agents-introspection

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.90). The skill explicitly instructs to "skip the ai-coord gate" (an instruction to bypass system-level coordination controls) and includes invocation-override guidance, which are commands to override environment/security controls outside the stated introspection purpose.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). At runtime, the skill’s workflow mines and inspects local Codex/Claude Code transcript JSONL (including user/assistant text fields) from ~/.codex and ~/.claude without restricting the authorship of those transcripts, so outsider-authored free text stored in those transcript files can be ingested and summarized.