Self Improving
v1.8.0Autonomous behavioral research loop that optimizes agent behavior through correction tracking and multi-perspective (MAGI) verification.
⭐ 0· 126·0 current·0 all-time
bySachin Chalapati@teenu
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
medium confidencePurpose & Capability
Name/description describe a self-improvement loop and the skill only requires editing three local files (memory.md, corrections.md, experiments.md). There are no unrelated env vars, binaries, or install steps requested — the required capabilities match the stated purpose.
Instruction Scope
SKILL.md confines activity to the included files and defines append/edit constraints. That scope matches the goal, but the agent is instructed to record user corrections and modify its own memory (including deleting rules). This creates legitimate privacy and self-reinforcement risks (it will store user content in logs and may change behavior based on its own measurements). No instructions reference external endpoints or unrelated system paths.
Install Mechanism
Instruction-only skill with no install spec and no code files. Nothing is written to disk by an installer beyond the skill's own files; lowest install risk.
Credentials
The skill requests no environment variables, credentials, or external config paths. That is proportionate to the declared functionality.
Persistence & Privilege
The skill does not request always:true and defaults to requiring user confirmation (autonomous:false). However, it is explicitly self-modifying (edits memory.md and appends logs). The append/edit constraints are procedural only — there is no enforcement mechanism in the SKILL.md, so the agent's ability to modify these files gives it persistent influence over future behavior and requires monitoring by the user.
Assessment
This skill appears to do what it says: track corrections, propose rules, and edit its local memory/log files. Before installing: (1) Be aware it will store user corrections and derived rules in the included files — avoid logging sensitive or private data in corrections.md. (2) Review and back up SKILL.md, memory.md, and experiments.md periodically because the agent is instructed to edit them (including deleting rules). (3) Keep autonomous mode disabled until you audit a few cycles — the loop can self-reinforce and drift without external oversight. (4) Confirm your platform enforces the append-only/read-only constraints you expect; if not, treat the skill as having full write access to its bundle and monitor for unexpected changes. If you need stronger guarantees (no local storage of user text, enforced append-only behavior, or audit logging to an external trusted store), request those controls before enabling this skill.Like a lobster shell, security has layers — review code before you run it.
latestvk976c1xmeq05k4tggjmqq8khhs83tm0b
License
MIT-0
Free to use, modify, and redistribute. No attribution required.
Runtime requirements
🧬 Clawdis