self-improvement

Original：🇨🇳 Chinese

Translated

Use this when unexpected failures occur in tasks, when users provide corrections, when requests for missing capabilities are made, or when reusable project patterns are discovered. It is responsible for depositing high-value experiences into `.learnings/` and deciding whether to elevate them to project rules or independent skills.

1installs

Sourcealleneee/skill-agent

Added on2026-08-22

NPX Install

npx skill4agent add alleneee/skill-agent self-improvement

SKILL.md Content (Chinese)

View Translation Comparison →

Self-Improvement

Convert high-value experiences from one-time conversations into reusable assets within the project.

Quick Reference

Scenario	Action
Command/Tool Failure	Record to `.learnings/ERRORS.md`
User Corrected You	Record to `.learnings/LEARNINGS.md` , category `correction`
User Requests Missing Capability	Record to `.learnings/FEATURE_REQUESTS.md`
External API Failure	Record to `.learnings/ERRORS.md` , including integration details
Outdated Knowledge	Record to `.learnings/LEARNINGS.md` , category `knowledge_gap`
Better Practice Discovered	Record to `.learnings/LEARNINGS.md` , category `best_practice`
Similar to Existing Entry	Link with `See Also` , consider raising priority
Widely Applicable Experience	Elevate to `CLAUDE.md` / `AGENTS.md`

File Structure

text

.learnings/
├── LEARNINGS.md
├── ERRORS.md
└── FEATURE_REQUESTS.md

When to Record

Prioritize recording in these situations:

Users corrected your judgment, assumptions, or implementation methods
Non-obvious failures occurred in tools, commands, or runtime environments
Users requested capabilities not directly supported by the current system
You discovered reusable project constraints, workflows, or best practices
Your knowledge is outdated or inconsistent with actual behavior

Do not record these situations:

Obvious one-time errors
Minor issues that can be resolved without review
General common sense unrelated to the current project

Recording Principles

Only retain high-signal information, avoid turning
```
.learnings/
```
into a noise accumulation area
First write "why it will be useful in the future", then write "what happened this time"
If it's an error, clarify reproduction clues and suggested repair directions
If it's a user correction, clarify where the original assumption was wrong
Search existing entries before recording, link duplicates with See Also
If the pattern has stabilized, consider elevating to
```
AGENTS.md
```
or extracting a new skill

Recording Format

Learning

markdown

## [LRN-YYYYMMDD-XXXXXX] correction | knowledge_gap | best_practice

**Logged**: 2026-03-23T10:00:00+00:00
**Priority**: low | medium | high | critical
**Status**: pending
**Area**: frontend | backend | infra | tests | docs | config | runtime

### Summary
One-sentence explanation of what was learned

### Details
Explain the wrong assumption at the time, trigger conditions, and correct approach

### Suggested Action
How to avoid making the same mistake in the future

### Metadata
- Source: user_feedback | error | conversation
- Related Files: path/to/file.py
- Tags: tag1, tag2
- See Also: LRN-20260320-001 (if related to existing entry)
- Pattern-Key: correction.api_format (optional, for repeating pattern tracking)
- Recurrence-Count: 1 (optional)

---

Error

markdown

## [ERR-YYYYMMDD-XXXXXX] auto_logged_error

**Logged**: 2026-03-23T10:00:00+00:00
**Priority**: medium | high
**Status**: pending
**Area**: runtime

### Summary
One-sentence explanation of where the failure occurred

### Error
```text
Specific error content
```

### Context
- Step: 3
- Tool: `bash`
- Task: Current task summary

### Suggested Fix
Explain how to review or correct it later

### Metadata
- Reproducible: yes | no | unknown
- Related Files: N/A
- Pattern-Key: tool_error.bash (optional)
- Recurrence-Count: 2 (optional)

---

Feature Request

markdown

## [FEAT-YYYYMMDD-XXXXXX] capability_name

**Logged**: 2026-03-23T10:00:00+00:00
**Priority**: medium
**Status**: pending
**Area**: runtime

### Requested Capability
What capability the user needs

### User Context
Why they need it

### Complexity Estimate
simple | medium | complex

### Suggested Implementation
Feasible minimal implementation idea

### Metadata
- Frequency: first_time | recurring
- Related Features: N/A

---

Entry Lifecycle

Status	Meaning
`pending`	New record, pending processing
`in_progress`	Being processed
`resolved`	Fixed/resolved
`promoted`	Elevated to CLAUDE.md / AGENTS.md / new skill
`wont_fix`	Decision not to process (explain reason in Resolution)

When resolving an entry, append a Resolution block:

markdown

### Resolution
- **Resolved**: 2026-03-24T09:00:00Z
- **Commit/PR**: abc123 or #42
- **Notes**: Brief description of what was done

Repeating Pattern Detection

Search existing entries before recording:

bash

grep -r "keyword" .learnings/

If similar entries are found:

Link with
```
See Also
```
Consider raising priority
If Recurrence-Count >= 3, strongly recommend elevating to project rules

Priority Guidelines

Priority	Applicable Scenarios
`critical`	Blocks core functionality, risk of data loss, security issues
`high`	Significant impact, common workflows blocked, recurring issues
`medium`	Moderate impact, has workaround
`low`	Minor inconvenience, edge cases

Area Tags

Area	Scope
`frontend`	UI, components, client-side code
`backend`	APIs, services, server-side code
`infra`	CI/CD, deployment, Docker, cloud
`tests`	Test files, testing tools
`docs`	Documentation, comments
`config`	Configuration files, environment variables
`runtime`	Agent runtime, tool execution

Elevation Path

When a learning has been verified multiple times and has cross-task value:

Learning Type	Elevation Target	Example
Project Convention	`CLAUDE.md`	"Use uv as package manager, not pip"
Workflow Rule	`AGENTS.md`	"Must regenerate client after API changes"
Reusable Pattern	New skill	Patterns complex enough to require dedicated guidance

After elevation, update the original entry's Status to

promoted

and note

**Promoted**: CLAUDE.md

.

Skill Extraction Criteria

A learning can be extracted as an independent skill if it meets any of the following conditions:

Has 2+ See Also links (recurring occurrence)
Status is resolved and the fix has been verified
Non-obvious, requires actual debugging to discover
Not limited to specific projects, usable across codebases
User explicitly requests "save this as a skill"

Relationship with Native Hook

The project's

SelfImprovementHook

automatically performs these actions:

Reminds you to use this skill when necessary before running
Automatically records a basic entry in
```
ERRORS.md
```
when a tool fails
Automatically raises priority and tracks Pattern-Key when the same tool fails repeatedly
Automatically records to
```
LEARNINGS.md
```
when user rejects/edits feedback
Automatically records to
```
FEATURE_REQUESTS.md
```
when feature_request feedback is received

Automatic recording is only a safety net. Truly high-quality learning still requires active organization by you before task completion.

Regular Review

Review

.learnings/

at these times:

Before starting a new large task
After completing a feature
When working in an area with historical learnings

bash

grep -h "Status\*\*: pending" .learnings/*.md | wc -l
grep -B5 "Priority\*\*: high" .learnings/*.md | grep "^## \["