Skip to content

SKILL.md

Latest commit

 

History

History
70 lines (49 loc) · 2.5 KB

SKILL.md

File metadata and controls

70 lines (49 loc) · 2.5 KB
Raw
name deepwork_rules
description Rules enforcement for AI agent sessions

deepwork_rules

Multi-step workflow: Rules enforcement for AI agent sessions

CRITICAL: Always invoke steps using the Skill tool. Never copy/paste step instructions directly.

Manages rules that automatically trigger when certain files change during an AI agent session. Rules help ensure that code changes follow team guidelines, documentation is updated, and architectural decisions are respected.

IMPORTANT: Rules are evaluated at the "Stop" hook, which fires when an agent finishes its turn. This includes when sub-agents complete their work. Rules are NOT evaluated immediately after each file edit - they batch up and run once at the end of the agent's response cycle.

  • Command action rules: Execute their command (e.g., uv sync) when the agent stops
  • Prompt action rules: Display instructions to the agent, blocking until addressed

Rules are stored as individual markdown files with YAML frontmatter in the .deepwork/rules/ directory. Each rule file specifies:

  • Detection mode: trigger/safety, set (bidirectional), or pair (directional)
  • Patterns: Glob patterns for matching files, with optional variable capture
  • Action type: prompt (default) to show instructions, or command to run a shell command
  • Instructions: Markdown content describing what the agent should do

Example use cases:

  • Update installation docs when configuration files change
  • Require security review when authentication code is modified
  • Ensure API documentation stays in sync with API code
  • Enforce source/test file pairing
  • Auto-run uv sync when pyproject.toml changes (command action)

Available Steps

  1. define - Create a new rule file in .deepwork/rules/

Execution Instructions

Step 1: Analyze Intent

Parse any text following /deepwork_rules to determine user intent:

  • "define" or related terms → start at deepwork_rules.define

Step 2: Invoke Starting Step

Use the Skill tool to invoke the identified starting step:

Skill tool: deepwork_rules.define

Step 3: Continue Workflow Automatically

After each step completes:

  1. Check if there's a next step in the sequence
  2. Invoke the next step using the Skill tool
  3. Repeat until workflow is complete or user intervenes

Handling Ambiguous Intent

If user intent is unclear, use AskUserQuestion to clarify:

  • Present available steps as numbered options
  • Let user select the starting point

Context Files

  • Job definition: .deepwork/jobs/deepwork_rules/job.yml