Back to skill
Skillv0.1.0
ClawScan security
Paul Graham 视角 · ClawHub's context-aware review of the artifact, metadata, and declared behavior.
Scanner verdict
BenignApr 10, 2026, 4:41 PM
- Verdict
- benign
- Confidence
- high
- Model
- gpt-5-mini
- Summary
- The skill is internally consistent with its stated purpose: it is an instruction-only advisor that uses bundled reference files to apply a 'Paul Graham' lens and does not request credentials, install software, or access external endpoints.
- Guidance
- This skill appears safe from a technical-permissions perspective: it runs purely from local instructions and bundled reference files and asks for no credentials or installs. Before installing, consider: (1) legal/persona risks — it presents a 'Paul Graham' lens but explicitly states it is not Paul Graham; avoid implying authorship if you publish outputs; (2) accuracy limits — the skill advises to 'fact-check' time-sensitive claims but doesn't provide a built-in fact-check mechanism, so verify facts externally for current events; (3) review the included reference files and the SKILL.md GitHub homepage link (present in the skill metadata) to confirm you trust their content and licensing. If you need the agent to query live data or external services, expect additional permissions to be requested at that time.
Review Dimensions
- Purpose & Capability
- okName and description match the contents: the skill is an advisor-style PG-perspective lens for writing/startups/products and the included references, prompts, and workflow implement that. Minor metadata inconsistency: registry metadata lists no homepage while SKILL.md metadata contains a GitHub URL — this is an administrative mismatch but not a security problem.
- Instruction Scope
- okSKILL.md instructs the agent to classify questions, load specific local reference files (which are present), apply 2–3 'lenses', and surface blind spots. It does not instruct reading unrelated system files, accessing environment variables, or sending data to external endpoints.
- Install Mechanism
- okNo install spec and no code files beyond static references — lowest-risk model (instruction-only). Nothing will be downloaded or executed on install.
- Credentials
- okThe skill declares no required environment variables, credentials, or config paths. The runtime instructions only reference bundled reference files, so no secret or external credential access is requested or implied.
- Persistence & Privilege
- okDefault privileges (always: false, agent-invocable allowed) are used. The skill does not request permanent/system-wide presence or modification of other skills' configs.