add-iron-proxy
Skill31k starsChanged 9 days agoMITView on GitHub
Install or refresh Iron Proxy and its official Iron Control web console for NanoClaw. Use when selecting the Iron gateway, adding its management UI, configuring credentials and grants, or restoring the proxy, approval bridge, and agent guidance.
- Reads credentials
We split this skill into the rules it gives an agent
11 rules. About 0 are common, 4 are distinctive, and 7 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 10
Use the bundled setup scripts and the source revisions in versions.json.
- Very specific: line 69
Source access fails: verify the pinned source is readable from this machine.
- Distinctive: line 72
A command times out: use the last printed stage to identify whether source download, image build, or console startup failed.
- Distinctive: line 84
Keep the database volume and encryption keys together.
- Very specific: line 90
If the folder was deleted by hand, the error prints the docker rm -f and docker volume rm commands that delete the old database.
- Very specific: line 135
Use the existing provider-auth entry point after /add-codex installs the provider.
- Very specific: line 156
After installing /add-opencode, use the same provider-auth entry point:
- Distinctive: line 184
Follow your provider's skill to restart the host and test a real reply.
- Very specific: line 190
Run the server on a fixed port, bound to 127.0.0.1 (Docker Desktop) or the Docker bridge address, often 172.17.0.1 (Linux).
- Very specific: line 194
Only that port and the OpenAI inference routes are reachable.
- Distinctive: line 198
Follow REMOVE.md. Stop only this copy's proxy and console services.
≈ “Common” is an estimate: at least 1 in 5 of the 308,232 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
Add Iron Proxy gateway
Install the official iron-control console and PostgreSQL alongside one central Iron Proxy. NanoClaw core supplies the generic gateway seam and human approval flow. Read docs/gateway-seam.md before changing the integration.
Use the bundled setup scripts and the source revisions in versions.json. The console is the upstream Rails application; its UI is not generated or copied into NanoClaw. This is a local Docker installation. Reuse this copy's recorded services and keys when refreshing it.
Install the provider payload
Copy the package's provider, approval middleware, tests, and agent guidance into their normal NanoClaw paths.
payload/src/gateway-providers/iron-proxy.ts -> src/gateway-providers/iron-proxy.ts
payload/src/gateway-providers/iron-proxy.test.ts -> src/gateway-providers/iron-proxy.test.ts
payload/src/gateway-providers/iron-proxy-allowlist.ts -> src/gateway-providers/iron-proxy-allowlist.ts
payload/src/gateway-providers/iron-proxy-local-model.ts -> src/gateway-providers/iron-proxy-local-model.ts
payload/src/gateway-providers/iron-proxy-local-model.test.ts -> src/gateway-providers/iron-proxy-local-model.test.ts
payload/src/gateway-providers/iron-proxy-approval.ts -> src/gateway-providers/iron-proxy-approval.ts
payload/src/gateway-providers/iron-proxy-approval.test.ts -> src/gateway-providers/iron-proxy-approval.test.ts
payload/src/gateway-providers/iron-proxy-credential-scope.ts -> src/gateway-providers/iron-proxy-credential-scope.ts
payload/src/gateway-providers/iron-proxy-credential-scope.test.ts -> src/gateway-providers/iron-proxy-credential-scope.test.ts
payload/src/gateway-providers/iron-proxy-transform.proto -> src/gateway-providers/iron-proxy-transform.proto
payload/container/skills/iron-proxy-gateway/SKILL.md -> container/skills/iron-proxy-gateway/SKILL.md
payload/container/skills/iron-proxy-gateway/instructions.md -> container/skills/iron-proxy-gateway/instructions.mdRegister once
The provider file makes the only product registration call. It declares idempotent sessions, typed runtime contributions, owned-resource cleanup, normalized approvals, network access, and agent guidance. NanoClaw core owns approval persistence, cards, clicks, authorization, and timeouts.
import './iron-proxy.js';Install the bridge dependencies
@grpc/grpc-js@1.14.5
@grpc/proto-loader@0.8.1Install the console and pinned proxy
Setup pulls the pinned official Iron Control image and database image, starts them on a dedicated Docker network, creates a local operator account, and registers this copy's proxy and principal through Iron's API. It stores credentials and encryption keys in owner-only files under data/session-materials/iron-control/. The database has its own persistent volume and no published port. Neither the console credentials nor its database are mounted into agents.
