Self Improving Agent 3.0.5
v1.0.0Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Clau...
⭐ 0· 125·17 current·19 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 (capture learnings, log errors, promote learnings) align with the included files: reminders, logging templates, hook handlers, activator and error-detection scripts, and a helper to extract skills. No environment variables, credentials, or binaries are requested that would be unrelated to logging and hook-based reminders.
Instruction Scope
Runtime instructions and scripts perform the expected scoped tasks: creating .learnings files, emitting reminder text, detecting error output, and scaffolding new skills. The error-detector script reads CLAUDE_TOOL_OUTPUT (a runtime-provided env var) to detect errors — that is proportionate to 'detect command failures', but you should be aware that learnings may reference command output and file paths. The hooks inject a virtual bootstrap file into session context; this is intended, but installing globally will surface the content across sessions.
Install Mechanism
No install spec (instruction-only plus included scripts). The SKILL.md suggests cloning from a GitHub repo or using clawdhub; there are no remote downloads or archive extracts in the package itself. The helper scripts are local and include safeguards (extract-skill.sh refuses absolute paths or '..').
Credentials
The skill declares no required env vars or credentials. The scripts rely on runtime context: CLAUDE_TOOL_OUTPUT (read-only) and event/session fields provided by OpenClaw hooks — these are relevant to error detection and hook handling. No unrelated secrets or broad environment access are requested.
Persistence & Privilege
always:false and model invocation is allowed (default). The skill offers optional hooks that the user must enable; enabling hooks and copying files to ~/.openclaw/hooks or ~/.openclaw/workspace gives it persistent presence across sessions (intentional for a workspace-level learning system). This is expected but you should opt for project-level configuration if you don't want global persistence.
Assessment
This skill appears to do what it says: it provides templates, a lightweight activator reminder, an error detector that watches tool output, and a helper to scaffold extracted skills. Before installing or enabling hooks: 1) Review the scripts (activator.sh, error-detector.sh, extract-skill.sh) to confirm you’re comfortable with their behavior; they do not exfiltrate data but do read runtime tool output and may encourage logging of command outputs to .learnings. 2) Prefer project-level (per-repo) hook configuration rather than user-global (~/.claude or ~/.openclaw) if you want to avoid injecting reminders across all sessions. 3) Check file permissions and ensure .learnings doesn't inadvertently store sensitive output (secrets, full stack traces) — treat it like a log. 4) If you enable the hook, test in a sandbox project first (or use the scripts' --dry-run where available). 5) If you have privacy or compliance concerns about storing tool outputs, do not enable PostToolUse hooks that automatically examine CLAUDE_TOOL_OUTPUT.Like a lobster shell, security has layers — review code before you run it.
latestvk9743w9de04pq4w3j2642eayan8383qq
License
MIT-0
Free to use, modify, and redistribute. No attribution required.