Hermes Agent

PASS consensus from 3 PASS.

The quickstart links installation, hermes setup --portal, CLI or TUI launch, and observable success criteria. The Quickstart gives one ordered path, install command per platform, hermes model or hermes setup --portal for the provider, hermes or hermes --tui to launch, and an explicit "What success looks like" checklist, and the Installation page repeats the same install command without contradiction. One unambiguous path: install via curl, configure with hermes setup --portal, launch with hermes.

PARTIAL consensus from 2 PARTIAL, 1 FAIL.

Hermes-owned limits are numeric, but the provider page explicitly leaves several plan-quota semantics undocumented. Hermes-owned limits are stated exactly but only on the page that owns each subsystem, API server max_concurrent_runs default 10, webhook 30 requests per minute per route, webhook body 1 MB, delegation max_concurrent_children default 3, max_concurrent_sessions null, while no page gives a single numeric model-provider plan quota, so the provider half of the question is answered only by "upgrade your provider plan". Hermes-owned concurrent-run cap (default 10) and webhook rate limit (30/min) are stated clearly, but provider plan quotas for OpenAI Codex and xAI inference are explicitly marked not currently documented, and Google Gemini has no consumer-plan path at all.

PASS consensus from 2 PASS, 1 PARTIAL.

The API page identifies the default 10-run cap and instructs clients to back off and retry. Cause, the exact knob (gateway.api_server.max_concurrent_runs, default 10, 0 disables, negatives clamp to 0), the release valve (POST /v1/runs/{run_id}/stop), and the observability hook (GET /health/detailed reports active API runs) are all on one page, though no Retry-After header or backoff interval is documented. The cause (max_concurrent_runs cap exceeded, default 10) and the response are stated, but the correction path is incomplete, no mention of increasing the cap or waiting for in-flight runs to complete.

PASS consensus from 3 PASS.

The webhook and blueprint pages provide setup, route creation, health checks, exact HMAC headers, and a signed POST. One page carries the full loop, enable via WEBHOOK_ENABLED/WEBHOOK_PORT/WEBHOOK_SECRET, the route table, the POST URL http://your-server:8644/webhooks/<route-name>, a health probe returning {"status": "ok", "platform": "webhook"}, per-source signature schemes for GitHub, GitLab, generic V2 and legacy V1, and a signature-failure troubleshooting section. End-to-end process is fully documented: enable webhooks, define routes, verify health at GET /health, and HMAC validation for GitHub, GitLab, and generic sources.

PARTIAL consensus from 1 PASS, 2 PARTIAL.

The table names all four parameters but omits boolean encoding, defaults, request syntax, and the response shape. This one table cell is the only place in the entire docs corpus where include_children appears, no types, defaults, maximums, response shape, or example request are given, source values must be borrowed from the CLI-oriented Sessions page, and the Web Dashboard page documents a same-path GET /api/sessions that takes no parameters at all. The API endpoint explicitly documents the pagination and filter parameters, and the CLI provides concrete examples using the same parameter patterns (hermes sessions list --source telegram --limit 50).