The installer streams stage names and elapsed-time updates. Source downloads stop after two minutes, the image build after twenty minutes, and console startup after six minutes. It never opens a Git credential prompt. The proxy is built locally from the pinned public upstream source, so installation does not require a GitHub account or access to a private proxy image. The console and build dependencies use their pinned public images. On failure, fix the reported access or service issue and rerun setup; keep existing database volumes and encryption keys together. Raw subprocess output is not streamed because it can contain credentials.
Setup builds unmodified upstream Iron Proxy and a separate NanoClaw approval front in the same image. No Iron fork or source patch is used. The front is the only network-facing listener. It authenticates session identities, inspects each HTTP request inside HTTPS tunnels, checks the allowlist, and waits for an explicit approval before forwarding to Iron on 127.0.0.1:18080. Empty, malformed, rejected or timed-out decisions fail closed. Iron's own dial-time loopback and link-local deny rules prevent DNS aliases from reaching the internal backend. Managed control-plane updates only change Iron's credential transforms; they cannot remove the front's checks.
The front builds and runs the pinned OneCLI helper in gateway-compat/onecli-summary; do not add app-specific rules. Only method, host, path, response status, the resulting OneCLI summary and whether the request sends a body or upgrade reach the approval bridge. Raw bodies, authorization headers, query strings and Iron transform traces do not. The helper sees a bounded pre-injection body prefix; the full original stream is preserved. Read the approval-presentation contract in docs/gateway-seam.md. The build tests stock Iron, the front, and the summary helper, then records an immutable image ID and source hash.
NanoClaw's approval service uses a private Unix socket on Linux. On macOS it uses loopback with mutual TLS and a proxy-only client certificate. Credentials pass directly from Iron Control to Iron Proxy.
With NANOCLAW_GATEWAY_UNCREDENTIALED_READS=true in .env, a GET or HEAD that sends no body or upgrade skips the card when no Iron credential rule could apply to its host and method. The bridge reads the rules from the proxy's config.yaml and, when managed, from Iron Control's effective config for the principal the proxy is assigned now. It reads them fresh for each such request and keeps every card for 30 seconds after its first read, because Iron applies grant changes on its own 10-second sync. Every rule it has seen stays in scope, recorded in seen-credential-rules.json beside config.yaml, because Iron can keep applying a grant that Iron Control has dropped. Delete that file and restart the host to forget revoked grants. The card also stays when Iron Control cannot be read, when the assignment changes mid-read, when a rule has an unknown shape, when config.yaml or proxy.env is newer than the running proxy (restart it), or when proxy.env sets a custom sync interval. Two gaps remain. A grant created in the moment between that read and the forward can be used without a card. So can a grant created and revoked between two reads, if Iron synced it in between, until Iron drops it. A front built before this change never attests a payload-free request, so re-run setup to rebuild it. See docs/gateway-seam.md for the data-leak trade-off.
NANOCLAW_IRON_PROXY_PORT in .env sets the internal proxy port (default 8080). Setup uses the same value for the front listener and the agent proxy URL. This does not publish a host port. Re-run setup and restart this NanoClaw copy after changing it.
NANOCLAW_IRON_CONTROL_PORT sets the console port (default 10257). Only 127.0.0.1 is published. Set it before setup if another install uses that port; use the URL printed by setup. The official image currently targets linux/amd64; Docker on Apple Silicon runs it with emulation. On another architecture, setup checks the engine before it pulls anything and stops, printing the command that enables amd64 emulation, when the engine cannot run that image.
pnpm exec tsx .claude/skills/add-iron-proxy/scripts/setup.ts --with-controlTroubleshooting
- Source access fails: verify the pinned source is readable from this machine.
