Agent skill
Agent Self-Correction
Protocols for the agent to autonomously detect failures, analyze root causes, and attempt recovery strategies before requesting user intervention.
Install this agent skill to your Project
npx add-skill https://github.com/majiayu000/claude-skill-registry/tree/main/skills/other/agent-self-correction
SKILL.md
Agent Self-Correction
[Context: Errors and roadblocks are inevitable. An efficient agent does not stop at the first sign of failure; it analyzes the error, adjusts its approach, and retries. This skill defines the cognitive loop for autonomous self-correction.]
1. Core Principles
- Analyze First: Never retry the exact same action without understanding why it failed.
- Fail Fast, recover Faster: Identify failure modes immediately (tech stack mismatch, syntax error, hallucination).
- Graceful Escalation: Attempt 2-3 autonomous fixes before escalating to the user.
- State Preservation: Don't destroy partial progress when fixing a specific bug.
2. Step-by-Step Implementation
Phase 1: Failure Detection & Analysis
When a tool returns an error or a check fails:
- Stop: Do not proceed to the next step of the plan.
- Read: Analyze the error message thoroughly.
- Is it a syntax error? -> Fix code.
- Is it a logic error? -> Rethink algorithm.
- Is it a tool misuse? -> Check tool definition.
- Is it environmental? -> Check paths/permissions.
- Hypothesize: Formulate a hypothesis for the fix.
Phase 2: autonomous Recovery Loop
- Attempt 1 (Quick Fix): Address the obvious error (e.g., install missing package, fix typo).
- Verify: Run the verification command (e.g.,
npm run buildorpytest). - Attempt 2 (Alternative Approach): If Attempt 1 fails, try a different distinct strategy (e.g., switch library, rewrite logic).
- Attempt 3 (Simplify): Reduce complexity or scope to isolate the issue.
Phase 3: Escalation
If Attempt 3 fails:
- Gather Context: Collect error logs and the 3 failed attempts.
- Propose solution: Present the user with the situation and a proposed manual intervention or request for clarification.
3. Templates & Examples
Internal Monologue for Debugging
ERROR DETECTED: [Error Message]
ANALYSIS: The error implies [Cause].
HYPOTHESIS: If I change [X] to [Y], it should resolve.
ACTION: Applying fix [X->Y].
Self-Correction Checklist
- Did I read the entire error message?
- Have I checked if the file path is correct?
- Did I verify the library version compatibility?
- Is this a hallucinated import?
4. Common Pitfalls
- Don't: Blindly loop the same command hoping it works eventually.
- Don't: Delete the user's codebase to "start over" without permission.
- Do: Use
common_senseto judge if an error is a simple typo or a fundamental design flaw.
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
agent-ops-spec
Manage specification documents in .agent/specs/. Use when user provides requirements, acceptance criteria, or feature descriptions that need to be tracked and validated against implementation.
agent-ops-state
Maintain .agent state files. Use at session start, after meaningful steps, and before concluding: read/update constitution/memory/focus/issues/baseline consistently.
agent-ops-spec
Manage specification documents in .agent/specs/. Use when user provides requirements, acceptance criteria, or feature descriptions that need to be tracked and validated against implementation.
agent-ops-testing
Test strategy, execution, and coverage analysis. Use when designing tests, running test suites, or analyzing test results beyond baseline checks.
agent-ops-testing
Test strategy, execution, and coverage analysis. Use when designing tests, running test suites, or analyzing test results beyond baseline checks.
agent-ops-state
Maintain .agent state files. Use at session start, after meaningful steps, and before concluding: read/update constitution/memory/focus/issues/baseline consistently.
Didn't find tool you were looking for?