Self-Improving OpenClaw
v1.0.0Log corrections, errors, feature requests, and recurring patterns into structured workspace learning files, then promote stable patterns into tiered memory a...
⭐ 0· 30·0 current·0 all-time
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
Name/description (self-improving/learning loop) match the content: files and instructions all relate to logging, reviewing, promoting, and maintaining workspace memory. There are no unrelated environment vars, binaries, or external services requested.
Instruction Scope
SKILL.md tells the agent to read/write specific local workspace files (.learnings/, .self-improving/) and to append/promote entries into workspace files (SOUL.md, AGENTS.md, TOOLS.md, MEMORY.md). All referenced paths are within the workspace and the guidance includes explicit safety rules (never store credentials, redact secrets, require private/local sessions). It does not instruct reading arbitrary system files or sending data externally.
Install Mechanism
No install spec; instruction-only skill plus a tiny init script (scripts/init-workspace.sh) that safely creates directories and copies bundled template files if they don't exist. The script is idempotent and does not download or execute remote code.
Credentials
Requires no environment variables or credentials. The skill's behavior (logging and promoting entries) does not justify any external credentials. The SKILL.md explicitly forbids storing secrets, which aligns with the lack of requested credentials.
Persistence & Privilege
The skill is allowed to write persistent changes into workspace files (append/promote items to HOT.md, SOUL.md, AGENTS.md, TOOLS.md, MEMORY.md). This is expected for a learning/promote workflow, but it is a lasting change to agent behavior and workspace contents — consider this intended privilege. always:false and normal autonomous invocation are used (the agent can run this skill autonomously), which increases the chance the agent will add/promote entries without an explicit human step unless you limit invocation or review promoted changes.
Assessment
This skill appears internally consistent and implements a local learning loop that reads and appends to workspace files; the included init script only creates folders and copies templates. Before installing, consider: (1) Run the init script in a test workspace to see exactly what files are created; (2) Back up SOUL.md / AGENTS.md / TOOLS.md / MEMORY.md if you don't want automatic appends; (3) Decide whether you want the agent to autonomously promote learnings — if not, avoid enabling autonomous invocation or require explicit confirmation for promotions; (4) Remember that redaction guidance is advisory (the agent must follow it): confirm by inspecting logs that secrets are not being written; (5) Review promotion/demotion rules in references/promotion-rules.md to ensure they match your expectations. If you want stronger safety, require manual review of REVIEW_QUEUE.md before any promotions are applied.Like a lobster shell, security has layers — review code before you run it.
latestvk97ey11jwsd2rvhy92jse6afh1848bdg
License
MIT-0
Free to use, modify, and redistribute. No attribution required.
Runtime requirements
🧠 Clawdis