The installer uses the public source in
versions.jsonand disables interactive Git prompts. Check connectivity and the pinned revision, then retry. - A command times out: use the last printed stage to identify whether source download, image build, or console startup failed. Check connectivity and Docker health before retrying. The installer terminates the timed-out process group.
- "Iron Control cannot run on this aarch64 Docker engine", or
exec format errorat the Iron Control step: the pinned console image is amd64 only. Run the printedtonistiigi/binfmtcommand against the Docker engine, or choose the OneCLI gateway; then re-run setup. The registration lives in the kernel and is gone after a reboot: re-run the command, or register it at boot (a systemd unit or your Docker host's boot script), or Iron Control restart-loops withexec format error. Setup only checks an engine running on this machine's own kernel; a VM or remote engine (Docker Desktop, Colima, aDOCKER_HOSTelsewhere) is not inspected and needs emulation enabled inside the engine. - The database exists but keys are missing: restore its matching
control.env. Keep the database volume and encryption keys together; do not generate replacement keys for an existing database.nanoclaw uninstallremoves both together: the containers carry this copy'snanoclaw-installandnanoclaw-role=gatewaylabels (gateway-owned: the gateway role and no session, so the update drain and residue reaping keep them), and the uninstaller removes their Compose project's volume and network withdata/. If the folder was deleted by hand, the error prints thedocker rm -fanddocker volume rmcommands that delete the old database; run them only if its credentials can go.
Validate
pnpm run buildpnpm exec vitest run src/gateway-providers/iron-proxy.test.ts src/gateway-providers/iron-proxy-approval.test.ts src/gateway-providers/iron-proxy-credential-scope.test.ts src/gateway-providers/gateway-provider-registry.test.ts src/gateway-approval-coordinator.test.ts .claude/skills/add-iron-proxy/scripts/control.test.ts .claude/skills/add-iron-proxy/scripts/setup.test.ts .claude/skills/add-iron-proxy/scripts/provider-credentials.test.ts .claude/skills/add-iron-proxy/scripts/credential-isolation.test.ts .claude/skills/add-iron-proxy/scripts/install-command.test.ts src/gateway-providers/iron-proxy-local-model.test.ts .claude/skills/add-iron-proxy/scripts/local-model.test.tsThe setup consumer writes NANOCLAW_GATEWAY_PROVIDER=iron-proxy only after every directive succeeds. Restart only this copy's NanoClaw service after an upgrade so its session contribution and approval bridge match the new installation. Check the proxy has synced its assigned principal before reporting the gateway ready.
The request order is front identity and allowlist → human approval → stock Iron credentials → upstream. The front also requires an explicit response decision before returning upstream data. HTTPS tunnels pin the target authority; each inner HTTP request is checked again. Streaming responses and WebSocket upgrades use this same request/response gate. Credentialed application requests use HTTPS; the approval bridge rejects plaintext HTTP destinations, except one keyless model on this machine pinned by host and port (see Serve a local model). Do not rewrite an HTTP request’s approval metadata as HTTPS to bypass that restriction. The bridge forwards every allowed HTTP request to core as a default approval request. Core uses the active agent provider’s model-domain declaration to permit model traffic without a card; other destinations retain human approval. CONNECT verifies identity; the inner HTTP request is the approval point. Existing standalone model credentials are moved into Iron Control during setup and removed from the old secret file after successful storage and grant.
Open and use the official console
Open the URL printed by setup. Give the operator the path to login.txt in the same private directory; keep passwords and API tokens out of chat and command logs. control.ts status prints only the URL and login-file location.
The pinned console provides Secrets, Credentials, OAuth Apps, and Principals. Use Secrets to create a credential, choose its source, and set its host, method, and path rules. The current console shows principals and grants but creates grants through its API. Apply a secret to this NanoClaw copy with the bundled helper:
pnpm exec tsx .claude/skills/add-iron-proxy/scripts/control.ts grant static <secret-id>Other supported kinds are gcp, aws, oauth, postgres, and hmac. The command only grants an existing credential to this install's recorded principal. Read references/control-plane.md for the exact supported UI, source links, policy example, and verification steps.
