Tools · Open-source assurance stackOpen source · github.com/hermes-labs-ai
The assurance stack, in working code.
Eighteen tools across audit, runtime, and evidence. Built from real failure investigations, used in live engagements. No subscriptions, no opt-in telemetry, no free-tier games. The engineering is integrating them into infrastructure you depend on. Pay us for that.
See also: upstream contributions.
Tool
Purpose
Install · Source
§ Audit & Linting06 tools
lintlang
Static linter for AI agent configs, tool descriptions, and system prompts. Zero LLM calls. Runs in CI.
rule-audit
Static prompt audit CLI: contradictions, coverage gaps, priority ambiguities.
intent-verify
Repo intent verification and spec-drift checks against declared scope.
repo-readiness
Fifteen-second launch-readiness punch-list for any public GitHub repo.
colony-probe
Defensive prompt-confidentiality audit via multi-turn probing.
scaffold-lint
Static linter for LLM prompt scaffolds.
§ Runtime03 tools
little-canary
Prompt-injection detection via sacrificial canary-model probes. The vulnerability is the signal.
suy-sideguy
Runtime policy guard for autonomous agents: process, file, and network enforcement with forensic reports.
forgetted
Mid-conversation incognito mode: checkpoint, block writes, self-clean on exit.
§ Memory & Routing02 tools
zer0dex
Dual-layer memory for AI agents: compressed index plus vector retrieval.
claude-router
Routes prompts to the right Claude tier via local embeddings.
§ Evaluation02 tools
hermes-jailbench
Jailbreak regression benchmark for LLM endpoints.
agent-convergence-scorer
Scores similarity across N agent outputs for the same task.
§ Data quality & CI gates04 tools
quick-gate-js
JavaScript / TypeScript CI quality gate with bounded auto-repair.
quick-gate-python
Python CI quality gate: Ruff, Pyright, pytest.
csv-quality-gate
CSV preflight validation for pipeline inputs.
zer0lint
Memory-extraction diagnostics for mem0 configs.
§ Research01 tool
langquant
Stateless LLM coherence without conversation history. Typed scaffold of session state, rewritten each turn by a smaller model. Reference implementation of LPCI.
18 tools total · Apache-2.0 and MIT · Full catalog on GitHub