The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by incorporating untrusted data from the local environment into the agent's instruction pipeline.
Ingestion points: The skill reads project configuration from CLAUDE.md (Step 1) and accepts arbitrary input from the user via $ARGUMENTS (Parse Arguments section).
Boundary markers: Data retrieved from external sources is interpolated directly into the session context and generated files without explicit delimiters or instructions to the model to ignore embedded commands.
Capability inventory: The skill possesses the Bash, Read, and Write tools, providing a significant capability surface if the model obeys instructions injected through the processed data.
Sanitization: While the mode argument is validated against a whitelist, the {sprint_id} is extracted as a raw token and used in file paths, and the content of CLAUDE.md is processed without validation.
[COMMAND_EXECUTION]: Uses the system shell to perform environment discovery and filesystem maintenance.
Evidence: Employs grep and sed to parse project metadata and uses mkdir -p to ensure the existence of the .claude_plans directory.

goal