grok-bot:ycaoC61Wx4dwlqkrlfGvw

Hermes Adversary

Adversarial review of code, plans and review pipelines, with the independent reviewer running on your own Hermes agent (a different model).

Overview

Adversarial review of code, plans, and review pipelines, with the independent reviewer running on your own Hermes agent (a different model). Findings come back proven, ranked, and either fixed or explicitly accepted as residual risk.

Contents

  • skill ship-security-gauntlet Use before shipping a vibe-coded site when you need a proven 14-attack security red team with separate test accounts, proof tables, and a one-pass fix.
  • skill ship-perf-gauntlet Use before shipping a vibe-coded site when you need a 20-check performance audit across images, frontend, database/API, and network, with Lighthouse before/after.
  • skill backward-adversarial-audit Use when the user wants a backward adversarial audit of code, a plan, or a doc: line-by-line breakage, no skim, ranked critical to trivial.
  • skill pre-mortem Use before committing to a plan, architecture, launch, or strategy: assume it already failed and list why (Klein pre-mortem).
  • skill verbalized-sampling-iterate Use when exploring designs, attack variants, or plan options and mode collapse would hide better alternatives: generate a probability-weighted sample, then iterate on the strongest tails.
  • skill review-coordinator Use when designing or running AI code-review pipelines that must debounce bursts of commits, dedupe inference, cancel stale reviews, and stay tied to a commit SHA.
  • skill agentic-ar-patterns Use when choosing which agentic adversarial-review patterns to apply to a change, pipeline, or team process. Distills the IT Revolution pattern language with credit.
  • skill hermes-security-audit Use for security questions, focused vulnerability review, or (only when explicitly asked) a full source security audit of a codebase: Cloudflare's six-phase method, with Grok hunting and Hermes as the independent cross-model verifier.
  • skill hermes-api-fleet-setup Use when wiring a Grok Bot to a Hermes agent over the native Hermes HTTP API on :8642 across a private Tailscale tailnet (the PRIMARY bridge), including restoring Tailscale after a box update, the userspace-networking forwarder, and proving the link with a health check and a PONG run.
  • skill hermes-bridge-method-chooser Use when deciding how a Grok Bot should reach Hermes: native Hermes API on :8642 over Tailscale first, SSH ask-relay only as fallback when the API is down or not enabled.
  • skill hermes-ar-relay Use when adversarially reviewing code, plans, or pipelines and the independent reviewer must be Hermes (different model), not another Grok instance: package the full AR methodology as one self-contained ask, relay it to Hermes (API runs first, SSH relay fallback), then triage findings.
  • skill hermes-ssh-relay-setup FALLBACK bridge: use when the Hermes API on :8642 is down or not enabled and a Grok Bot must reach a Desktop-primary Hermes host through the SSH ask-relay connector over a private network.
  • skill desktop-v1-runs-escape Use when a Grok connector ask to Desktop-primary Hermes dies around ~120s (-32001) — escape via host-local POST /v1/runs; also restarting a gateway from a work order and telling provider failure from host failure. Not a fleet ask path.
  • skill getting-started-hermes-adversary First conversation after importing Hermes Adversary: orient the owner, wire the Hermes bridge (API on :8642 over Tailscale first, SSH fallback), prove it with health + PONG, ask setup questions one at a time, then start cross-model AR.
  • memory Memory entries · 7 items
  • routine Hermes AR webhook Fires when the owner's Hermes host or CI POSTs a review request, so Hermes runs an independent adversarial review and the ranked findings come back in chat.
  • plugin Plugin references · 1 items

Dependencies

Attachments

No separate attachments are published for this record.