The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection as it orchestrates data flow from task definitions to sub-agents. This is an inherent property of the workflow rather than a specific vulnerability.
Ingestion points: External data such as 'Work Packages', 'Contract clauses', and 'Constitution rules' are ingested in SKILL.md and passed to sub-agents (Capos and Soldiers).
Boundary markers: The skill lacks explicit boundary markers or instructions for sub-agents to ignore potential malicious content within the Work Packages.
Capability inventory: The skill dispatches sub-agents via gangsta:the-capo and gangsta:drill-tdd which possess the capability to modify files and execute tests.
Sanitization: No sanitization or validation of the ingested task metadata is performed before interpolation into sub-agent prompts.
[COMMAND_EXECUTION]: The skill references the execution of tests via the gangsta:drill-tdd skill. This is the intended behavior for verifying software implementations and does not involve arbitrary or unauthorized command execution.

the-hit