agent-os
Skill4 starsChanged 7 months agoMITView on GitHub
Drive the conversational launch of agent-os, OR orient yourself in an existing repo. Detects which mode based on whether the repo has been bootstrapped.
- Pipes a download into a shell
- Reads credentials
2 other projects use a version of this.
Adopted by
Most recently taken up by
- 2867203296/hermes-agenttheir own version, first seen Aug 19, 2026
- christiankoertel-sketch/adam-skillstheir own version, first seen Mar 20, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- 2867203296/hermes-agent★ 0same name
- christiankoertel-sketch/adam-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
Show its rules and text
We split this skill into the rules it gives an agent
19 rules. About 0 are common, 5 are distinctive, and 14 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
14
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Distinctive: line 8
Run this check first:
- Distinctive: line 23
Before repo bootstrapping, check whether Hermes Agent itself is installed:
- Distinctive: line 31
If Hermes is missing, install it first with the official installer:
- Distinctive: line 41
Ask the user, in one short message, the minimum setup questions:
- Very specific: line 54
Use AskUserQuestion if you have it.
- Very specific: line 66
If they picked Slack: SLACK_BOT_TOKEN, SLACK_APP_TOKEN, SLACK_SIGNING_SECRET.
- Very specific: line 68
If Telegram: TELEGRAM_BOT_TOKEN.
- Very specific: line 70
If web voice: LIVEKIT_* triple, plus voice provider key (OPENAI_API_KEY or GEMINI_API_KEY).
- Very specific: line 72
If Enterprise and deployment discovery is requested: RAILWAY_TOKEN and/or DIGITALOCEAN_ACCESS_TOKEN.
- Very specific: line 85
Use the Edit tool to fill in each value the user provided.
- Very specific: line 97
After each, summarize the result for the user in one line.
- Very specific: line 107
If they want to proceed with the Hermes wiring NOW, offer to drop into Stage 2 — that's a focused 1–2 hour session where you wire …
- Very specific: line 134
Run agent-os manifest to refresh the system graph.
- Distinctive: line 136
Read ARCHITECTURE.md for the orchestrator/runtime model.
- Very specific: line 137
Read ETHOS.md for the unbreakable rules (don't edit vendor/, don't add a 15th framework).
- Very specific: line 139
Use /route --tags ... to confirm which runtime your work would dispatch to.
- Very specific: line 145
If the work spans packages, write a short plan in vault/decisions/<date>-<topic>.md first.
- Very specific: line 155
Single-state guarantee: Every channel writes through vault_memory — no per-channel state.
- Very specific: line 158
Update setup instructions with every update: If you add a tool, runtime, tier, deployment path, or workflow, update README/LAUNCH/setup …
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Mode detection
Run this check first:
test -f .env && test -d .venv && echo "bootstrapped" || echo "fresh"- fresh → run Launch Mode (below).
- bootstrapped → run Operating Mode (further down).
Launch Mode (fresh repo, drop-the-link flow)
The user just dropped this repo's URL into your session. You are the conversational installer. Your job: walk them through LAUNCH.md Path A and docs/hermes-first-install-walkthrough.md in plain English. Don't dump documentation at them — converse.
Before repo bootstrapping, check whether Hermes Agent itself is installed:
command -v hermes || true
hermes --version || true
hermes doctor || trueIf Hermes is missing, install it first with the official installer:
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bashThen run or guide hermes setup before continuing the Super Agent repo setup.
Step 1 — Confirm scope
Ask the user, in one short message, the minimum setup questions:
- Which commercial tier? Operator / Pro Operator / Enterprise.
- What business/project/offer is this for? Include first 3 workflows the agent should own.
- Which channels do they want active first? Slack? Telegram? Web text? Web voice? Any combination.
- Which model setup should Hermes use? Default suggestion: connect both GPT-5.5 and Claude Opus 4.6/4.7 for architecture/debug/security/auth/unit-test/deployment-code review, plus a cheaper worker model like DeepSeek when available.
- Where do they want to deploy? Local-only / Railway / DigitalOcean VPS / Docker Compose / Fly.io.
- Where is the shared brain? Obsidian vault path and Notion integration/database IDs, if available.
Example phrasing:
"I'll walk you through Super Agent setup like a commercial onboarding — should take 15–30 minutes depending on keys. Quick setup questions: (1) tier: Operator, Pro Operator, or Enterprise? (2) what business/project is this for and what first 3 workflows should the agent own? (3) channels: Slack, Telegram, web chat, voice, or a mix? (4) model setup: GPT-5.5 + Claude Opus 4.6/4.7 together for architecture/debug/security/auth/tests/deployment logic, plus cheaper workers? (5) deploy target: local, Railway, DigitalOcean/VPS, Docker Compose, or Fly? (6) Obsidian vault path and Notion workspace/database map?"
Use AskUserQuestion if you have it; otherwise plain prose.
Step 2 — Collect keys
Based on their channel choices, ask for ONLY the keys actually needed. Don't ask for everything in .env.example if they only picked Slack. Reference .env.example for the canonical list.
Required regardless of channel: at least one model/provider key. Preferred commercial setup is both GPT-5.5-capable OpenAI/OpenRouter access and Claude Opus 4.6/4.7-capable Anthropic/OpenRouter access, plus an optional cheaper worker model such as DeepSeek.