The pinned source has no general egress Policies screen or audit search. Do not promise the screens shown in newer or hosted documentation. A grant's request rules govern credential use; the local egress allowlist remains separate.
To allow another destination explicitly, the operator runs:
pnpm exec tsx .claude/skills/add-iron-proxy/scripts/setup.ts --allow-host <hostname-or-*.domain>For a standalone proxy without the console, omit --with-control on a fresh install. A recorded console is preserved on refresh. Both modes build the pinned public source. To reuse an independently built image, build the exact commit from versions.json
and label the image org.opencontainers.image.revision with that commit. Then run
pnpm exec tsx .claude/skills/add-iron-proxy/scripts/setup.ts --local-image <image>.
Setup checks the revision and records the immutable local image ID. All installs require the exact bundled approval-front label; a stock Iron image alone is not the full NanoClaw integration.
Codex authentication
Use the existing provider-auth entry point after /add-codex installs the
provider. Browser sign-in, device pairing, and API-key entry are the same for
all gateways; scripts/credential-store.ts supplies Iron's custody adapter.
Iron Control stores API keys and manages subscription refresh-token rotation
through its native broker. Agents receive only a synthetic auth.json file
with placeholder tokens. For a subscription login it also carries the login's
ChatGPT workspace id, which Codex 0.156.0 and later look up before any model
request; the id is an identifier, not a credential, and the agent can read it.
Setup records it next to the other non-secret Iron IDs. A login saved before
the id was recorded keeps the placeholder, which Codex 0.156.0 and later
reject, and the host logs a warning; Iron cannot read stored values back, so
run the command below once to sign in again. A subscription login uses a new,
dedicated Codex session; never copy personal Codex credentials. The setup flow
and provider picker stay unchanged.
pnpm exec tsx setup/index.ts --step provider-auth codexOpenCode authentication
After installing /add-opencode, use the same provider-auth entry point:
pnpm exec tsx setup/index.ts --step provider-auth opencodeThe OpenCode setup flow supports ChatGPT sign-in and API keys through Iron
Control. It installs no OneCLI service and needs no OneCLI management settings.
ChatGPT arrives as the seam's chatgpt OAuth profile: Iron creates three
records, a native broker from OpenCode's public OAuth client and refresh token,
a broker-backed bearer secret, and a separate granted secret that carries the
ChatGPT-Account-Id header; any other OAuth profile is rejected. The agent sees only placeholders. Initial sign-in and reauthentication wait for the
native broker to refresh successfully (up to two minutes) before setup continues. API keys use each backend's declared header
scheme. Setup grants the secrets to this install's principal and permits the
model hostname. Rotation and reauthentication keep IDs and grants; reauthentication
also resets a dead broker with the new refresh token. Moving a key to another
host requires confirmation and re-entering its value: Iron's update API replaces
a secret's source whenever its rules change, so a blank answer keeps a key only
on its existing host. Records use install-scoped foreign IDs, so an interrupted
save is retried on the same IDs, and missing grants are reconciled without
reading values. Before keeping or overwriting a record, setup rechecks its
ownership and rules and stops if they no longer match. Broker refresh may continue during login; a change to the broker's
client binding or the secrets' rules stops setup.
Native backends and keyless or custom HTTPS endpoints on port 443 are supported.
Setup refuses plain HTTP (except a local model, below), other ports, IP addresses
and private names such as *.home.arpa, because Iron trusts only public
certificates. Setup adds the model host to Iron's allowlist.
Follow your provider's skill to restart the host and test a real reply.
Serve a local model
A keyless model on this machine can use http://host.docker.internal:<port>/v1. Traffic still goes through Iron.
