← Back to Skills Marketplace
363
Downloads
0
Stars
0
Active Installs
4
Versions
Install in OpenClaw
/install active-self-improvement
Description
Active self-improvement loop that reads learnings, errors, batch outputs, and memory — detects patterns — and UPDATES skills/protocols/behavior automatically...
README (SKILL.md)
Auto-Improve
Reads logs, detects patterns, rewrites the playbook. Not passive logging — this ACTS on what it learns.
SCAN (read logs) ──► PROPOSE (specific edits) ──► APPLY (low-risk auto, high-risk flag)
Input Sources
| Source | What It Contains |
|---|---|
.learnings/ERRORS.md |
What broke and how it was fixed |
.learnings/LEARNINGS.md |
Corrections, insights, knowledge gaps, batch outcomes |
workspace/OUTSTANDING.md |
Ranked ideas and opportunities |
memory/permanent/*.md |
Current knowledge state |
workspace/DELEGATION_PLAN.md |
Atom timing data (if delegation was used) |
Step 1: SCAN
Detect:
- Repeated errors — same mistake 3+ times → needs a prevention rule
- Repeated corrections — user keeps fixing the same thing → behavior change needed
- Emerging patterns — 3+ items connecting → thesis forming
- Stale knowledge — facts in permanent memory contradicted by recent sessions
- Unused wins — high-value items that haven't been acted on
Step 2: PROPOSE
For each detected pattern:
PROPOSAL: [short title]
EVIDENCE: [file#line references]
CHANGE: [exact edit — old text → new text]
RISK: [low/medium/high]
REVERSIBLE: [yes/no]
Pattern-Key: [hash(error+fix) for dedup]
| Pattern Type | Action | Target File |
|---|---|---|
| Repeated error | Add prevention rule | relevant skill's ## Learned section |
| Repeated correction | Update behavior guideline | SOUL.md or AGENTS.md |
| Emerging thesis | Write thesis + next steps | OUTSTANDING.md |
| Stale knowledge | Update the fact | memory/permanent/*.md |
| Unused win | Create ticket or reminder | NEXT_TICKET.md or cron |
Step 3: APPLY
- Low risk + reversible: Apply immediately. Log the change.
- Medium risk: Apply but notify user on next interaction.
- High risk: Write to
OUTSTANDING.mdand wait for approval. - Dry-run mode (
--dry-run): Propose all changes but apply none. Output a report.
Use 3-occurrence threshold before proposing pattern-based changes. Track recurrence with Pattern-Key and Recurrence-Count.
Error→Skill Feedback Loop
After SCAN, for each error in ERRORS.md:
- Extract the
Contextcolumn value - Match against skill names (fuzzy: "SiteBlitz CSS" →
webdev-sop) - If match found and skill doesn't already have the fix in
## Learned:## Learned - [date] [error summary] → [fix]. Source: .learnings/ERRORS.md#L[N] - Use
Pattern-Key: hash(error+fix)to prevent duplicates
Skills self-heal: every failure improves the relevant skill.
Delegation Feedback
After delegation plan completes:
- Read atom timing data from DELEGATION_PLAN.md
- Atom actual time > 2× estimated → flag estimation drift
- Atom model upgraded (flash→sonnet) → update routing suggestion in MODEL_ROUTING_PROTOCOL.md
- Append summary to
.learnings/LEARNINGS.md
Usage Guidance
This skill does what it says — it will scan agent logs and memory and can automatically edit other skills and persistent memory. That's powerful but risky because mistakes or buggy heuristics could alter behavior across your agents. Before installing: 1) Require dry-run by default and review proposed changes; never auto-apply low/medium changes without human approval unless you trust the environment. 2) Restrict its write scope to a sandboxed directory or a git branch and enable automatic backups/versioning so edits can be reverted. 3) Add explicit approval gates for 'medium' and 'high' risk changes and keep an audit log of every change and its Pattern-Key. 4) Run it in a test agent with representative data first and validate proposals against a test suite. 5) Limit its scheduling/autonomy (avoid enabling unattended runs) until you are confident in its proposals. These mitigations will reduce the chance it silently alters other skills or agent memory in undesirable ways.
