The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data that could contain malicious instructions designed to influence subagent behavior.
Ingestion points: The skill ingests user-provided input via the $ARGUMENTS variable and researches the codebase using foreground subagents, the results of which are then used to build worker prompts.
Boundary markers: The instructions do not define specific delimiters or "ignore instructions" warnings to wrap the interpolated research results or user arguments when passing them to worker agents.
Capability inventory: The orchestration involves spawning background agents (Agent tool) that have the capability to execute shell commands (e.g., npm test, pytest, gh pr create) and commit code.
Sanitization: There is no evidence of sanitization, escaping, or filtering of the ingested content before it is interpolated into the prompts for background workers.
[SAFE]: The skill implements a structured multi-phase workflow (Research, Plan, Execute) which includes a manual approval step (ExitPlanMode) before any worker agents are spawned, providing a human-in-the-loop security checkpoint.

batch