Back to skill
Skillv0.2.0

ClawScan security

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

Scanner verdict

BenignMar 8, 2026, 8:38 PM
Verdict
benign
Confidence
high
Model
gpt-5-mini
Summary
The skill is internally consistent: it contains only read-only sqlite3 queries against the local Apple Books SQLite databases on macOS and requests no external credentials or installs — but it accesses sensitive personal data and requires Full Disk Access, so only install if you trust the author.
Guidance
This skill appears to do exactly what it says — run read-only sqlite3 queries against your local Apple Books databases — but it accesses sensitive personal content (books, highlights, notes) and requires granting Full Disk Access to the process that runs these commands. Before installing: (1) verify you trust the skill's source/homepage and the publisher (provenance in _meta.json differs from registry metadata); (2) prefer running the shown sqlite3 commands locally yourself in Terminal first so you can inspect results and confirm behavior; (3) do not grant Full Disk Access to untrusted agents or skills; (4) be aware that SEARCH_TERM/ASSET_ID are substituted directly into SQL strings — avoid entering untrusted or unexpected values if you don't want broader reads; and (5) if you need stricter isolation, run these commands in a separate macOS user account or machine where you control permissions.

Review Dimensions

Purpose & Capability
okName/description match the runtime instructions: the SKILL.md shows sqlite3 SELECT queries against Apple Books DB files under ~/Library/Containers/com.apple.iBooksX. Required binary (sqlite3) and macOS-only constraint are appropriate. Minor provenance inconsistency: _meta.json lists ownerId "local" and version 0.1.0 while registry metadata shows a different owner and version 0.2.0; this is a bookkeeping/provenance mismatch but does not contradict the skill's stated purpose.
Instruction Scope
noteAll instructions are limited to reading local SQLite files and explicitly warn against writes. No network endpoints or external uploads are used. However: (1) the skill requires Full Disk Access for the executing process — a sensitive OS permission that grants access to user data; (2) queries are built by simple string substitution (e.g., replacing SEARCH_TERM or ASSET_ID) rather than parameterized statements, so crafted inputs could cause broader/unexpected SQL reads within those databases; and (3) because this reads personal highlights/notes, it exposes sensitive content if the agent prints or transmits results. These behaviors are expected for the stated purpose but have privacy implications.
Install Mechanism
okNo install spec and no code files (instruction-only). This minimizes installation risk — nothing is downloaded or written to disk by the skill itself.
Credentials
okThe skill requests no environment variables or external credentials, which is appropriate. The notable privilege is an OS-level requirement: Full Disk Access is necessary to read the Books DB files; this is proportionate for the stated function but high-impact for privacy.
Persistence & Privilege
okalways:false (not force-included) and normal model invocation permissions. The skill does not request persistent system changes, nor does it modify other skills or system-wide config. Autonomous invocation is allowed by platform default — combined with Full Disk Access this increases potential blast radius only if you grant that permission to a compromised/untrusted skill.