The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow where the agent ingests external data (specified in the 'Read first' section) and executes tasks based on that data. This creates an attack surface where instructions embedded in those files could redirect the agent's behavior.
Ingestion points: Project files, PLAN.md, or issue trackers specified in the Read first field of the goal contract.
Boundary markers: Absent. The template does not instruct the agent to use delimiters or ignore instructions found within the files it reads.
Capability inventory: The agent is authorized to modify files and execute shell commands defined in the Validate field.
Sanitization: No sanitization or validation logic is defined for the ingested data.
[COMMAND_EXECUTION]: The core mechanism of the skill requires the definition and execution of a 'Validation command' (such as pytest or pnpm test). This involves the execution of arbitrary shell strings provided within the goal contract.

goal-loop