Skip to content

AGENTS.md

Latest commit

 

History

History
43 lines (30 loc) · 3.03 KB

AGENTS.md

File metadata and controls

43 lines (30 loc) · 3.03 KB

Do-Work Skill Project

Before Every Commit

Always bump the version in actions/version.md before committing.

The version is on the line that starts with **Current version**: - increment it using semver:

  • Patch (0.1.0 → 0.1.1): Bug fixes, minor tweaks
  • Minor (0.1.0 → 0.2.0): New features, behavior changes
  • Major (0.1.0 → 1.0.0): Breaking changes

When in doubt, bump the patch version.

Immediately after bumping the version, update CHANGELOG.md.

Add a new entry at the top of the file (below the header), following this format:

## X.Y.Z — The [Fun Two-Word Name] (YYYY-MM-DD)

[1-2 sentences. Casual, clear, and concise. What changed and why it matters. No fluff, but personality is welcome.]

- [Bullet points for specifics — what was added, changed, or fixed]

Rules for changelog entries:

  • Newest on top. The file reads top-to-bottom as newest-to-oldest.
  • Give every version a name. A short, fun title after the em dash (e.g., "The Organizer", "Typo Patrol"). This isn't a novel — two or three words max.
  • Date every entry. Add the date in parentheses after the name, formatted as (YYYY-MM-DD). Use today's date.
  • Lead with the value, not the implementation. Say what the user gets, not what files changed. "The archive tidies itself now" beats "added cleanup.md".
  • Keep it brief. One short paragraph + a few bullets. If you're writing more than 5 bullets, you're over-explaining.
  • Match the voice. Conversational, not corporate. Imagine you're telling a friend what shipped. No jargon walls, no passive voice marathons.
  • Every version gets an entry. No skipping. Even a patch fix deserves a line.

Agent Compatibility

This skill is designed to work with any agentic coding tool, not just one specific platform. When writing or editing action files:

  • Use generalized language. Say "use your environment's ask-user prompt/tool" rather than naming a specific tool API. Say "spawn a subagent" rather than referencing a specific tool's agent mechanism.
  • Hint at advanced features, don't require them. Subagents, multi-agent workflows, and structured question UIs improve results when available. The actions must still work in a single-session tool that reads the markdown as a prompt.
  • Each action file should work as a standalone prompt. If someone pastes do.md into a basic chat interface with file access, the instructions should be clear enough to follow without the SKILL.md routing layer or any skill-runner infrastructure.
  • No tool-specific APIs in action files. Tool-specific names and APIs belong in the tool's own integration layer, not in the skill's action files. Use platform-specific details only as clearly-labeled examples (e.g., "Example: Claude Code caches images at ~/.claude/...").
  • Design for the floor, not the ceiling. The least sophisticated agent that can read/write files and run shell commands should be able to execute these actions correctly. Advanced agents benefit from subagents and parallel execution, but the baseline must work without them.