Capability Analysis
Type: OpenClaw Skill
Name: active-self-improvement
Version: 1.3.0
The 'active-self-improvement' skill implements a self-modifying logic loop that automatically rewrites the agent's core instructions (e.g., SOUL.md, AGENTS.md) and skill files based on logs and error patterns. While the stated intent is optimization and 'self-healing,' this capability introduces a high risk of indirect prompt injection, where malicious input captured in logs (.learnings/ERRORS.md) could be promoted into permanent behavioral rules. The automated application of 'low-risk' changes and scheduled execution via cron further increase the risk of unattended logic shifts.
Capability Assessment
Purpose & Capability
The name/description claim an automatic self-improvement loop; the SKILL.md implements exactly that (scan logs, propose edits, apply changes). No unrelated environment variables or binaries are requested, so capability and purpose are aligned.
Instruction Scope
The instructions direct the agent to read repository and agent-state files (.learnings/ERRORS.md, memory/permanent/*.md, workspace/*) and to write edits to other skill documents (e.g., add '## Learned' entries, update SOUL.md/AGENTS.md, memory files, OUTSTANDING.md, NEXT_TICKET.md). It also prescribes automatic application rules (apply low-risk immediately; medium notify later; high-risk wait) and fuzzy matching of errors to skill names. This grants broad discretion to modify other skills and persistent memory with limited human review.
Install Mechanism
Instruction-only skill with no install spec or code files. Lowest install risk — nothing is downloaded or executed beyond what the agent is instructed to do at runtime.
Credentials
No environment variables, credentials, or external config paths are requested. The skill only references repository and agent-local paths, which is proportional to an auto-improvement function.
Persistence & Privilege
Although always:false, the skill is intended to autonomously run on schedules or triggers and to modify other skills' files and memory. The instructions explicitly direct changing other skills' content (writing '## Learned' sections, updating memory) which is precisely the kind of cross-skill modification the policy flags as a privilege concern without explicit guardrails.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install active-self-improvement - After installation, invoke the skill by name or use
/active-self-improvement - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.3.0
Added error-to-skill feedback loop and delegation timing integration
v1.2.0
active-self-improvement v1.2.0
- Renamed skill to "active-self-improvement" for clarity and consistency
- Added dry-run mode to preview proposed changes without applying them
- Improved pattern detection: now triggers on three occurrences instead of two
- Introduced recurrence tracking metadata for pattern detection
- Enhanced risk classification distinguishing reversible from permanent changes
- Simplified trigger integration, especially after the Recorder skill
v1.1.0
Refined trigger conditions, clearer input sources, improved loop documentation. Tested across 30+ sessions.
v1.0.0
Auto-Improve v1.0.0 – Active Self-Improvement Engine
- Introduces an active self-improvement loop that automatically reads learnings, errors, outputs, and memory to detect patterns and update skills, protocols, or behavior.
- Runs in the background at key workflow completion points, and on request, acting immediately on low-risk improvements.
- Distinguishes itself from passive logging by making changes based on detected mistakes or new insights, not just recording them.
- Includes robust safety and logging protocols, ensuring traceability and user control over high-risk changes.
- Enables smarter, pattern-driven adaptation between simple error correction and predictive behavior.
Metadata
Frequently Asked Questions
What is Active Self-Improvement?
Active self-improvement loop that reads learnings, errors, batch outputs, and memory — detects patterns — and UPDATES skills/protocols/behavior automatically... It is an AI Agent Skill for Claude Code / OpenClaw, with 363 downloads so far.
How do I install Active Self-Improvement?
Run "/install active-self-improvement" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Active Self-Improvement free?
Yes, Active Self-Improvement is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Active Self-Improvement support?
Active Self-Improvement is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Active Self-Improvement?
It is built and maintained by KairoKid (@dodge1218); the current version is v1.3.0.
More Skills