The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes task instructions and shell command definitions directly from the card-loop/board.md file within the active repository. This data-driven execution model allows content in the board file to influence agent behavior and command execution, posing a risk if the file is modified by external actors (e.g., via pull requests).
Ingestion points: The agent reads shell command strings from the <TEST_CMD>, <TYPECHECK_CMD>, and <LINT_CMD> fields in the card-loop/board.md markdown table.
Boundary markers: Absent; there are no clear delimiters or "ignore instructions" warnings when extracting command strings from the configuration table.
Capability inventory: The skill is authorized to perform shell command execution (for tests and tools), write files (creating cards and plans), and manage Git operations (branching, committing, and pushing).
Sanitization: Absent; the instructions direct the agent to read the value in the table row and execute it as a command without validation.
[COMMAND_EXECUTION]: The /tek-do-card command is designed to invoke arbitrary shell tools based on repository-specific settings provided in the markdown templates. While this is the intended purpose of the developer workflow, it provides an execution pathway for any command string retrieved from the project metadata.

tek-skill