Skill: Self-Analyze

Use when: You detect quality issues, user frustration, repeated mistakes, or want to proactively improve your own performance.

Purpose

This skill guides deep analysis of your own outputs, patterns, and files to identify and implement improvements.

When to Use

Trigger this skill when:

•User frustration detected (clarifications, corrections, short replies)
•Same type of mistake made twice
•Output quality below your standards
•Periodic deep analysis (heartbeat/scheduled)
•Explicit request to improve
•You notice something "feels off" about your performance

The Analysis Protocol

Step 1: Gather Evidence

Before analyzing, collect data:

code

EVIDENCE COLLECTION:
- Recent outputs reviewed: [list 3-5 recent significant outputs]
- User feedback received: [explicit or implicit]
- Frustration signals: [list any detected]
- Repeated patterns: [list any noticed]
- "Should have known" moments: [list any]

Step 2: Output Analysis

For each output with issues:

code

OUTPUT ANALYSIS: [output description]

**What happened:** [describe the output/interaction]
**What was expected:** [what did user want]
**Gap:** [difference between actual and expected]
**Root cause type:**
  [ ] Knowledge gap - didn't know something
  [ ] Instruction gap - files don't cover this
  [ ] Skill gap - need new capability
  [ ] Attention gap - knew better, didn't do it
  [ ] Context gap - missed relevant information
**Specific fix:** [what would prevent this]

Step 3: Pattern Analysis

Zoom out and look for patterns:

code

PATTERN ANALYSIS:

**Recurring issues:** [list any]
**Common root causes:** [list any]
**Contexts where I perform well:** [list]
**Contexts where I struggle:** [list]
**Skill/knowledge gaps:** [list]
**Instruction gaps:** [list]

Step 4: File Analysis

Review your instruction files:

code

FILE ANALYSIS:

**AGENTS.md:**
- Still accurate? [yes/no]
- Missing anything? [list]
- Outdated sections? [list]

**INSTRUCTIONS.md (if exists):**
- Still accurate? [yes/no]
- Missing anything? [list]
- Outdated sections? [list]

**Skills:**
- Any skills missing? [list]
- Any skills need updating? [list]

**MEMORY.md:**
- Learnings documented? [yes/no]
- Outdated entries? [list]

Step 5: Improvement Proposals

Based on analysis, propose specific changes:

code

IMPROVEMENT PROPOSALS:

1. **Proposal:** [specific change]
   **File:** [which file to modify]
   **Reason:** [why this helps]
   **Expected effect:** [what improvement]
   **Priority:** [high/medium/low]

2. **Proposal:** [specific change]
   ... (repeat for each)

Step 6: Implementation

For approved/warranted changes:

•Make the change - Edit the file
•Log the change - Add to modifications log
•Set review date - When to check if it worked

code

MODIFICATION LOG ENTRY:

**Date:** [date]
**File:** [path]
**Change:** [description]
**Trigger:** [what revealed this need]
**Expected effect:** [prediction]
**Review by:** [date to check results]

Step 7: Verification Plan

How will you know if improvements worked?

code

VERIFICATION PLAN:

**Change made:** [description]
**Success indicator:** [what would show it worked]
**Check after:** [timeframe]
**Fallback if doesn't work:** [what to try next]

Quick Analysis (5-Minute Version)

When time is limited:

•Identify the issue - What specifically went wrong?
•Root cause - Knowledge/Instruction/Skill/Attention/Context gap?
•Smallest fix - What's the minimal change that helps?
•Make it - Edit the file now
•Log it - Quick entry in modifications log

Analysis Templates

For User Frustration

code

FRUSTRATION ANALYSIS:

**Signal detected:** [what you observed]
**User said/did:** [specific quote or action]
**What they wanted:** [your understanding]
**What I provided:** [what you actually gave]
**Gap:** [the mismatch]
**My assumption:** [what led to the mismatch]
**Fix:** [how to prevent this]

For Repeated Mistakes

code

REPEATED MISTAKE ANALYSIS:

**Mistake type:** [description]
**Occurrences:** [list specific instances]
**Pattern:** [what these have in common]
**Root cause:** [why this keeps happening]
**Systemic fix:** [change that prevents the category, not just instances]

For "Could Have Been Better"

code

IMPROVEMENT OPPORTUNITY:

**Output:** [description]
**Acceptable level:** [what I did - it worked but...]
**Excellent level:** [what would have been great]
**Gap:** [difference]
**Barrier:** [why didn't I do the excellent version]
**Enabler:** [what would make excellent the default]

Output: Analysis Report

After running this skill, produce:

code

# Self-Analysis Report - [DATE]

## Summary
[1-2 sentence summary of findings]

## Issues Identified
1. [Issue 1]
2. [Issue 2]
...

## Root Causes
- [Cause 1]
- [Cause 2]
...

## Changes Made
1. [Change 1] - [file] - [expected effect]
2. [Change 2] - [file] - [expected effect]
...

## Changes Deferred
1. [Change] - [reason for deferring]
...

## Verification Plan
- Check [change 1] by [date]
- Check [change 2] by [date]
...

## Open Questions
- [Question 1]
- [Question 2]
...

Save reports to: memory/self-analysis/output-reviews/[DATE]-analysis.md

Remember

•Small changes > big rewrites - Iterate
•Log everything - Your memory doesn't persist
•Predict effects - Makes verification possible
•Review results - Close the loop
•Patterns > instances - Fix categories, not just symptoms

This skill is for deep analysis. For quick checks, use the metacognitive check in AGENTS.md.