My Agent Skills, written once and consumed by several LLM harnesses: omp, opencode, openclaw, hermes. Most of them cover the parts of the workflow around the code: implementing a plan as a subagent, reviewing it, committing, and getting a PR through review.
llm-skills
Collection of my custom LLM agent skills.
One skills folder, every harness.
LLM toolingToolkitMarkdown, Python, ShellMIT
What it is
How it works
make install # or: ./tools/install.shThe install is idempotent; I re-run it after every git pull. It lints every skill and aborts on problems, symlinks skills/<name> into the shared hub ~/.agents/skills while pruning links for removed skills, and wires the harnesses that don't read the hub natively. omp, opencode and openclaw read it on their own; hermes gets the hub added to skills.external_dirs in ~/.hermes/config.yaml. SKILLS_HUB=/path moves the hub.
The skills
| Skill | Use when |
|---|---|
agent-implement | acting as the implementation subagent, executing a plan step by step |
agent-review | acting as the independent reviewer/QA subagent |
imperative-commits | committing, splitting work into commits, writing commit messages |
pr-comments | triaging review comments on a GitHub PR or GitLab MR |
pr-babysit | watching one PR/MR in the background until merge |
pr-ship | finishing a branch: verify green, tidy history, push, open the PR |
pr-summary | writing a PR description for the current branch |
quick-code-review | fast read-only review of uncommitted changes |
quick-fix | fixing issues found earlier in the conversation, user-invoked only |
visual-code-review | rendering a PR or branch review as a step-by-step HTML walkthrough |
The README also lists external skill collections I recommend, among them pbakaus/impeccable and mattpocock/skills.