The Agent Skills Directory

[SAFE]: The skill includes a Python script scripts/validate_deep_dive_artifacts.py which is used for verifying the structure of generated markdown files. The script uses standard libraries and performs no dangerous operations.
[SAFE]: The instructions proactively mitigate indirect prompt injection risks. It mandates the use of <trace-context> delimiters and explicitly instructs the agent to treat data retrieved from codebases as non-executable advisory context rather than instructions.
[SAFE]: While the skill accesses runtime configuration files (e.g., ~/.claude/settings.json), it does so for the legitimate functional purpose of resolving the active environment and applying project-specific settings.
[SAFE]: File operations are restricted to specific directories like .omc/specs/, .omx/specs/, and .agents/specs/. These paths are validated by the included script to ensure artifacts are not written to sensitive or unexpected locations.

deep-dive