- Run the server on a fixed port, bound to
127.0.0.1(Docker Desktop) or the Docker bridge address, often172.17.0.1(Linux). Not0.0.0.0: that exposes it to your network. - Enter the URL at the provider's endpoint prompt, and answer that it needs no key.
- Restart the host.
Only that port and the OpenAI inference routes are reachable. Don't grant an Iron credential for host.docker.internal: Iron would send it over plain HTTP. A model that needs a key, or runs on another machine, needs https on a public DNS name.
Remove
Follow REMOVE.md. Stop only this copy's proxy and console services. Keep the database volume and encryption keys together when preserving data.
Provider credentials on shared model hosts
OpenCode and Codex use distinct non-secret markers. Iron replaces only the matching
header marker, so each runtime retains its own account on a shared HTTPS host.
Claude's managed model marker remains distinct. These replacements use
require: false: a request from another provider must pass without substitution.
The upstream API still rejects an unavailable or unmatched placeholder.
Refresh the installed provider payloads and rebuild the agent image before using this version. Reconnect older Codex credentials to replace their host-wide injection rules. If setup identifies a conflicting legacy or manual grant, reconnect that provider with this version (including Claude's model credential), or remove the conflicting grant in Iron Control. Stored secrets are write-only and are not silently rewritten. Setup checks direct and role grants, including inactive OAuth brokers. The installation's Iron principal remains the authorization boundary; these markers select credentials, not permissions for separate untrusted tenants.
Trending
Steady: 0 new projects took it in the last 4 weeks, none the 4 weeks before.thin data
Who's talking about qwibitai/nanoclaw
- Jordan Urbs“How to Use NanoClaw (Better Than OpenClaw?)”YouTube · Mar 6, 2026 · 30k views
- Yar Malik“OpenClaw vs ZeroClaw vs PicoClaw vs NullClaw vs NanoBot vs TinyClaw vs NanoClaw — Full Comparison”YouTube · Feb 27, 2026 · 9.8k views
- Aashish Pahwa“3 Free Alternatives To Openclaw”YouTube · Mar 11, 2026 · 3.2k views
- Mike's Ai Forge“NanoClaw + Docker: Build a 100% Secure AI Agent”YouTube · Mar 20, 2026 · 1.9k views
- Prism Labs“Anthropic Is Eating the OpenClaw Ecosystem (And They Know It)”YouTube · Mar 19, 2026 · 1.7k views
The biggest 5 of 79 mentions linking to the repository that our crawl found. Each line is their own headline.
See all 79
Talked about on GitHub, YouTube, Bluesky and Hacker News (79 times).
On GitHub · 50
- Feb 22, 2026Nanoclaw needs a skill marketplace/registry — nanocoai/nanoclaw17 reactions · 9 comments
- Mar 6, 2026🦞 OpenClaw 生态日报 2026-03-06 — duanyytop/agents-radar3 reactions
- Mar 3, 2026🦞 OpenClaw 生态日报 2026-03-03 — duanyytop/agents-radar3 reactions · 1 comment
- Mar 25, 2026feature: make it possible to select alternatives for software we installing (e.g. for openclaw) — Osmantic/ODS2 reactions · 4 comments
- Mar 3, 2026🦞 OpenClaw 生态日报 2026-03-03 — duanyytop/agents-radar2 reactions
45 more on GitHub
- May 4, 2026NanoClaw — community-scripts/DevScripts1 reaction
