Use fresh agents for focused subtasks to avoid context pollution
The English summary is being prepared. Showing the original description.
See how to installReview the instructions the agent will receive before you install.
FRESH AGENT = FRESH CONTEXT
Use subagents for focused work without baggage
| Situation | Why Subagent Helps |
|---|---|
| Complex subtask | Focused context, clear scope |
| Going in circles | Fresh perspective breaks loops |
| Parallel work | Multiple things at once |
| Context pollution | Main agent too cluttered |
| Specialized task | Pass only relevant info |
| Research/exploration | Keep main context clean |
The Handoff:
1. DEFINE clear scope - what exactly to do
2. PROVIDE needed context - files, constraints, requirements
3. SPECIFY deliverable - what to return
4. SET success criteria - how to know it's done
Template:
Task: [Specific action to take]
Context:
- Working in [repo/directory]
- Relevant files: [list]
- Constraints: [any limits]
Deliverable:
- [Exact output expected]
Success when:
- [Criteria 1]
- [Criteria 2]
❌ Vague handoffs
Bad: "Look into this bug"
Good: "Find root cause of TypeError in user.ts:45"
❌ Context dumping
Bad: Passing entire conversation history
Good: Passing only relevant files and specific question
❌ No success criteria
Bad: "Make it better"
Good: "Refactor to reduce duplication, all tests must pass"
❌ Too broad scope
Bad: "Implement the whole feature"
Good: "Implement the validation logic for email field"
✅ Research:
✅ Focused fixes:
✅ Generation:
✅ Analysis:
When subagent returns:
1. READ the full response
2. VERIFY against success criteria
3. INTEGRATE results into main context
4. CONTINUE from where you left off
Before launching subagent:
scope-boundary-checker → Is scope clear and bounded?assumption-checker → Are handoff assumptions valid?After receiving results:
fact-checker → Verify subagent claimspre-action-verifier → Before using resultsMain agent responsibilities:
Subagent responsibilities:
No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Complete methodology for preventing AI hallucinations. Use when accuracy is critical and you need Claude to verify before claiming, cite before asserting, and admit uncertainty instead of guessing.
The English summary is being prepared. Showing the original description.
Transform rough ideas into solid designs through structured questioning
The English summary is being prepared. Showing the original description.
Canvas API Connection Skill
The English summary is being prepared. Showing the original description.
Forces every code-related claim to include file:line citations. No citation means the claim must be verified or removed. Use when building documentation, writing code reviews, or any output that others will rely on.
The English summary is being prepared. Showing the original description.
Assign confidence scores (0-100) to every claim in a response. Helps users understand which parts are verified facts and which are educated guesses. Use when the user needs to know how much to trust each part of the answer.
The English summary is being prepared. Showing the original description.
Forces Claude to extract exact quotes from files before making claims about them. Prevents hallucinations by grounding every statement in actual text from the codebase. Use when working with large files or unfamiliar code.
The English summary is being prepared. Showing the original description.