The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local script using the command python3 ${CLAUDE_SKILL_DIR}/scripts/cook.pyz artifact-path specs <slug>, where <slug> is a variable derived from user-provided input or file paths. This pattern presents a potential command injection vulnerability if the input is not strictly validated before being interpolated into the shell command.
[PROMPT_INJECTION]: The orchestrator reads and acts upon data in .cheese/ markdown files to drive the pipeline's logic, specifically the status: and next: fields.
Ingestion points: The skill reads .cheese/<phase>/<slug>.md files generated by automated sub-agents during the pipeline execution.
Boundary markers: No explicit delimiters or boundary markers are mentioned for the content being parsed.
Capability inventory: The skill can execute shell commands, interact with GitHub repositories via the /gh tool, and spawn sub-agents with full tool and filesystem access.
Sanitization: There is no evidence of sanitization or validation of the data read from the handoff slug files before it is used to determine the next orchestrator action.
[COMMAND_EXECUTION]: The skill automates git operations by invoking the /gh tool to commit and push changes to remote repositories. While this is a documented feature for the pipeline's terminal phase, it involves autonomous network operations and modifications to the codebase.

ultracook