writing-for-agents
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation and modification of agent-controlling instructions (AGENTS.md, skills), which represents a potential attack surface for indirect prompt injection if the processed content is malicious.
- Ingestion points: SKILL.md and references/project-instructions.md describe reading existing instructions and sources of truth to formulate new instructions.
- Boundary markers: The skill mandates a "Project-instruction gate" requiring explicit human approval for all proposed changes or diffs before any file writes occur.
- Capability inventory: The skill is designed to write and update instruction files across the project scope.
- Sanitization: The skill relies on manual human verification and approval of the final text/diff as a primary defense against malicious content.
- [SAFE]: The skill promotes security best practices by explicitly instructing to keep secrets and private payloads out of instructions. It also prohibits performing side-effect operations such as package installations, service configurations, or deployments during the documentation review and verification process.
Audit Metadata