The Agent Skills Directory

[PROMPT_INJECTION]: The references/evals.yaml file contains simulated prompt injection test cases. These are used strictly for evaluating the agent's robustness against adversarial inputs and are not part of the operational instructions.
[DATA_EXFILTRATION]: The skill references the official Linear API documentation URL (https://linear.app/developers/graphql). This is a well-known technology service and the reference is for informational purposes only.
[COMMAND_EXECUTION]: The skill includes Python scripts (scripts/render_plan_template.py and scripts/check_plan_template_drift.py) used for maintaining documentation templates. These scripts perform standard file read/write operations within the skill's local directory and do not execute external or untrusted code.
[DATA_EXFILTRATION]: The skill identifies user-provided specifications and bug reports as untrusted data ingestion points in SKILL.md and references/contract.yaml. The instructions explicitly include a boundary marker directing the agent to treat this input as untrusted, mitigating indirect prompt injection risks.

ce-plan