Pluginv0.1.0

ClawScan security

summit entertainment · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

BenignApr 24, 2026, 7:57 PM

Verdict: benign
Confidence: high
Model: gpt-5-mini
Summary: The skill's code, hooks, and instructions are consistent with a self-improvement logging helper and do not request unrelated credentials or perform network downloads — behavior matches the stated purpose.
Guidance: This skill appears to do what it says: provide short reminders and help capture learnings to local .learnings/ files. Before enabling it, consider: (1) The activator and error-detector are opt-in hooks — only enable them if you trust the workspace/user config. (2) The error detector reads CLAUDE_TOOL_OUTPUT which may contain sensitive command output; keep the policy to redact secrets and avoid automatic logging of full transcripts. (3) scripts/extract-skill.sh will create files under the current workspace; it contains checks to prevent absolute/../ paths but review the target path before running. (4) Review/adjust file permissions (chmod +x) only if you trust these scripts. If you want to be extra cautious, install the skill in a sandboxed or project-level workspace (not a global/user-level settings file) and inspect .learnings contents before promoting entries to shared workspace files.

Review Dimensions

Purpose & Capability: okName/description (self-improvement / capture learnings/errors/feature-requests) aligns with the included files: reminder hook, small shell scripts to produce lightweight reminders, an extractor that scaffolds local skill files, and documentation. There are no unrelated requested binaries, env vars, or external services.
Instruction Scope: okSKILL.md instructs the agent to create and append to local .learnings/* files, avoid logging secrets, and optionally install an opt-in hook. The runtime instructions and scripts only read provided context (bootstrapFiles, CLAUDE_TOOL_OUTPUT) and create files under the workspace — all consistent with the stated logging/promotion purpose.
Install Mechanism: okNo install spec/payload download is present. The repository contains local scripts and hook handlers; scripts run locally and create files under the current workspace. No remote code fetches, URL downloads, or package manager installs are used.
Credentials: okThe skill declares no required env vars or credentials. The error-detector script reads CLAUDE_TOOL_OUTPUT (documented) to detect failures; this is proportional to its purpose but requires care because CLAUDE_TOOL_OUTPUT can contain sensitive output — the skill explicitly warns not to log secrets.
Persistence & Privilege: okalways:true is not set; the hook is opt-in and the handler only injects a virtual reminder file at agent bootstrap. Scripts write only to relative paths under the workspace/skills directories and the extractor validates and forbids absolute/.. paths. The skill does not modify other skills' configurations or request global privileges.