The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to run local Python scripts (capture.py, brain.py) using uv run. User input provided via $ARGUMENTS is interpolated into these shell commands (e.g., in the 'KILL' branch). This pattern is vulnerable to command injection if the input contains shell metacharacters such as ;, &, or |.\n- [PROMPT_INJECTION]: The skill demonstrates an indirect prompt injection surface by ingesting untrusted data (from $ARGUMENTS and findings.md) and passing it to multiple sub-agents (validator, poc-builder, report-writer) without using boundary markers or sanitization.\n
Ingestion points: $ARGUMENTS in SKILL.md and content from findings.md.\n
Boundary markers: Absent in the prompt templates.\n
Capability inventory: Subprocess execution via uv run, and automated evidence capture/recording.\n
Sanitization: None described or performed on the ingested content.

validate