Skillv1.0.0

ClawScan security

Poetry Daily Art · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

ReviewApr 12, 2026, 4:20 PM

Verdict: Review
Confidence: medium
Model: gpt-5-mini
Summary: The skill's stated purpose (generate daily poem art) is plausible, but the implementation has several inconsistencies and privacy risks (hardcoded user paths, undeclared API/transport credentials, and reading arbitrary 'memory' files) that you should fix or clarify before installing.
Guidance: Before installing, ask the author to: (1) remove or parameterize hardcoded absolute paths (use a workspace or env variable instead of /Users/hwang/...), (2) explicitly declare the required credentials and environment variables (e.g., MINIMAX_API_KEY, TELEGRAM_TOKEN) in the skill metadata, (3) avoid searching general memory/*.md files (or document and limit exactly which files are read) because those may contain sensitive notes, (4) document the exact delivery mechanism and any third-party CLI (mmx) including where to obtain it and why it is trusted, and (5) test in a sandboxed account to confirm behavior. If the author provides a corrected version that uses relative/parameterized paths, declares env vars, and limits file access, the assessment would likely move to benign. Right now the inconsistencies and potential privacy exposures make the skill suspicious.

Review Dimensions

Purpose & Capability: noteThe skill's name/description (generate poem art cards) matches the included script and instructions: they build an image prompt from a poem title, generate an image via an image CLI, and send it as a message. However the SKILL.md advertises use with a cron and a 'message' tool (Telegram delivery) and mentions MiniMax; those required credentials/tools are not declared in the skill metadata. This is a mismatch between declared requirements (none) and what the skill actually needs.
Instruction Scope: concernSKILL.md says to read data/poem_study_progress.json (relative) and archive/poem/, but the script hardcodes absolute paths under /Users/hwang/.openclaw/workspace (progress file, memory/*.md, output dir). The script also falls back to scanning memory markdown files (memory/*.md) to find poem titles — that can read arbitrary user notes. These file accesses go beyond the documented relative paths and may expose unrelated sensitive user data.
Install Mechanism: noteThere is no install spec (instruction-only) which is low risk for supply-chain downloads. The script requires an external CLI ('mmx') and instructs 'npm install -g mmx-cli' as a prerequisite in SKILL.md, but this is not enforced or declared in metadata. Lack of an explicit install step is acceptable, but the skill depends on an external, third-party CLI (mmx) and its trustworthiness should be validated by the user.
Credentials: concernThe skill metadata declares no required environment variables or credentials, yet SKILL.md and the script require a MiniMax API/CLI and imply a MiniMax API key and a messaging transport (Telegram) for delivery. The script also reads files from a specific user's workspace (absolute /Users/hwang/...), which is a disproportionate and user-specific access pattern. Required secrets (MiniMax API key, Telegram token) should be declared and justified; currently they are missing.
Persistence & Privilege: okThe skill is not marked 'always:true' and does not modify other skills or system-wide settings. It writes output files to a workspace directory and removes previous images; this is expected for an image-generation helper and does not indicate elevated privileges.