Skip to content

skills.md

Latest commit

 

History

History
121 lines (76 loc) · 4.15 KB

skills.md

File metadata and controls

121 lines (76 loc) · 4.15 KB

Skills

Skills are specialized instructions that agents load for specific workflows. dot-agents keeps them under .agents/skills/ and, for Claude Code projects, links them into .claude/skills/ when .claude/ already exists.

Available Skills

Skill Trigger Purpose
adapt Run adapt Analyze project, fill in AGENTS.md
agent-work Create a work item Create and maintain .agents/work/ context
feature-planning Create a plan, write a handoff prompt Turn context into plans and new-thread prompts
research Research [topic] Investigate and save work-local or reusable findings
tmux tmux, background process Manage background processes

adapt

Analyzes your project and fills in AGENTS.md with:

  • Project overview and tech stack
  • Build/test/lint commands
  • Code conventions
  • Project structure
  • dot-agents work-item workflow guidance

Invoke: Run adapt

Details: .agents/skills/adapt/SKILL.md

agent-work

Creates and curates durable work item folders:

.agents/work/<category>/<slug>/index.md

Use it to create work items, list active work, place artifacts deliberately, and migrate legacy plans when requested.

Invoke: Create a new work item for user authentication

Details: .agents/skills/agent-work/SKILL.md

feature-planning

Turns work-item context into implementation-ready plans and paste-ready handoff prompts.

Use it to:

  • Create or refine a short requirements brief (prd.md) only when alignment is needed
  • Create or refine the active plan file
  • Validate stale assumptions before implementation
  • Generate a new-thread handoff prompt
  • Stress-test a plan when explicitly asked

Invoke: Write a handoff prompt for .agents/work/feature/user-authentication

Details: .agents/skills/feature-planning/SKILL.md

research

Investigates technical questions using available local, web, and repository evidence.

Research is saved:

  • Work-locally at .agents/work/<category>/<slug>/research.md or .agents/work/<category>/<slug>/research/<topic>.md when it supports one work item
  • Reusably at .agents/research/<topic>.md when it should guide future unrelated work

Invoke: Research authentication patterns for this work item

Details: .agents/skills/research/SKILL.md

tmux

Background process management:

  • Spawns long-running processes such as servers or watchers
  • Captures output for review
  • Sends interrupts or kills background windows when needed

Invoke: Use tmux to run the dev server in the background

Details: .agents/skills/tmux/SKILL.md

Adding Custom Skills

Create a new skill by adding a directory under .agents/skills/:

.agents/skills/my-skill/
└── SKILL.md

SKILL.md Format

---
name: my-skill
description: "Brief description. Use when relevant context applies. Triggers on: keyword1, keyword2."
---

# My Skill

Instructions for the agent when this skill is loaded.

The name and description fields are used for skill discovery. Keep descriptions quoted, concise, and trigger-rich.

Preserving Custom Skills

Custom skills in .agents/skills/ are preserved during sync.sh updates. Only upstream core skills are updated or retired by dot-agents.

Claude Code Project Skill Discovery

Claude Code discovers project skills in .claude/skills/<skill>/SKILL.md. dot-agents keeps .agents/skills/ as the source of truth, so when install.sh or sync.sh detects an existing .claude/ directory it creates directory symlinks such as:

.claude/skills/adapt -> ../../.agents/skills/adapt
.claude/skills/agent-work -> ../../.agents/skills/agent-work

Directory symlinks expose the whole skill, including optional supporting files like assets/, references/, and scripts/. The installer skips user-owned Claude Code skills and only removes dot-agents-managed symlinks during uninstall or retired-skill cleanup.