Back to skill
Skillv1.0.3
ClawScan security
Home Fix-It · ClawHub's context-aware review of the artifact, metadata, and declared behavior.
Scanner verdict
SuspiciousMar 31, 2026, 9:40 PM
- Verdict
- suspicious
- Confidence
- medium
- Model
- gpt-5-mini
- Summary
- The skill's behavior is largely coherent with a local photo-based home-repair assistant, but some packaged files (dashboard spec, manifest, cross-sell) imply optional cloud services and credentials that conflict with the README/SECURITY claim of '100% local processing' — this mismatch and the setup prompt's filesystem operations deserve user attention before installing.
- Guidance
- What to check before installing: - The core skill appears coherent and runs locally, but inspect the SETUP-PROMPT and confirm your agent runtime will actually perform strict path canonicalization and enforce chmod/chown as described (to prevent directory escape via symlinks). - The package includes a dashboard kit that requires Supabase/NextAuth and environment variables — only provide DB or auth credentials if you intentionally enable the dashboard, and review its deployment code and access controls first. - Verify the agent truly keeps photos local (run a test with non-sensitive images and monitor network activity). The README/SECURITY claim of "100% Local Processing" conflicts with bundled dashboard instructions that assume external services. - If you enable integrations (DocuScan, dashboard), follow least-privilege principles: use per-service accounts, private storage buckets, and rotate keys; enable row-level security as recommended. - Back up any existing workspace data before letting the skill create files, and run the setup in a controlled environment (not on a sensitive production host) until you confirm behavior. If you want, I can produce a short checklist of exact runtime permissions and filesystem commands to review with your platform operator.
Review Dimensions
- Purpose & Capability
- noteThe name, README, and SKILL.md all describe a local, vision-capable home-repair assistant that stores data under a workspace 'home/' directory — that is coherent with the declared lack of required binaries and environment variables. However, the included dashboard-kit and manifest describe a Supabase-backed dashboard, authentication requirements, and environment variables for DB/auth keys. That suggests an optional cloud-backed component not reflected in the top-level claims (README/SECURITY state '100% Local Processing'). This is likely an optional feature but is an inconsistency the user should be aware of.
- Instruction Scope
- noteThe SKILL.md and SETUP-PROMPT are explicit and narrowly scoped: they tell the agent to canonicalize the workspace root, create a 'home/' directory and specific files, and re-canonicalize before each read/write. Those filesystem actions are reasonable for local persistence and within the stated purpose. The skill also includes robust prompt-injection defenses and a clear safety classification. Note: the SETUP-PROMPT instructs the agent to execute path resolution and filesystem commands — the actual agent runtime must implement those safely (canonicalization checks, no symlink escape), so verify the runtime honors those constraints.
- Install Mechanism
- okInstruction-only skill with no install spec or downloaded code. This minimizes supply-chain risk since nothing is written to disk by an installer outside the explicit SETUP-PROMPT filesystem actions.
- Credentials
- noteThe skill declares no required environment variables or credentials, which matches the local-processing claim. However, the dashboard-kit and DASHBOARD-SPEC explicitly mention Supabase, NextAuth, private storage buckets, and environment variables for keys/URLs. If you enable the dashboard or the DocuScan/manual-retrieval integration, you'll need to supply external credentials, which contradicts the packaging's 'no external transmission' messaging. Treat those dashboard components as optional plugins that require additional secrets and careful review.
- Persistence & Privilege
- okThe skill is not always-enabled and can be invoked by the user. Its persistent footprint is limited to creating/maintaining a 'home/' directory within the user's workspace (as defined by SETUP-PROMPT). That behavior is proportional for a maintenance-tracking tool, provided the runtime enforces the described canonicalization and permissions (chmod 700/600).