The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from external sources which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Processes "live requirements", "relevant comments", "repository instructions", and "current issue or document" (identified in SKILL.md under Research and Decide, Step 1).
Boundary markers: None specified; the skill does not instruct the agent to use delimiters or ignore embedded instructions within these inputs.
Capability inventory: The skill is authorized to read from and write to trackers (e.g., GitHub issues) and read from the repository. It is instructed to settle implementation decisions and generate plans.
Sanitization: No sanitization or validation of the external content is mentioned before it is incorporated into the planning process.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute specific shell commands using a local script if it is available in the target environment.
Evidence: Instructions in SKILL.md (Research and Decide, Step 7) direct the use of scripts/plan-header.mjs digest and plan-header.mjs check to compute fingerprints and validate plan readiness. While the script does not ship with the skill, the execution of local files represents a command execution surface.

write-plan