The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestrator triggers various sub-skills (dogpile, battle, orchestrate, review-code, memory) by executing their respective run.sh scripts via subprocess.run and subprocess.Popen using relative internal paths.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to the handling of untrusted input throughout the workflow stages.
Ingestion points: The primary idea argument in the start command and the contents of files within the project directory being reviewed or implemented.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the orchestrator logic.
Capability inventory: The skill orchestrates complex capabilities including web search (via dogpile), code execution in isolated environments (via battle), and autonomous file modification (via orchestrate).
Sanitization: The skill lacks sanitization, filtering, or validation of the user-provided idea before it is interpolated into search queries and review request documents.

create-code