aradotso/hermes-labyrinth-observability

Hermes Labyrinth

Skill by ara.so — Daily 2026 Skills collection.

Hermes Labyrinth is a read-only observability dashboard plugin for Hermes Agent. It turns autonomous agent runs into a navigable map of crossings (prompts, tool calls, tool results, failures, model switches, subagents, approvals, memory hits, redactions, context compression, cron runs) with exportable evidence. It is not a chat UI — it is a black-box recorder for agents moving through unknown work.

---

Install

Plugin Directory Install

mkdir -p ~/.hermes/plugins
git clone https://github.com/stainlu/hermes-labyrinth.git ~/.hermes/plugins/hermes-labyrinth

Start or restart the Hermes dashboard:

hermes dashboard

If the dashboard is already running, rescan plugins without restarting:

curl http://127.0.0.1:9119/api/dashboard/plugins/rescan

Open the dashboard in your browser and select the Labyrinth tab.

Optional Theme

mkdir -p ~/.hermes/dashboard-themes
cp ~/.hermes/plugins/hermes-labyrinth/theme/hermes-labyrinth.yaml ~/.hermes/dashboard-themes/

---

What Labyrinth Tracks

| View | Contents | |---|---| | Journey index | Recent CLI, dashboard, gateway, cron, and delegated work | | Labyrinth map | Ordered crossings through a selected agent journey | | Inspector | Input, output, duration, status, evidence, guideposts per crossing | | Guideposts | Generated observations backed by local evidence | | Skill atlas | Bundled, optional, external, and user skill inventory | | Cron gate | Scheduled autonomy, next runs, last failures, workdirs | | Model ferry | Model/provider transitions across sessions | | Reports | Redacted Markdown and JSON exports for one journey |

---

API Surface

All endpoints are read-only. The plugin API is served by Hermes dashboard at:

http://127.0.0.1:9119/api/plugins/hermes-labyrinth/

Endpoints

GET /api/plugins/hermes-labyrinth/health
GET /api/plugins/hermes-labyrinth/journeys
GET /api/plugins/hermes-labyrinth/journeys/{journey_id}
GET /api/plugins/hermes-labyrinth/journeys/{journey_id}/crossings
GET /api/plugins/hermes-labyrinth/skills
GET /api/plugins/hermes-labyrinth/cron
GET /api/plugins/hermes-labyrinth/guideposts
GET /api/plugins/hermes-labyrinth/reports/{journey_id}.json
GET /api/plugins/hermes-labyrinth/reports/{journey_id}.md

Example: Fetch All Journeys

curl http://127.0.0.1:9119/api/plugins/hermes-labyrinth/journeys | jq .

Example: Fetch Crossings for a Journey

JOURNEY_ID="your-journey-id"
curl "http://127.0.0.1:9119/api/plugins/hermes-labyrinth/journeys/${JOURNEY_ID}/crossings" | jq .

Example: Export a Journey Report as Markdown

JOURNEY_ID="your-journey-id"
curl "http://127.0.0.1:9119/api/plugins/hermes-labyrinth/reports/${JOURNEY_ID}.md" > report.md

Example: Export a Journey Report as JSON

JOURNEY_ID="your-journey-id"
curl "http://127.0.0.1:9119/api/plugins/hermes-labyrinth/reports/${JOURNEY_ID}.json" > report.json

Example: Health Check

curl http://127.0.0.1:9119/api/plugins/hermes-labyrinth/health

---

Python API Client Examples

The plugin backend lives at dashboard/plugin_api.py. You can also call the HTTP API from any language. Here are Python examples:

import urllib.request
import json

BASE = "http://127.0.0.1:9119/api/plugins/hermes-labyrinth"

def get_journeys():
    with urllib.request.urlopen(f"{BASE}/journeys") as r:
        return json.loads(r.read())

def get_crossings(journey_id: str):
    with urllib.request.urlopen(f"{BASE}/journeys/{journey_id}/crossings") as r:
        return json.loads(r.read())

def get_report_json(journey_id: str):
    with urllib.request.urlopen(f"{BASE}/reports/{journey_id}.json") as r:
        return json.loads(r.read())

def get_report_md(journey_id: str) -> str:
    with urllib.request.urlopen(f"{BASE}/reports/{journey_id}.md") as r:
        return r.read().decode("utf-8")

