optimize-agents-md
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill reads and processes AGENTS.md and other repository files. This ingestion of untrusted repository data poses a risk of indirect prompt injection where malicious instructions in the repo could manipulate the optimization process.
- Ingestion points: AGENTS.md, AGENTS.override.md, and repo source code as described in SKILL.md and references/repo-agents.md.
- Boundary markers: The skill mentions preserving principles but lacks explicit prompt delimiters to isolate untrusted data.
- Capability inventory: File writing capabilities via the apply tool and remote code execution via agi-mode.
- Sanitization: No explicit sanitization or validation of repository content is specified.
- [REMOTE_CODE_EXECUTION]: The skill utilizes remote execution environments (via SSH or Docker) and a remote service referred to as 'Codex' for evaluating instruction sets.
- [COMMAND_EXECUTION]: The skill relies on local execution of tools such as Python, uv, and agi-mode utilities to orchestrate the optimization pipeline.
- [EXTERNAL_DOWNLOADS]: The execution flow involves interacting with external remote control planes and Codex services for authentication and evaluation tasks.
Audit Metadata