -s plan-decision-grill -g
# later, pull the latest version
npx skills updateThis symlinks the skills into your agent dirs (~/.claude/skills, ~/.codex/skills, …) and records them in .skill-lock.json. Pass --copy if you want copies instead of symlinks.
On the machine where you author these skills, don't use npx skills add — it copies the files, decoupling them from this working tree (edits no longer flow to git). Instead clone and run link.sh:
link.sh symlinks every skill here into the shared hub (~/.agents/skills/<name>) and into each agent dir (~/.claude/skills, ~/.codex/skills). Edits to a skill are then live immediately, and git push publishes them. Re-run ./link.sh any time after adding a new skill folder (it's idempotent).
It creates this symlink chain (arrows = "points at"):
~/.claude/skills/<name> ──┐
├──→ ~/.agents/skills/<name> ──→ <repo>/<name>
~/.codex/skills/<name> ──┘ (hub) (real, git working tree)
Per-agent dirs point at the hub (relative), and the hub points at this repo (absolute). link.sh creates all three links.
Pass skill names to link only those:
./link.sh # all skills (default)
./link.sh plan-execute # just one