- Mar 7, 2026🦞 OpenClaw 生态日报 2026-03-07 — duanyytop/agents-radar1 reaction
- Mar 2, 2026🦞 OpenClaw 生态日报 2026-03-02 — duanyytop/agents-radar1 reaction
- Feb 26, 2026🦞 OpenClaw 生态日报 2026-02-26 — duanyytop/agents-radar1 reaction
- Feb 22, 2026Research: Architecture for standalone agent app with model layer, agentic loops, and messaging integration — laurentenhoor/devclaw1 reaction · 3 comments
- Feb 20, 2026fix: signal active containers to close when a scheduled task is enqueued — nanocoai/nanoclaw1 reaction · 4 comments
- Feb 14, 2026[SECURITY] [Upstream PR #216] Secret sanitization bypass via /proc and Read tool — nanocoai/nanoclaw1 reaction · 1 comment
- Jul 27, 2026Weekly upstream sync 2026-07-27 — vweaver/nanoclaw0 reactions · 1 comment
- Jun 18, 2026[Security] Non-owner members can create persistent child agents without approval in NanoClaw owner-initialized groups — nanocoai/nanoclaw0 reactions
- Jun 14, 2026[Security] Local gateway approval bypass via unauthenticated loopback webhook — nanocoai/nanoclaw0 reactions
- May 19, 2026fix(webhook) + docs: восстановление webhook + откат NanoClaw v0.1 + research — botkin-health/Botkin0 reactions
- May 9, 2026Setup script silently deletes existing groups/*/CLAUDE.md without warning or backup — nanocoai/nanoclaw0 reactions
- May 6, 2026feat: add NanoClaw v2 script — community-scripts/DevScripts0 reactions · 1 comment
- May 6, 2026feat: add NanoClaw v2 script — community-scripts/DevScripts0 reactions · 1 comment
- May 4, 2026New Script: NanoClaw — community-scripts/DevScripts0 reactions · 1 comment
- Apr 29, 2026Research: Memory approach of common non-coding Agents — 8monkey-ai/hebo-platform0 reactions · 20 comments
- Apr 27, 2026feat(channels): agent-channels:{telegram,discord,slack} layers — minimal MIT-cribbed adapters — tangle-network/starter-foundry0 reactions · 2 comments
- Apr 19, 2026feat(setup): add register-task step (carry until qwibitai#1857 lands) — sagri-tokyo/nanoclaw0 reactions
- Mar 30, 2026NanoClaw の Fork とローカル初期セットアップ — Helicoir/nanoclaw0 reactions
- Mar 30, 2026Feature: Native typing indicator via iLink sendtyping API — WEIFENG2333/ilink-bot0 reactions
- Mar 30, 2026Feature: Native typing indicator via iLink sendtyping API — co-pine/wx-robot-ilink0 reactions
- Mar 25, 2026fix: LID JID fallback via senderPn, warn on unmatched messages — nanocoai/nanoclaw-whatsapp0 reactions · 1 comment
- Mar 24, 2026Setup: IPC fix + Dev Agent configuration — yaelandau22/nanoclaw0 reactions
- Mar 23, 2026🦞 Bản tin hàng ngày hệ sinh thái OpenClaw 2026-03-23 — compasify/agents-radar0 reactions
- Mar 22, 2026🦞 Bản tin hàng ngày hệ sinh thái OpenClaw 2026-03-22 — compasify/agents-radar0 reactions
- Mar 21, 2026🦞 Bản tin hàng ngày hệ sinh thái OpenClaw 2026-03-21 — compasify/agents-radar0 reactions
- Mar 20, 2026skill: add-group-persona — per-group agent personas from WhatsApp group description — nanocoai/nanoclaw-whatsapp0 reactions
- Mar 20, 2026🦞 Bản tin hàng ngày hệ sinh thái OpenClaw 2026-03-20 — compasify/agents-radar0 reactions
- Mar 19, 2026🦞 Bản tin hàng ngày hệ sinh thái OpenClaw 2026-03-19 — compasify/agents-radar0 reactions
- Mar 18, 2026Prefer injected credentials and preserve service env — nanocoai/nanoclaw0 reactions · 2 comments
