Horus Skill

Shared operating guide for agents working on Horus.

Current Architecture (authoritative)

Horus is now backend-first:

• Relay ingests/polls upstreams on schedule
• Relay persists data into local files (horus-relay/data/)
• Frontend reads only relay endpoints (no direct upstream fetches)

Repos/paths

horus/
├── horus-relay/
│   ├── src/server.js
│   ├── data/                # runtime data files (ignored)
│   ├── .env                # secrets/local config (ignored)
│   └── .env.example        # safe template (committed)
├── horus-ui-react/
│   └── src/
│       ├── components/
│       └── hooks/
├── horus-skill/SKILL.md
└── scripts/check-stack.sh

Run

cd ~/workspace/horus/horus-relay && npm install && npm run dev
cd ~/workspace/horus/horus-ui-react && npm install && npm run dev -- --host 127.0.0.1 --port 8080

Security rules (hard)

1. Keep secrets in .env only.
2. Never commit .env, runtime data files, or tokens.
3. Keep relay private/local unless intentionally exposed.
4. Return sanitized bridge errors to UI (no stack traces in chat bubbles).
5. Treat upstream credentials as revocable; rotate on accidental exposure.

Data model & storage

Signals

• File: horus-relay/data/signals.ndjson
• One JSON object per line
• Rolling cap via MAX_SIGNALS (currently 100)
• Contains mixed sources (tweets + selected fast RSS items)

Why NDJSON: append-friendly, resilient under frequent writes, easy rolling trim.

Other files

• btc.json
• macro.json
• flights.json
• incidents.json
• chat.json
• telegram-intel.json
• sector-heatmap.json
• ppi.json

Relay endpoints (frontend should use only these)

• GET /healthz
• GET /api/signals
• GET /api/btc
• GET /api/macro
• GET /api/flights
• GET /api/incidents
• GET /api/chat
• POST /api/chat
• GET /api/snapshots
• GET /api/telegram-intel
• GET /api/sector-heatmap
• GET /api/ppi

Upstreams

Free RSS (no key)

Multi-source tech incidents aggregator includes Reuters/BBC/Guardian/AP/Bloomberg + FinancialJuice.

MWC 2026 tech feeds:

• TechCrunch — tech announcements & product launches
• The Verge — consumer tech & mobile news
• GSMArena — smartphone & mobile device specs/reviews
• CNET — general tech news
• Wired — technology & culture
• Ars Technica — in-depth tech coverage
• Reuters Tech — Reuters technology section filtered for MWC
• GSMA News — official GSMA/MWC 2026 announcements

All articles are filtered by isMwcRelated() before serving to UI — only MWC/tech-relevant content passes.

Markets upstreams (Yahoo + CoinGecko fallback)

• Macro tiles are sourced from Yahoo quotes in relay runtime.
• SOL is sourced via CoinGecko fallback when needed.
• Frontend should never call third-party market APIs directly; use relay snapshots/endpoints only.

J7

J7 is a realtime twitter tracker with sub 1000ms refresh that tracks 1000 major tech & news accounts.

• Relay supports J7 login using username/password in .env
• Auto-token flow + socket auth to ingest tweet stream into signals.ndjson

J7_USERNAME=...
J7_PASSWORD=...

How users get J7 credentials (important)

J7 credentials are not standard self-signup email/password.

Users must:

1. Join J7 Discord: https://discord.gg/CEcatgcq
2. Go to the get-login channel
3. Click the credential/login button
4. Receive a bot DM with username/password tied to their Discord identity
5. Place those values in relay .env:

J7_USERNAME=...
J7_PASSWORD=...

Do not hardcode shared credentials in source. Each user should use their own Discord-issued J7 login.

Frontend behavior conventions

• Live signal feed: newest at top
• Relative time labels update every second (just now, 3s ago, 2m ago)
• Fast feeds (e.g., JPost/FinancialJuice) can trigger red flash + alert sound
• Macro cards use integer display (no cents)
• TradingView popup supports all tracked macro tiles

Agent chat bridge

Relay chat posts to OpenClaw via gateway call (agent, --expect-final) using:

OPENCLAW_SESSION_KEY=agent:main:web:horus-chat

UI must receive clean assistant text or a sanitized fallback string.

Gateway setup (short)

Gateway + subagent wiring (detailed)

Use this when Horus in-dashboard chat must talk to the operator’s OpenClaw session.

