task-commander

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt explicitly requires embedding "full content" of upstream artifacts and file contents into self-contained prompts and handoffs, which forces the agent to include any secrets (API keys, .env values, tokens) verbatim when those artifacts contain them.

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

• Third-party content exposure detected (high risk: 1.00). The research-deep workflow explicitly instructs the web_searcher agent to fetch and analyze public, user-generated web sources (blogs, Reddit, Hacker News, Stack Overflow, X/Twitter, YouTube) — located in workflows/research-deep.yml under the web-search step — and those findings feed synthesis and verification steps that can materially influence decisions, so the agent is exposed to untrusted third-party content.