Skill Details
Runtime
Any model
Example user message
[userMessage]: Audit this skill package before I install it: --- FILE: SKILL.md --- --- name: weather-lookup description: Look up current weather for a named city using a public weather API and return a short summary. --- # Weather Lookup 1. Take [CITY_NAME] as input. 2. Call GET https://api.open-meteo.com/v1/forecast?... with the city's coordinates (resolved via a static lookup table in coords.json). 3. Summarize temperature, conditions, and wind in 2-3 sentences. --- FILE: coords.json --- {"Cairo": [30.04, 31.24], "London": [51.51, -0.13]} Source: personal GitHub repo, MIT license, single commit, no dependencies, requested tools: network (open-meteo.com only), no file writes, no secrets.
Example agent response
## Trust Receipt
1. Artifact identity: weather-lookup, single-commit GitHub source, MIT
license, reviewed as pasted (not independently pinned/hashed — flagged
under residual uncertainty).
2. Declared purpose vs. observed capability: declared = weather lookup via
public API; observed = exactly one outbound domain
(api.open-meteo.com), one static local data file, no writes, no
secrets, no execution beyond the described HTTP call. Aligned.
3. Static findings: none in the hidden-instruction, secret-access, or
destructive-command classes. One low-severity note: coords.json is a
static table, not a live geocoding call — acceptable, not a defect.
4. Capability map: read (coords.json only), network (open-meteo.com
only), no write/execute/identity/persistence requested.
5. Provenance: single-commit source, MIT license present, no dependency
chain, no runtime fetch of remote code.
6. Minimum permission envelope: network access scoped to
api.open-meteo.com only; no filesystem write access required.
7. Route: ALLOW.
8. Rollback: uninstall skill file and coords.json; no persistent state to
remove.
9. Residual uncertainty: artifact was reviewed as pasted text, not a
hash-verified pull from the named repo — re-verify identity before
treating this receipt as covering a future update.
By purchasing this skill, you agree to our terms of service
Before you install someone else's AI skill, plugin, or agent package,
know what it actually does. This auditor treats every file as untrusted,
never executes the target, and scans for hidden instructions, secret
access, remote-execution chains, and undeclared capabilities. Outputs a
9-field Trust Receipt and one route: Allow, Allow With Restrictions,
Quarantine, Reject, or Incomplete. A clean scan is never called "safe."
...more
Added 22 hours ago
