secrets-scan

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The agent prompt requires returning a JSON "content" field with raw file snippets (example: "pwd=x") and explicitly forbids redaction, so the LLM must include secret values verbatim in its output.

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

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md Phase 1–2, the workflow builds a list of git-tracked files from the current repository via git ls-files and then spawns subagents that read/classify contents of those files, meaning outsider-authored free text in a user-supplied repo is ingested at runtime.