Shared brain fields: OBSIDIAN_VAULT_PATH, NOTION_API_KEY, and Notion database IDs for conversations, actions, decisions, agents, companies/offers, deployments, approvals, and costs when available. Notion is mandatory for the final operating system, but if credentials are not available during setup, mark Notion sync as pending rather than blocking the whole install.
Business setup fields: SUPER_AGENT_TIER, BUSINESS_NAME, BUSINESS_TYPE, FIRST_WORKFLOWS, HUMAN_APPROVAL_REQUIRED.
If they picked Slack: SLACK_BOT_TOKEN, SLACK_APP_TOKEN, SLACK_SIGNING_SECRET. Link them to https://api.slack.com/apps if they don't have an app yet.
If Telegram: TELEGRAM_BOT_TOKEN. Link to @BotFather.
If web voice: LIVEKIT_* triple, plus voice provider key (OPENAI_API_KEY or GEMINI_API_KEY).
If Enterprise and deployment discovery is requested: RAILWAY_TOKEN and/or DIGITALOCEAN_ACCESS_TOKEN. If the user asks about Orgo/managed cloud computers, explain it is optional and only collect ORGO_API_KEY when the customer/workspace needs an isolated visible desktop.
For each key the user doesn't have, ASK if they want to skip it (and disable that channel/runtime for now) or pause to grab it. Do not block.
Step 3 — Stage .env
Once keys are collected, write them to .env:
cp .env.example .env
# then edit with the values gathered aboveUse the Edit tool to fill in each value the user provided. Leave the rest blank (they're optional).
Step 4 — Bootstrap
Run, in order:
./scripts/bootstrap.sh # installs uv if missing, syncs Python, installs pnpm
uv run pytest -q # 23 tests should pass
uv run agent-os manifest # builds the system graph; writes vault/graph/system.yamlAfter each, summarize the result for the user in one line. If something fails, debug interactively — most likely a missing dep or a key that didn't make it into .env.
Step 5 — Boot Hermes
uv run agent-os bootThis is currently a stub (Stage 2 of docs/EXECUTION-PLAN.md). It returns a JSON describing the boot intent. Tell the user this honestly. The skeleton is here; the real Hermes wiring is the next session's job.
If they want to proceed with the Hermes wiring NOW, offer to drop into Stage 2 — that's a focused 1–2 hour session where you wire vendor/hermes-agent into src/agent_os/orchestrator/boot.py for real.
Step 6 — Verify single-state (when channels are live)
Once at least two channels are wired (post Stage 7 of the execution plan), run:
uv run pytest tests/integration/test_single_state_smoke.pyConfirm the test passes. This is the load-bearing acceptance criterion for the accessibility layer.
Step 7 — Hand-off
Give the user a 5-line summary:
- What's running (channels, model, deploy target).
- Tier/business/workflows and approval rules.
- What's stubbed and which stage of
docs/EXECUTION-PLAN.mdlights it up. - What the daily loops do (upgrader at 02:00, agi-audit at 02:30, heartbeat every 5 min).
- The three operating skills they have now (
/agent-os,/explain,/route). - A single concrete next move ("ship Stage 2 next session, then Stage 3 after").
Operating Mode (bootstrapped repo, day-to-day)
Run agent-os manifest to refresh the system graph. Then:
- Read ARCHITECTURE.md for the orchestrator/runtime model.
- Read ETHOS.md for the unbreakable rules (don't edit
vendor/, don't add a 15th framework). - Identify which package the work belongs to (
src/agent_os/orchestrator,src/agent_os/runtimes/<name>,src/agent_os/manifest,src/agent_os/quality,src/agent_os/upgrader,src/agent_os/channels,packages/webapp,packages/dashboard). - Use
/route --tags ...to confirm which runtime your work would dispatch to.
For commercial/multi-business decisions, read docs/commercial-packaging.md and docs/portfolio-agent-architecture.md before recommending another Hermes instance or a shared memory layout.
For Railway/DigitalOcean discovery, follow runbooks/deployment-access.md: read-only inventory first, no mutations without explicit approval.
If the work spans packages, write a short plan in vault/decisions/<date>-<topic>.md first.
For introspection ("what's running?", "who wrote this output?", "how does X connect to Y?"), use /explain.
Hard rules (apply in both modes)
- Never edit
vendor/. It breaks the upgrader. Open an upstream PR or wrap insrc/agent_os/runtimes/. - Never start another framework wrapper. New ideas go into a runtime adapter, a Hermes skill, or upstream.
- Single-state guarantee. Every channel writes through
vault_memory— no per-channel state. - Smoke tests are non-negotiable for the upgrader. A bad upstream commit silently promoted is the failure mode that takes the system down.
- Default to Hermes. Specialist runtimes are exceptions, not defaults.
- Update setup instructions with every update. If you add a tool, runtime, tier, deployment path, or workflow, update README/LAUNCH/setup docs in the same change.
- Separate serious businesses. One primary Hermes can orchestrate the portfolio, but project/customer agents need isolated vaults, secrets, deploy targets, and skills.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

