swarm-executor

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The Tier 3 conflict-resolver prompt injects full file contents (git show outputs) into the model and requires producing the complete merged file, which forces the LLM to see and output any secret values present in those files verbatim.

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

• Third-party content exposure detected (high risk: 0.85). The skill ingests the swarm plan file from a user-supplied path (Step 2: “Read the plan file fully”), and that plan text is outsider-authored free-form markdown that the orchestrator then places into the subagent prompt context (Subagent Prompt Template includes “Plan: [filename]” and “Goal/Risks/Description/Acceptance Criteria/Validation” extracted from the plan).