# Usage
journeys = get_journeys()
for j in journeys:
    print(j["id"], j.get("status"), j.get("started_at"))

Iterate Crossings and Inspect Tool Calls

import urllib.request
import json

BASE = "http://127.0.0.1:9119/api/plugins/hermes-labyrinth"

def inspect_tool_crossings(journey_id: str):
    with urllib.request.urlopen(f"{BASE}/journeys/{journey_id}/crossings") as r:
        crossings = json.loads(r.read())

    for crossing in crossings:
        if crossing.get("type") == "tool_call":
            print(f"Tool: {crossing['tool']}")
            print(f"  Status:   {crossing.get('status')}")
            print(f"  Duration: {crossing.get('duration_ms')}ms")
            print(f"  Input:    {json.dumps(crossing.get('input', {}))[:200]}")
            print()

inspect_tool_crossings("your-journey-id")

Download and Save All Reports for Recent Journeys

import urllib.request
import json
import pathlib

BASE = "http://127.0.0.1:9119/api/plugins/hermes-labyrinth"
OUT = pathlib.Path("./labyrinth-reports")
OUT.mkdir(exist_ok=True)

with urllib.request.urlopen(f"{BASE}/journeys") as r:
    journeys = json.loads(r.read())

for j in journeys[:10]:  # last 10 journeys
    jid = j["id"]
    try:
        with urllib.request.urlopen(f"{BASE}/reports/{jid}.json") as r:
            (OUT / f"{jid}.json").write_bytes(r.read())
        with urllib.request.urlopen(f"{BASE}/reports/{jid}.md") as r:
            (OUT / f"{jid}.md").write_bytes(r.read())
        print(f"Saved reports for {jid}")
    except Exception as e:
        print(f"Failed {jid}: {e}")

---

JavaScript / Frontend API Examples

The frontend plugin bundle lives in dashboard/dist/. If you're extending the UI or writing a custom integration:

const BASE = "http://127.0.0.1:9119/api/plugins/hermes-labyrinth";

async function fetchJourneys() {
  const res = await fetch(`${BASE}/journeys`);
  if (!res.ok) throw new Error(`HTTP ${res.status}`);
  return res.json();
}

async function fetchCrossings(journeyId) {
  const res = await fetch(`${BASE}/journeys/${journeyId}/crossings`);
  if (!res.ok) throw new Error(`HTTP ${res.status}`);
  return res.json();
}

async function fetchReportMarkdown(journeyId) {
  const res = await fetch(`${BASE}/reports/${journeyId}.md`);
  if (!res.ok) throw new Error(`HTTP ${res.status}`);
  return res.text();
}

// Example: log all failed crossings in the most recent journey
async function logFailures() {
  const journeys = await fetchJourneys();
  if (!journeys.length) return;
  const crossings = await fetchCrossings(journeys[0].id);
  const failed = crossings.filter(c => c.status === "failure" || c.status === "error");
  console.table(failed.map(c => ({
    type: c.type,
    tool: c.tool ?? "-",
    duration_ms: c.duration_ms,
    error: c.error?.slice(0, 120),
  })));
}

logFailures();

---

Build & Development

The frontend is built from src/parts/*.js + src/labyrinth.css into dashboard/dist/. The demo index.html is generated with content-hash query strings.

# Build dashboard/dist and index.html
npm run build

# Run reproducibility and parse checks
npm run check

# Run browser smoke tests (headless Chrome)
npm run smoke

# Smoke-test the deployed GitHub Pages demo
npm run smoke:live

# Run all tests (build checks, fixture tests, smoke)
npm test

Full Test Suite

npm test

Runs:

• Reproducible build checks for dashboard/dist and index.html
• Frontend JavaScript parse checks
• Backend Python parse checks
• API normalization fixture tests (including numeric Hermes timestamps)
• Packed-artifact and dead-control regressions
• Headless Chrome smoke coverage for map modes, route changes, search, dataset switching, and threshold filter

