The Agent Skills Directory

[SAFE]: The skill utilizes local Python helper scripts (artifacts.py, native_reports.py) that rely exclusively on standard library modules. This avoids the risks associated with external dependency installation or remote script execution.
[SAFE]: File system operations in scripts/artifacts.py include extensive security checks to prevent unauthorized access. These include path normalization, verification that targets remain within the designated run directory, and specific protections against symlinks, junctions, and hardlink attacks.
[SAFE]: The purge functionality is designed with a least-privilege approach, requiring a specific authorization JSON record that must match the run identity and include a user-provided reference before any files are deleted.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a coordinator that ingests data from various specialist agents (e.g., test results, static analysis). While this creates an inherent attack surface for indirect prompt injection, the skill's instructions require the agent to wrap all raw reports with provenance metadata and to distinguish between independently derived evidence and echoed context, effectively mitigating the risk of the coordinator being misled by malicious specialist input.

lifecycle-method