Skillv1.0.4

ClawScan security

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

Scanner verdict

BenignMar 24, 2026, 9:37 AM

Verdict: Benign
Confidence: high
Model: gpt-5-mini
Summary: The skill is internally coherent for a personalized Bazi (八字) daily-analysis tool: it only uses local classic texts and an internal calendar, reads user_id/user_timezone from session context, and reads/writes a per-user profile via the platform heartbeat API — there are no unrelated credentials, external download/install steps, or hidden endpoints.
Guidance: This skill appears to do what it says: personalized daily Bazi analysis using local classic text files and a built-in calendar. Before installing, note: (1) it will read your session user_id and user_timezone and will store your four-pillar (year/month/day/hour) profile in long-term memory via the platform heartbeat API — decide if you want that personal data persisted; (2) each request will record mandatory logging fields (user_id, timezone, today_local, memory_hit, calendar_hit, heartbeat statuses, etc.), so check your privacy policy and retention rules; (3) the package includes tools to regenerate classic texts and to import calendar xlsx files — running those requires Python (pypdf, openpyxl) and local PDF/XLSX sources and is an admin action, not automatic; (4) the included classic-text cache has documented coverage gaps, so some analyses will explicitly mark missing-source caveats rather than silently fabricating; and (5) no external network endpoints or secret environment variables are requested by the skill. If you accept storing the simple profile data and the logging behavior, the skill is coherent and reasonable for its purpose.

Review Dimensions

Purpose & Capability: okName/description (daily Bazi analysis) match what the files and instructions do: use local classical texts A/B/C, a bundled daily calendar SQL, and per-user four-pillar profiles. Requiring access to session user_id/user_timezone and a heartbeat get/upsert to store/retrieve four-pillar profiles is proportionate to the stated personalized service.
Instruction Scope: noteSKILL.md explicitly instructs the agent to extract user_id and user_timezone from session context, compute today_local, call internal heartbeat events (bazi_profile_get/upsert), query the local bazi_daily_calendar, and read local classic text files. These actions are necessary for personalization and correct date-scoped lookups, but they do mean the skill will access user identifiers and persist four-pillar data to long-term memory and will log several request-level fields. The skill also mandates not fabricating missing calendar/classic data and to surface coverage gaps — this is a safe guard but may cause frequent partial results until classics are regenerated.
Install Mechanism: okNo install spec (instruction-only) and bundled scripts/data are local files. The two utility scripts (PDF text extraction and xlsx->SQL import) require common Python libraries (pypdf, openpyxl) but are maintenance tools, not invoked automatically by the skill at runtime. No external downloads, shorteners, or arbitrary extract/install steps are present.
Credentials: noteThe skill declares no required env vars or external credentials. It does depend on internal platform services (heartbeat and an internal DB/table bazi_daily_calendar) and requires writing/reading per-user profile keys and logging fields. These are proportionate to the purpose, but users should be aware that personal data (four pillars, user_id, timezone) will be stored and logged per the skill's mandatory logging contract.
Persistence & Privilege: okalways:false and normal autonomous invocation settings. The skill deliberately writes to long-term memory (heartbeat upsert) and requires persistent built-in calendar data; this is expected. There is no attempt to modify other skills or system-wide configs. The mandatory logging fields require recording user_id and other metadata — a privacy consideration but not a privilege escalation.