---

Repository Layout

.
├── dashboard/
│   ├── manifest.json        # Hermes dashboard plugin manifest
│   ├── plugin_api.py        # Read-only API over local Hermes state
│   └── dist/                # Generated dashboard plugin bundle
├── docs/
│   ├── CONCEPT.md
│   ├── DESIGN_BRIEF.md
│   └── FUNCTIONAL_SPEC.md
├── scripts/
│   ├── build-plugin.mjs     # Builds dashboard/dist and index.html
│   ├── smoke-demo.mjs       # Browser smoke test for public demo
│   ├── test-plugin-api.py   # Fixture tests for API normalization
│   └── verify.mjs           # Local verification checks
├── src/
│   ├── demo/                # GitHub Pages demo source
│   ├── parts/               # Ordered frontend source chunks
│   └── labyrinth.css        # Frontend CSS source
├── theme/
│   └── hermes-labyrinth.yaml
├── index.html               # Generated GitHub Pages demo
└── package.json

---

Architecture

Hermes local state
  ├─ state.db sessions/messages
  ├─ skills directories
  └─ cron config
        ↓
dashboard/plugin_api.py
        ↓
/api/plugins/hermes-labyrinth/*
        ↓
src/parts/*.js + src/labyrinth.css
        ↓ npm run build
dashboard/dist/*
        ↓
Hermes dashboard tab: Labyrinth

---

Data Policy (Important)

• Read-only by design: does not start, stop, resume, mutate, or create Hermes sessions.
• Secret redaction: applied to previews and reports.
• Unknown fields: stay unknown — not silently dropped.
• Reports: generated from local Hermes state only.
• Public demo: uses sample/mocked data — not live telemetry.

---

Common Patterns

Check Plugin Health Before Querying

import urllib.request, json

def is_labyrinth_healthy() -> bool:
    try:
        with urllib.request.urlopen(
            "http://127.0.0.1:9119/api/plugins/hermes-labyrinth/health",
            timeout=3
        ) as r:
            data = json.loads(r.read())
            return data.get("status") == "ok"
    except Exception:
        return False

if not is_labyrinth_healthy():
    print("Labyrinth plugin not reachable — is `hermes dashboard` running?")

Filter Journeys by Type

import urllib.request, json

BASE = "http://127.0.0.1:9119/api/plugins/hermes-labyrinth"

with urllib.request.urlopen(f"{BASE}/journeys") as r:
    journeys = json.loads(r.read())

# Filter to only cron-triggered journeys
cron_journeys = [j for j in journeys if j.get("origin") == "cron"]

# Filter to only failed journeys
failed_journeys = [j for j in journeys if j.get("status") in ("failure", "error")]

Summarize Crossing Types in a Journey

from collections import Counter
import urllib.request, json

BASE = "http://127.0.0.1:9119/api/plugins/hermes-labyrinth"

def summarize_journey(journey_id: str):
    with urllib.request.urlopen(f"{BASE}/journeys/{journey_id}/crossings") as r:
        crossings = json.loads(r.read())
    counts = Counter(c.get("type", "unknown") for c in crossings)
    total_ms = sum(c.get("duration_ms", 0) for c in crossings)
    print(f"Journey {journey_id}: {len(crossings)} crossings, {total_ms}ms total")
    for ctype, n in counts.most_common():
        print(f"  {ctype}: {n}")

---

Troubleshooting

Plugin tab not appearing in dashboard

1. Confirm clone location: ls ~/.hermes/plugins/hermes-labyrinth/dashboard/manifest.json
2. Restart hermes dashboard or rescan: curl http://127.0.0.1:9119/api/dashboard/plugins/rescan
3. Check manifest.json is valid JSON.

API returns 404

• Verify the plugin loaded: curl http://127.0.0.1:9119/api/plugins/hermes-labyrinth/health
• Confirm hermes dashboard is running on port 9119.

Build artifacts out of date

npm run build
# Then verify reproducibility
npm run check

Smoke tests fail locally

• Requires headless Chrome/Chromium available in PATH.
• Run npm run smoke for local, npm run smoke:live for the deployed Pages demo.

Reports contain redacted fields

This is expected. Labyrinth applies secret redaction to all previews and exports by design. Raw values are only accessible inside the Hermes process itself.

Numeric timestamps in API responses

The API normalization layer handles numeric Hermes timestamps automatically (covered by fixture tests in scripts/test-plugin-api.py). If you consume the API directly, expect either ISO 8601 strings or Unix epoch integers in timestamp fields.

---

Links

• Live demo: https://stainlu.github.io/hermes-labyrinth/
• Hermes Agent: https://github.com/NousResearch/hermes-agent
• Release v0.1.0: https://github.com/stainlu/hermes-labyrinth/releases/tag/v0.1.0
• License: MIT