The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes task descriptions and requirements from external markdown files (e.g., plan.md, spec.md) and interpolates them into its execution logic. Maliciously crafted task descriptions could attempt to override agent instructions or execute unauthorized actions within the automated task loop.
Ingestion points: conductor/workflow.md, conductor/tracks/{trackId}/spec.md, and conductor/tracks/{trackId}/plan.md (SKILL.md)
Boundary markers: The skill uses markdown task list markers (e.g., [ ], [~], [x]) to identify tasks, but does not explicitly sanitize or validate the text content within those descriptions.
Capability inventory: The skill has the capability to perform git commits (git commit), write to local configuration and progress files, and execute shell commands for testing (npm test, pytest).
Sanitization: No explicit sanitization, escaping, or filtering of task description content is performed before processing.
[COMMAND_EXECUTION]: The skill executes shell commands for version control and automated testing as part of the TDD workflow.
Evidence: Uses git add -A and git commit -m "..." to automatically manage project state during task completion.
Evidence: Invokes project-defined test runners such as npm test or pytest based on the project environment.

conductor-implement