Captures learnings, errors, corrections, and feature requests to enable continuous improvement. Use when: (1) User corrects Claude ('No, that's wrong...', 'Actually...'), (2) User requests a capability that doesn't exist, (3) Claude realizes its knowledge is outdated or incorrect, (4) A better approach is discovered for a recurring task, (5) Receiving a Handoff block from self-healing (a recurring verified heal with Recurrence-Count at least 3) to distill into a memory file or new skill. For ACTIVE runtime failures where the agent needs to apply and verify a fix mid-task, use `self-healing` instead (it files HEAL- entries with proof; self-improvement promotes accumulated patterns). Also review learnings before major tasks. For CI-only/headless learning capture, use self-improvement-ci.
97
—
Does it follow best practices?
Impact
97%
4.40xAverage score across 3 eval scenarios
—
The risk profile of this skill
Multi-type self-improvement log entries
.learnings/ directory
0%
100%
Correction category
0%
100%
Learning required fields
0%
100%
Error entry exists
0%
100%
Verbatim error message
0%
100%
Reproduction steps
0%
100%
Error required fields
0%
100%
Feature request entry exists
0%
100%
Feature request required fields
0%
100%
ID format compliance
0%
100%
Area tag values
0%
100%
Priority values
0%
100%
Error metadata Reproducible field
0%
100%
Promotion workflow for broadly applicable learning
.learnings directory
0%
100%
LEARNINGS.md created
0%
100%
Correct ID format
0%
100%
ISO-8601 Logged timestamp
0%
100%
All required fields present
0%
100%
Appropriate category
0%
100%
Status set to promoted
0%
100%
Promoted annotation present
0%
100%
Project memory file created
100%
100%
Concise rule in project memory
53%
46%
Simplify & Harden feed ingestion and promotion decision
.learnings/ directory
100%
100%
LRN ID format
0%
100%
Required entry fields
10%
100%
Source annotation
0%
100%
Pattern-Key dead_code
37%
100%
Pattern-Key null_checks
37%
100%
Recurrence counts
77%
100%
First-Seen / Last-Seen fields
80%
100%
null_checks promoted status
80%
100%
null_checks Promoted annotation
40%
100%
dead_code NOT promoted
100%
100%
CLAUDE.md concise rule
90%
100%
8a71d70
Table of Contents
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.