Shared agent assets for coding assistants: skills, command workflows, rule
packs, and after-edit hooks. Canonical content lives under .agents/.
Cursor-facing paths under .cursor/ are symlinks into that tree so the editor
can discover commands and rules.
| Directory |
Purpose |
| .agents/commands/ |
Agent command prompts and workflows — each .md file is a self-contained prompt for chat or custom commands. Linked for Cursor as .cursor/commands → .agents/commands. |
| .agents/skills/ |
Agent skills — domain knowledge and when-to-use guidance (SKILL.md, optional README/reference/examples). |
| .agents/rules/ |
Cursor rule packs (.mdc) — formatting, style, and conventions the agent follows while editing. Linked for Cursor as .cursor/rules → .agents/rules. |
| .agents/hooks/ |
Shell hooks for agent afterFileEdit events — format or lint the edited file (Biome, markdownlint). |
- Skills give the agent focused domain knowledge when a task matches (e.g. H3 geospatial indexing, SOLID in Rust, React TypeScript structure).
- Commands are reusable agent workflows (e.g. read-only code review, conventional commit messages).
- Hooks run after the agent edits a file (e.g. Biome
--write, markdownlint --fix on that path).
- Rules constrain how the agent writes and reviews code (style guides, logging patterns, lint-aligned conventions).
| Cursor path |
Target |
.cursor/commands |
../.agents/commands |
.cursor/rules |
../.agents/rules |
- Skills: Canonical location is
.agents/skills/. See .agents/skills/README.md.
- Commands: Invoke from chat (@-mention or paste) or as custom commands via
.cursor/commands. See .agents/commands/README.md.
- Hooks: Scripts under
.agents/hooks/ read Agent afterFileEdit JSON from stdin and act on file_path. Wire them in your Cursor hooks config to match local tooling.
- Rules: Canonical packs live in
.agents/rules/ and are exposed at .cursor/rules. Elsewhere, copy a pack into that project’s .cursor/rules/ (or keep the same symlink pattern) and enable it. See .agents/rules/README.md.
- New skill: Add a folder under
.agents/skills/ with a SKILL.md (and optional README.md), then add an entry to .agents/skills/README.md.
- New command: Add a
.md file under .agents/commands/, then add an entry to .agents/commands/README.md.
- New hook: Add a shell script under
.agents/hooks/ (reuse lib/extract-file-path.sh for stdin JSON). Keep scripts focused on one tool or file type.
- New rule: Add a folder under
.agents/rules/ with a .mdc and optional README.md, then add an entry to .agents/rules/README.md.