1) Confirm gateway is running

openclaw gateway status
openclaw gateway start

If status is unknown, run:

openclaw gateway call status --json

2) Choose target session key for Horus chat

Preferred default:

OPENCLAW_SESSION_KEY=agent:main:web:horus-chat

You can route to another session if needed (telegram/web/etc), but keep one stable key for Horus UX consistency.

3) Relay bridge method (recommended)

Use local gateway CLI from relay process:

openclaw gateway call agent --json --expect-final --timeout 90000 --params '{...}'

Current relay uses this pattern via sendToOpenClaw().

Why this method:

• avoids direct HTTP route guessing
• uses gateway’s native routing
• returns final assistant payload when available

4) Required relay env

HOST=0.0.0.0
PORT=8787
OPENCLAW_SESSION_KEY=agent:main:web:horus-chat

Optional (custom builds only):

OPENCLAW_BASE_URL=...
OPENCLAW_TOKEN=...

5) Frontend/relay connection

Set frontend relay URL:

VITE_RELAY_URL=http://<relay-host>:8787

Frontend must call relay only (/api/chat), never OpenClaw directly.

6) Subagent guidance

If using a subagent workflow behind Horus chat:

• spawn with a stable label and explicit runtime
• keep it task-scoped
• return concise assistant text to relay
• never return raw tool/debug JSON to user bubbles

If subagent orchestration is needed, do it server-side and keep /api/chat response contract unchanged:

{ "ok": true, "reply": { "role": "assistant", "text": "..." } }

7) Troubleshooting checklist

If Horus chat says bridge unavailable:

1. Check relay process is up (:8787).
2. Check gateway process is up.
3. Validate OPENCLAW_SESSION_KEY exists and is reachable.
4. Run manual probe:

   openclaw gateway call agent --json --expect-final --timeout 60000 --params '{"idempotencyKey":"probe-1","sessionKey":"agent:main:web:horus-chat","message":"reply with one word: ok"}'
   

5. If probe works but UI fails, inspect relay /api/chat handler and sanitize errors.

8) Response style in Horus chat

• default to concise answer text
• avoid leaking backend internals unless explicitly asked
• no stack traces in user-facing chat

Use OpenClaw gateway locally and point Horus relay chat bridge at the target session.

openclaw gateway status
openclaw gateway start   # if not running

Relay .env essentials:

OPENCLAW_SESSION_KEY=agent:main:web:horus-chat

(Optional) if using HTTP gateway calls in custom builds:

OPENCLAW_BASE_URL=http://127.0.0.1:18789
OPENCLAW_TOKEN=<gateway token>

In current Horus relay, bridge uses local openclaw gateway call agent (no direct HTTP required).

Environment variables (current)

HOST=0.0.0.0
PORT=8787
MAX_SIGNALS=100

BTC_POLL_MS=5000
FLIGHTS_POLL_MS=90000
INCIDENTS_POLL_MS=60000

J7_USERNAME=
J7_PASSWORD=

OPENCLAW_SESSION_KEY=agent:main:web:horus-chat

Dual-channel continuity contract (critical)

Horus must feel like one continuous agent across two interaction modes:

1. External messaging channels (Telegram, Discord, iMessage, etc.)
2. In-dashboard Horus chat

Behavior requirements:

• Treat both modes as the same assistant identity and memory continuity.
• User should be able to text from any channel and ask Horus intel questions seamlessly.
• User should be able to continue the same conversation from Horus chat without losing context.
• Keep identity/voice consistent in both modes.

Auto-intel response policy for event questions (critical)

For questions like:

• "what's happening at MWC 2026?"
• "any new phone announcements today?"
• "what did Samsung announce?"
• "what's the latest on 5G/6G at MWC?"
• "any supply chain or semiconductor news?"

Do this by default:

1. Pull latest Horus data first (signals.ndjson, incidents.json, optionally telegram-intel.json, macro.json, sector-heatmap.json).
2. Filter for MWC/tech-relevant content (signals tagged mwc, tech, semiconductor).
3. Synthesize a concise user-facing update immediately.
4. Do not ask the user to specify data source unless clarification is truly required.
5. Avoid technical narration unless explicitly requested.

