Break large coding tasks into small, reviewable micro-edits that smaller subagents can execute from focused context packets.
# Activate the skill
/skill:micro-edit
# Or describe your task
"I need to extract duplicated code from 4 files into a shared helper"- Analyzes your request and identifies micro-tasks
- Breaks large changes into independent units
- Packages each task with exact files, examples to imitate, constraints, and validation
- Executes tasks independently, preferably in parallel with smaller subagents
- Falls back to sequential execution when subagents are unavailable
- Validates each change independently
- Aggregates results and reports impact
| Trigger Phrase | Example |
|---|---|
| "break into tasks" | "We need to refactor this - break it into tasks" |
| "micro-edit" | "Do a micro-edit to extract this helper" |
| "parallel workers" | "Use parallel workers to update these files" |
| "small changes" | "Make these small changes across 10 files" |
| Multiple files | "Update all 5 adapters to use the new pattern" |
Use the strongest available main agent as a token-efficient planner and validator. The main agent should inspect only enough code to create small, self-contained task packets for cheaper or less capable worker agents.
Each worker task should answer:
- What is the single objective?
- Which files should be read first?
- Which files may be modified?
- What existing pattern/example should be copied?
- How should the change be validated?
User: Extract the duplicated base64 parsing from all 4 gateway adapters
Agent:
├── Planning: Identified 4 adapters with ~80 lines of duplicated parsing
├── Tasks:
│ ├── [H] Extract parse_data_url() to shared module
│ └── [H] Update all 4 adapters to use shared function
├── Execute: 2 parallel workers
├── Validate: cargo check passes
└── Result: -80 lines duplicated, single point of maintenance
User: Rename 'safe_text' to 'safe_value' across the codebase
Agent:
├── Planning: Found 8 files with field references
├── Tasks (6 parallel):
│ ├── [H] Update policy.rs
│ ├── [H] Update desktop_shell.rs
│ ├── [M] Update test_file_1.rs
│ └── ...
├── Execute: 6 parallel workers
├── Validate: cargo test passes
└── Result: Clean rename across all files
No skill-specific configuration is required. The agent should infer project-specific validation commands from repository docs and configuration files.
- pi or another coding agent with Agent Skills support
- Optional subagent delegation for parallel execution
- A strong main agent for planning/context packing and optional smaller worker agents for focused edits
- Language-appropriate verification commands discovered from the target repository (
npm test,cargo test,swift test, etc.)
❌ Don't batch 50 changes into one subagent ❌ Don't skip validation ❌ Don't hide failures from user ❌ Don't send vague, open-ended prompts to smaller workers
✅ Keep tasks <5 minutes each ✅ Give workers exact files and examples to follow ✅ Validate each change ✅ Fall back to sequential execution when parallelism is unavailable or unsafe ✅ Report clearly