- Mar 18, 2026## Summary — Shinogreat/docs0 reactions
- Mar 17, 2026Add image support — send and receive images via Telegram — kylenessen/nanoclaw0 reactions
- Mar 16, 2026🦞 Bản tin hàng ngày hệ sinh thái OpenClaw 2026-03-16 — compasify/agents-radar0 reactions
- Mar 15, 2026🦞 Bản tin hàng ngày hệ sinh thái OpenClaw 2026-03-15 — compasify/agents-radar0 reactions
- Mar 14, 2026🦞 Bản tin hàng ngày hệ sinh thái OpenClaw 2026-03-14 — compasify/agents-radar0 reactions
- Mar 13, 2026🦞 Bản tin hàng ngày hệ sinh thái OpenClaw 2026-03-13 — compasify/agents-radar0 reactions
- Mar 12, 2026🦞 Bản tin hàng ngày hệ sinh thái OpenClaw 2026-03-12 — compasify/agents-radar0 reactions
- Mar 11, 2026🦞 Bản tin hàng ngày hệ sinh thái OpenClaw 2026-03-11 — compasify/agents-radar0 reactions
- Mar 10, 2026🦞 Bản tin hàng ngày hệ sinh thái OpenClaw 2026-03-10 — compasify/agents-radar0 reactions
- Mar 9, 2026🦞 Bản tin hàng ngày hệ sinh thái OpenClaw 2026-03-09 — compasify/agents-radar0 reactions
- Mar 8, 2026🦞 Bản tin hàng ngày hệ sinh thái OpenClaw 2026-03-08 — compasify/agents-radar0 reactions
- Mar 7, 2026🦞 OpenClaw 生态日报 2026-03-07 — compasify/agents-radar0 reactions
- Mar 5, 2026fix: Expose ANTHROPIC_MODEL from .env to agent — nanocoai/nanoclaw0 reactions · 1 comment
- Mar 4, 2026持续跟进前沿技术更新,不断优化marrow-core — zrr1999/marrow-core0 reactions
- Mar 3, 2026SKILL: unified channel media support (2-way, lazy retrival) — nanocoai/nanoclaw0 reactions · 3 comments
On YouTube · 22
- Mar 6, 2026How to Use NanoClaw (Better Than OpenClaw?) — Jordan Urbs30k views · 43 comments
- Feb 27, 2026OpenClaw vs ZeroClaw vs PicoClaw vs NullClaw vs NanoBot vs TinyClaw vs NanoClaw — Full Comparison — Yar Malik9.8k views · 33 comments
- Mar 11, 20263 Free Alternatives To Openclaw — Aashish Pahwa3.2k views · 2 comments
- Mar 20, 2026NanoClaw + Docker: Build a 100% Secure AI Agent — Mike's Ai Forgesays it tested it1.9k views · 3 comments
- Mar 19, 2026Anthropic Is Eating the OpenClaw Ecosystem (And They Know It) — Prism Labs1.7k views · 1 comment
17 more on YouTube
- Mar 19, 2026Every company needs to figure out their AI agent strategy. — NanoClaw1.4k views
- Apr 6, 202624시간 일하는 AI 개발자 만들기 — NanoClaw + Discord 셋업 가이드#NanoClaw #Claude #디스코드봇 #AI에이전트 #Docker #ClaudeCode — 빠른개발소식1.4k views · 9 comments
- Apr 23, 2026Announcing NanoClaw v2, in partnership with @VercelHQ ✨ — NanoClaw1.2k views
- Feb 14, 2026NanoClaw Lightweight Secure AI Agent Assistant for WhatsApp — Eddy Says Hi #EddySaysHi1.2k views
- May 5, 2026Building Small, Secure AI Agents with NanoClaw's Creator — NanoClaw853 views · 1 comment
- Mar 13, 2026NanoClaw is a safer OpenClaw — NEXT with John Koetsier783 views · 5 comments
- Mar 24, 2026How I move knowin' my NanoClaw agents don't hold any of my API keys — NanoClaw601 views
- May 3, 20263 Tips for Building Better AI Agent Skills — NanoClaw481 views
- Feb 28, 2026Nano Claw vs OpenClaw: Cheaper, Safer, or Just Hype? — Vibe Code Wiki367 views · 1 comment
- May 3, 2026Give your AI agent actual guardrails — NanoClaw339 views
- Feb 22, 2026[Product Reviw] 신뢰 가능한 OpenClaw? NanoClaw 🦞 — LOADING_265 views