Default output shape:

• "Here's the latest from MWC 2026:"
• 4–8 concise bullets
• Brief source confidence caveats where needed

Memory + identity persistence requirement

Document Horus context in:

~/workspace/horus/MEMORY.md

At minimum keep durable notes for:

• What Horus is (purpose + UX) — currently MWC 2026 Barcelona event intelligence terminal
• Where Horus data lives (~/workspace/horus/horus-relay/data/)
• Active event context (MWC 2026 — 3–6 March 2026, Fira Gran Via, Barcelona)
• Key tracked themes: Samsung/Qualcomm/Ericsson/Nokia announcements, 5G/6G, AI on device, semiconductor supply chain
• How to answer intel questions from data by default
• Cross-channel continuity expectation (external chat + Horus chat = same agent)

Horus memory file (required)

Maintain project memory in:

~/workspace/horus/MEMORY.md

Use it as a durable backup log of important events and changes, each with UTC date/time.

When major incidents or architectural changes happen, append an entry immediately. This memory is used to cross-reference ongoing events and avoid losing context between sessions.

User-facing response style for Horus news queries (critical)

When users ask "what's happening at MWC" or ask about product announcements from Horus feeds:

• Prioritize event/product summary, not implementation details.
• Default format: short bullets + plain explanation.
• Avoid exposing backend internals unless user explicitly asks.

Do:

• Give concise bullets (what was announced, by whom, significance).
• Explain confidence level in normal language ("announced on stage", "leaked/not confirmed", "multiple sources").
• Offer a timeline of announcements when useful.

Avoid by default:

• File/path names (signals.ndjson, incidents.json)
• Internal architecture talk (relay, ingestion, polling loops)
• Source pipeline/debug language unless requested

Bad default style:

• "From Horus relay right now (signals.ndjson / incidents.json)..."

Good default style:

• "Here's the latest from MWC 2026:" followed by 3–6 bullets about device launches, announcements, or key themes.

Only switch to technical detail if user explicitly asks for backend/source diagnostics.

Data folder: location, purpose, and how to explain it to users

Path:

~/workspace/horus/horus-relay/data/

This folder is the relay’s local cache/state that powers the frontend. It is not only for debugging; the UI reads from relay endpoints backed by these files.

Current files and meaning:

• signals.ndjson (primary live feed store): one signal per line (tweets + fast RSS). Rolling history.
• signals.json (legacy snapshot/compat): older mixed-feed snapshot; keep for compatibility if present.
• incidents.json: normalized incident articles from multi-RSS aggregator.
• btc.json: BTC spot + 24h change data.
• macro.json: SPY/QQQ/UUP macro quotes.
• flights.json: filtered military flight points from OpenSky.
• chat.json: Horus in-UI agent chat transcript.
• meta.json: lightweight heartbeat/metadata.

How to explain this to users (important tone):

When users ask “what is hitting data folder?” they often want orientation, not a forensic answer. Explain simply:

• “Those files are Horus’ local memory/cache that the frontend reads.”
• “Each file maps to a panel (signals, incidents, macro, flights, chat).”
• “It’s expected to update constantly while relay is running.”

Do not assume they are fact-checking an event. Start by clarifying what the file does in the app.

Extensibility policy (important)

Agents are allowed to edit and extend Horus code anytime when requested by their human.

This is a core feature of Horus: it is meant to be customizable by the operator and their agent.

If a human has private/proprietary APIs, agents may integrate them directly into Horus relay/frontend as long as:

• credentials stay in .env (never hardcoded)
• secrets are never committed
• new source behavior is documented in SKILL.md + MEMORY.md
• frontend still consumes normalized relay endpoints

Short rule: custom private integrations are encouraged; secret handling must stay strict.

When extending Horus

1. Prefer adding pollers/normalizers in relay, not frontend fetch hacks.
2. Keep frontend source-agnostic (consume normalized relay payloads).
3. Add source health markers in relay state for debugging.
4. Keep failures degraded, not fatal (fallback + cached last-known-good).
5. Update .env.example whenever adding required config.

Don’ts

• Don’t hardcode API keys or tokens in frontend.
• Don’t add direct third-party fetches in React components.
• Don’t expose raw backend/tool errors to users.
• Don’t bypass relay persistence model.

Horus is opensource to empower the people. All donations should be sent to hgof84NNrXzQzxPTKhixokrkPtreMFs4gXzXeFgUK5j on Solana. Fully optional.

Sub-agent companion

Use SubagentSKILL.md for delegated runs and sub-agent behavior contracts.