No description
787710f1de
Submodule pointer updates were previously committed like any other change, which in practice produces useless messages such as "update submodule" because the diff in the parent repository is just a hash change. The commit-push skill now handles them explicitly across its steps. During review, the agent runs git diff --submodule=log to learn the old and new commits and the upstream subjects the bump brings in, and verifies the new commit exists on the submodule remote before committing, because a parent commit pointing at an unpushed submodule commit breaks every other checkout. When composing the message, a bare "update submodule" is forbidden: the message must name the submodule, give the old and new short hashes, and distill what the range brings in and why the parent wants it, with backwards moves called out as rollbacks. A matching guardrail makes the unpushed-pointer case a hard stop, and the skill description now mentions submodule bumps so requests like "commit the submodule update" trigger it. |
||
|---|---|---|
| .github | ||
| commit-push | ||
| create-skill | ||
| sync-agents-md | ||
| .clinerules | ||
| .cursorrules | ||
| .gitignore | ||
| .windsurfrules | ||
| AGENTS.md | ||
| CLAUDE.md | ||
| GEMINI.md | ||
| LICENSE | ||
| README.md | ||
Agent skills
Personal collection of skills for AI coding agents, written in the
tool-agnostic SKILL.md format so they work with any AI tool. Cloned to
~/.claude/skills, where user-level skills are discovered, but any agent
tool can point at the same checkout.
Layout
One directory per skill, each containing a SKILL.md:
<skill-name>/
SKILL.md # frontmatter (name, description) + instructions
... # optional supporting files, referenced by relative path
Adding a skill
- Create
<skill-name>/SKILL.mdwith YAML frontmatter:name(kebab-case, matching the directory) anddescription(the trigger text — describe when the skill should fire, in the words a user would actually use). - Write the body as imperative, self-contained, vendor-neutral instructions for the agent — no dependency on a specific tool or model.
- Verify the skill appears in the tool's skill list.
See AGENTS.md for the full conventions.