The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the eval command to execute shell output generated by the bundled fluxctl tool during the setup-review phase. This is used to dynamically set environment variables for the session, but it executes tool-generated shell code at runtime.
[PROMPT_INJECTION]: The workflow incorporates an automated "Fix Loop" that ingests and acts upon feedback from an external AI backend to modify and commit code changes without human review or confirmation.
Ingestion points: External review feedback captured in the REVIEW_RESPONSE variable from SKILL.md and workflow.md.
Boundary markers: The instructions lack specific delimiters or safety prompts to prevent the agent from following malicious instructions embedded in the review feedback.
Capability inventory: Across SKILL.md and workflow.md, the skill permits file system writes ("Fix code"), test execution, and repository modifications (git add, git commit).
Sanitization: There is no evidence of sanitization or human-in-the-loop validation for the issues parsed from external responses before they are acted upon.

flux-opencode-epic-review