What the reviewer found

LLM/prompt-injection pentest guide: all flagged payloads (hidden instructions, /etc/shadow exfil message, system-prompt leak prompt) are example attacks to test against a target LLM app, not instructions aimed at the installer's own agent.

What was read

The file as it ships in uphiago/recon-skills:

• redteam/hunt-llm-ai/SKILL.md

What the static scan said

The scan flagged 8things. The reviewer kept 0 and dismissed 8 as false.

• P1Instruction-override phrasing — false positive
• P2Hidden instructions — false positive
• P6Asks the agent to reveal its instructions — false positive
• E1Sends data to an external URL — false positive
• PE3Reaches for credential files — false positive
• SSRF1Cloud metadata endpoint — false positive
• NETMakes network calls — false positive
• SHRuns shell commands — false positive

How this review was made

Sonnet 5 read the files above on 6 September 2026 and answered three questions: is it dangerous to whoever installs it, is each scanner finding real, and what should the installer know. The verdict is bound to the file's hash; when the file changes, it is scanned afresh and reviewed again. A script that changes while the definition does not is not re-reviewed — that is a known gap. How the scan and the review work.

Produced by agentmods.dev · Sonnet 5 · 6 September 2026