writing-effect

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

• Third-party content exposure detected (high risk: 0.75). The required runtime workflow is to clone/read the Effect v4 repository into .repos/effect-v4/ (via pnpm setup:repos), and that cloned repo’s own text files (e.g., AGENTS.md/LLMS.md) are explicitly “worth a skim,” meaning outsider-authored free text from a public GitHub repo can be ingested into the agent’s LLM context through that path.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill instructs running pnpm setup:repos / bash scripts/setup-repos.sh to clone the Effect v4 repository and then treats the cloned repo's AGENTS.md and LLMS.md as authoritative for agent behavior (see the referenced setup script https://github.com/effect-ts/effect-smol/blob/main/scripts/worktree-setup.sh), so externally fetched repository content is used at runtime to control agent instructions.