The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection from implementation plan documents.\n
Ingestion points: The skill ingests plan files from the local filesystem to be analyzed by sub-agents.\n
Boundary markers: Not implemented. The prompts to the TDD, Reality Check, and Drift Detection agents do not include delimiters or instructions to ignore instructions embedded within the analyzed content.\n
Capability inventory: The orchestrator has access to shell execution (bash), file modification (TodoWrite), and GitHub API interactions (gh CLI).\n
Sanitization: The skill does not sanitize the contents of the plan files before processing them or sending them to external tools.\n- [DATA_EXFILTRATION]: Content from local files is uploaded to GitHub issues.\n
The skill reads the entire content of a user-specified file and uploads it to a GitHub repository.\n
While the process requires user confirmation, it could lead to accidental data exposure if a sensitive file is targeted or if a malicious plan file tricks the user into a remote submission.\n- [COMMAND_EXECUTION]: Interaction with the system shell and GitHub CLI.\n
The skill utilizes bash scripts to verify the environment and extract metadata from plan files for use in the gh CLI.\n
Although variables are appropriately quoted in the documentation, the processing of untrusted file content for CLI arguments represents a potential surface for disruption.

validating-plans