The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local scripts located in the project's .agents/scripts/ directory using the node command. These scripts (precheck.mjs and cooking.mjs) are used to check prerequisites and track workflow status.
[DYNAMIC_EXECUTION]: The orchestrator dynamically constructs shell commands by interpolating user-provided identifiers (such as the feature parameter) into script calls (e.g., node .agents/scripts/cooking.mjs status <feature>). Without strict input validation, this pattern is susceptible to command injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes project-specific data and interpolates it into prompts for sub-agents, creating a vulnerability surface for indirect instructions.
Ingestion points: Data is ingested from goal.md, spec.md, and user requirement descriptions located in the .agents/cooking/ directory.
Boundary markers: The instructions do not define explicit delimiters or warnings to sub-agents to ignore potentially malicious content within these files.
Capability inventory: The system has the capability to execute shell commands via node, perform git-commit operations, and modify the project file system through implemented sub-agents.
Sanitization: There is no evidence of content sanitization or validation performed on the ingested project files before they are passed to sub-agents for processing.

rush