The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local script scripts/audit_plugin_paths.py for its auditing phase. The script's content is missing from the provided files, making its behavior unverifiable.
[COMMAND_EXECUTION]: The skill invokes an external evaluation script evaluate.py located in a different plugin directory (agent-agentic-os). This creates a dependency on external code that was not included in the audit.
[DATA_EXFILTRATION]: Artifacts in evals/.lock.hashes contain hardcoded absolute paths (e.g., /Users/richardfremmerlid/...), exposing details of the developer's local environment.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes and modifies external files provided by the user.
Ingestion points: Arbitrary files targeted for auditing and fixing via the Read tool and audit scripts.
Boundary markers: Missing markers to separate untrusted file content from the agent's instructions.
Capability inventory: Bash (script execution), Write, Edit, and Read tools.
Sanitization: No mechanisms are defined to sanitize or validate the content of the files before they are read or edited.

fix-plugin-paths