The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using uv run to interface with a backlog management tool. The command string incorporates a user-supplied variable (#N) directly into a shell execution string as a selector. Without explicit parameterization or sanitization of this input, the skill is vulnerable to command injection if a malicious string is provided as the selector argument.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from external sources (backlog items) to determine whether a project plan is approved or blocked. The ingestion process lacks explicit boundary markers or content sanitization, making the agent susceptible to instructions embedded within the backlog data that could maliciously influence the completeness assessment logic.
Ingestion points: Backlog item title, description, and acceptance criteria loaded via the backlog view command.
Boundary markers: No delimiters or isolation markers are used when interpolating external content into the assessment prompt.
Capability inventory: The skill can write back to the backlog (backlog groom) and has the authority to block or approve the subsequent engineering planning pipeline.
Sanitization: No sanitization or filtering of external content is specified before processing.

rt-ica