Hermes Labs

A guided, project-local first experience for Claude Code that helps beginners start making something real.

Recall locally stored agent memories through the Fidelis MCP server.

Local recovery anchors for Claude Code and Codex sessions, plus evidence-gated evaluation prompts

Receipt-bound completion rail for Claude Code: SessionStart injects the completion contract, Stop runs the cached fast gate advisory-only, and PreToolUse validates supported literal Git/PR commands. See docs/CONFIGURATION.md#hook-command-scope for parser and repository-context limits.

Jailbreak regression benchmark for LLM endpoints with repeatable known-pattern attacks and deterministic scoring

Assess named artifacts with evidence-cited Hermes Rubric results and explicit coverage limits.

Lint agent-facing interfaces: instructions, prompts, MCP and function-tool definitions, and parameter schemas. Local, deterministic audits with advisory feedback after supported Claude Code edits.

Blocks a Claude Code turn when a local Little Canary server rejects the submitted prompt. Also installs on Codex CLI, which must trust the hook before it screens.

Run Quick Gate CLI 0.4.0 as one deterministic JavaScript/TypeScript gate and read its gate-result/v1 verdict

Python quality gate CLI for Ruff, Pyright, and pytest with bounded auto-repair and escalation artifacts

Bounded public search evidence with source-explicit supersearch.search.v1 receipts (pinned hermes-supersearch 0.11.1); evidence, not answers