The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill is instructed to load relevant named skills using discovered paths from the implementation plan or project directory. Dynamically loading executable components from computed paths based on external data is a risk if a project configuration file (such as prd.json or spec.md) is maliciously crafted to point to compromised tools.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and acts upon content from external project files which may contain instructions that override intended agent behavior.
Ingestion points: The skill reads spec.md, prd.json, and any other documents declared as normative by the project specification.
Boundary markers: The instructions lack explicit boundary markers or directives to treat the documentation content as data only, increasing the risk that the agent might follow instructions embedded within the PRD.
Capability inventory: The agent possesses capabilities to read and write files, execute shell commands for testing and linting, and perform Git operations including staging and committing files.
Sanitization: There are no described sanitization steps to filter or escape instructions found within the project files before they are used to drive implementation or tool selection.
[COMMAND_EXECUTION]: To fulfill its purpose, the skill executes various shell-level commands for testing, linting, and Git management. The logic and parameters for these commands are directly influenced by the contents of the processed implementation plans.

workflows-work