- May 1, 2026How I Set Up My Personal AI Assistant (Claude + NanoClaw + Telegram) — Tino Muchenje | AI in Practice251 views · 2 comments
- May 6, 2026Nanoclaw: Containerized Claude Agent for Slack and WhatsApp — VibeDev205 views
- Feb 19, 2026The NEW #NanoClaw is Finally Here! — AskGuru187 views
- May 1, 2026I Built an AI Chief of Staff. One Month In. | AI in Practice — Tino Muchenje | AI in Practice167 views
- Mar 18, 2026The Future of Open-Source Contributions in the AI Age — Packet Pushers115 views
- Apr 24, 2026Best AI Agent Product Launch Video | Nanoclaw — Luceo Studio92 views
On Bluesky · 6
- Feb 12, 2026Uh so I highly recommend setting this up (I used nanoclaw: github.com/qwibitai/nan... combined with Telegram). Most of the benefits of… — @paulgp.com19 likes and reposts · 2 replies
- Feb 21, 2026Self-modifying AI agents: "There are no configuration files to learn. Just tell Claude Code what you want. Or run /customize for guided… — @narphorium.com2 likes and reposts
- Apr 6, 2026📦 qwibitai / nanoclaw ⭐ 26,620 (+91) 🗒 TypeScript A lightweight alternative to OpenClaw that runs in containers for security. Connects to… — @github-trending-js.bsky.social1 likes and repost
- Mar 16, 2026Próbáld inkább ezt: github.com/qwibitai/nan... — @kispocok.hu1 likes and repost · 1 reply
- Feb 24, 2026NanoClaw github.com/qwibitai/nan... — @annjose.com1 likes and repost
1 more on Bluesky
- Feb 24, 2026Check it out here: github.com/qwibitai/nan... — @jainamparmar.bsky.social0 likes and reposts · 1 reply
On Hacker News · 1
- Feb 27, 2026Show HN: Badge that shows how well your codebase fits in an LLM's context window88 points · 41 comments
Other versions3 projects use a version of this.
Word for word · 2
- gavrielc/nanoclaw★ 31k100% same
- nanocoai/nanoclaw★ 31k100% same
Edited · 1
- prathish-ks/isthmus★ 289% same
Wrote their own version · 0
None found.
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.
In this repository55 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- container/CLAUDE.mdCLAUDE.md
- add-anydocSkill
- add-atomic-chat-toolSkill
- add-clidashSkill
- add-codexSkill
- add-cursorSkill
- add-dashboardSkill
- add-deltachatSkill
- add-dial-numberSkill
- add-dialSkill
- add-dial-toolSkill
- add-discordSkill
- add-emacsSkill
- add-gchatSkill
- add-githubSkill
- add-imessageSkill
- add-karpathy-llm-wikiSkill
- add-linearSkill
- add-macos-statusbarSkill
- add-matrixSkill
- add-mattermostSkill
- add-mnemonSkill
- add-ollama-providerSkill
- add-ollama-toolSkill
- add-onecliSkill
- add-opencodeSkill
- add-resendSkill
- add-rtkSkill
- add-signalSkill
- add-slackSkill
- add-tavily-toolSkill
- add-teamsSkill
- add-telegramSkill
- add-vercelSkill
- add-webexSkill
- add-wechatSkill
- add-whatsapp-cloudSkill
- add-whatsappSkill
- customizeSkill
- debugSkill
- init-first-agentSkill
- learnSkill
- manage-channelsSkill
- manage-mountsSkill
- migrate-from-openclawSkill
- migrate-from-v1Skill
- migrate-memorySkill
- migrate-nanoclawSkill
- migrate-slack-agentsSkill
- setupSkill
- slack-a2a-roomsSkill
- slack-agent-flowSkill
- update-nanoclawSkill
- update-skillsSkill
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
Reviews can't be read right now.
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.

