The Agent Skills Directory

[SAFE]: The skill includes an explicit 'Prompt Injection Safety Notice' and constraints that instruct the agent to ignore any commands embedded in analyzed documents. This is a defensive best practice and the static detector alert for 'PI_IGNORE_INSTRUCTIONS' is a false positive triggered by these protective measures.
[SAFE]: Data processing is limited to reading local files using a restricted toolset (Read, Grep, Glob). The skill does not have access to the network or shell execution tools, effectively preventing data exfiltration or unauthorized system modifications.
[SAFE]: The skill maintains a legitimate purpose (compliance auditing) and its behavior matches its stated metadata and description. No suspicious dependencies, obfuscation, or persistence mechanisms were detected.
[SAFE]: Ingestion of untrusted data (Category 8 surface) is mitigated by the absence of dangerous capabilities.
Ingestion points: Local documents and configs accessed via Read, Grep, and Glob tools.
Boundary markers: Explicit instructions in the 'Constraints' and 'Prompt Injection Safety Notice' sections to treat data as untrusted.
Capability inventory: File reading only; no network, write, or shell execution tools are listed in allowed-tools.
Sanitization: Instructions to ignore embedded directives within analyzed content.

iso27001-gap