What the reviewer found
A working jailbreak toolkit (GODMODE/Parseltongue/ULTRAPLINIAN) that writes a jailbreak system prompt and prefill messages into the agent's own config (~/.hermes/config.yaml, prefill.json) so every future session runs with safety filters disabled, and includes a script to obfuscate trigger words to evade input classifiers. This is not a description of a technique for a gated engagement, it is a self-installing, persistent instruction override the agent is told to apply to itself and to third-party model APIs including Claude, up to and including phishing-template generation.
prompt-injection— tries to steer the agentpersistence— changes the machine outside the projectobfuscated— hides what it runsdual-use— a security tool that can be misused
What was read
The file as it ships in MilkyWay008/Hermes-OTG:
data/skills/security/godmode/SKILL.md
What the static scan said
The scan flagged 3things. The reviewer kept 3 and dismissed 0 as false.
P1Instruction-override phrasing — realAR1Tells the agent never to refuse — realAR3Nullifies safety policies — real
How this review was made
Sonnet 5 read the files above on 7 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.