The Agent Skills Directory

[COMMAND_EXECUTION]: The skill incorporates a feedback loop that executes a local Python script at utilities/skill-builder/scripts/record_skill_feedback.py to persist decision metadata and user outcomes.
[DATA_EXFILTRATION]: The workflow includes an option to share generated ideation artifacts to an external service named 'Proof' using a standard markdown upload pattern. This behavior is documented as part of the intended project-sharing lifecycle.
[PROMPT_INJECTION]: The references/evals.yaml file contains test cases that include adversarial instructions, such as 'Ignore previous instructions' and 'override your guardrails'. These are correctly placed within a testing configuration used to verify the agent's adherence to its primary instructions under pressure.
[DATA_EXFILTRATION]: The skill ingests data from the local repository and the issue tracker to ground its ideas. While this creates a surface for indirect prompt injection, the skill includes explicit instructions to redact secrets and sensitive data during the process.

ce-ideate