agentleFS
ARCASSystemsARCASSystems/FounderOS/skills/connect/SKILL.md

connect

Skill7 starsChanged 19 days agoMITView on GitHub

Connect a tool to Founder OS by asking in plain English. Trigger on "connect Telegram", "connect my calendar", "connect my email", "set up notifications", "hook up <tool>", "wire up <tool>", or any request to link an external tool. Also trigger on "connect gemini" and "connect elevenlabs", which the voice and mouth skills send here. Routes by connector type: env-key tools (Telegram, ElevenLabs, Gemini) get their key stored in the gitignored .env and a real reachability check; account-level MCP tools (calendar, email) are guided to the Claude Code MCP setup because the OS cannot hold their tokens; manual-link tools store a reference URL. A secret never lands in a tracked file - the writer refuses any target that is not gitignored, and the pre-commit guard blocks token-shaped strings as a backstop. A skipped connector leaves a visible marker the SessionStart brief surfaces.

  • Reads credentials

64 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use connect-apps, skill-share and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • ArogyaReddySkill

    connect-apps

    Connect Claude to external apps like Gmail, Slack, GitHub. Use this skill when the user wants to send emails, create issues, post messages, or take actions in external services.

    Also in 19 projects that use this

  • ArogyaReddySkill

    skill-share

    A skill that creates new Claude skills and automatically shares them on Slack using Rube for seamless team collaboration and skill discovery.

    Also in 17 projects that use this

  • ComposioHQSkill

    agent-deep-links

    Build, validate, and troubleshoot deep links for Codex, Cursor, VS Code, Visual Studio, and similar tools. Use when users ask for clickable links (especially in Slack) that open threads, files, folders, or app settings.

    Also in 3 projects that use this

  • ComposioHQSkill

    helium-mcp

    Search real-time news with bias scoring, get live stock/ETF/crypto data with AI analysis, ML options pricing, balanced news synthesis, and meme search via the Helium MCP server.

    Also in 3 projects that use this

  • trendinuxSkill

    twitter-algorithm-optimizer

    Analyze and optimize tweets for maximum reach using Twitter's open-source algorithm insights. Rewrite and edit user tweets to improve engagement and visibility based on how the recommendation system ranks content.

    Also in 12 projects that use this

  • LangChainSkill

    langsmith-fetch

    Fetches LangSmith traces for debugging agent behavior. Use when troubleshooting agent issues, reviewing conversation history, or investigating tool calls.

    Also in 18 projects that use this

  • +1 more from ComposioHQ
  • +5 more from ArogyaReddy

The file

We split this skill into the rules it gives an agent

8 rules. 0 are common, 2 are distinctive, and 6 are very specific to this project.

  • 0

    Common

    Most of the 56 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 6

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If core/identity.md does not exist, stop with: Founder OS not set up here.

    line 19
  2. Very specific:

    Run python scripts/connect.py registry to see the live registry.

    line 24
  3. Very specific:

    Ask for the share URL, store it where the relevant skill expects it (or in stack.json as a plain reference - never a secret), and record …

    line 59
  4. Very specific:

    Skipped connector: if the user declines or it fails, record the skip so it stays visible: python scripts/connect.py status set <name> "not …

    line 61
  5. Very specific:

    If a connected tool later throws a 401, route to reconnect-prompt, not back here.

    line 71
  6. Distinctive:

    Never report "connected" off an API 200 the user cannot confirm.

    line 84
  7. Very specific:

    Never echo a token back to the user or into a file you write.

    line 85
  8. Distinctive:

    No em dashes or en dashes in anything you write.

    line 86

“Common” is measured: kept by at least 6 in 10 of the 56 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Connect

Runs on: local-writes - stores keys in your gitignored .env and writes a status marker; env-key connectors also run a local script. On a surface that cannot write local files it guides only and says so.

Link an external tool to Founder OS by asking for it. The skill routes by how the tool authenticates and never lets a secret reach a tracked file.

Pre-flight

  • If core/identity.md does not exist, stop with: Founder OS not set up here. Run /founder-os:setup first.
  • If scripts/connect.py does not exist, stop with: Connector helper not found. Run /founder-os:update to install it.

The connector registry

Run python scripts/connect.py registry to see the live registry. Connectors fall into three classes:

  • env-key (telegram, elevenlabs, gemini): the OS can walk the steps and store the key locally. The key goes ONLY to the gitignored .env.
  • guide-only (calendar, email): account-level MCP tools. The OS does NOT control these, cannot run their OAuth, and cannot store their tokens. It hands you the Claude Code MCP-add steps and records the status. OAuth scopes are owned by the MCP / account layer - the OS does not pretend to set them.
  • manual-link (docs and similar): store a reference URL you paste. Honest that it is a bookmark, not a live integration.

Procedure

  1. Identify the tool the user named and its class (python scripts/connect.py registry). If the tool is not in the registry, say so and offer the closest class (env-key, guide-only, or manual-link).
  2. env-key class (Telegram, ElevenLabs, Gemini): one procedure for every key. Only the names change:

    SayThe line in .envWhere the key comes fromThe check after "done"
    connect telegramTELEGRAM_BOT_TOKEN@BotFather in Telegram: /newbot, copy the token, then message the new bot once so it has a chat to reply topython scripts/connect.py telegram-test, then ask "did the test message arrive in Telegram?"
    connect elevenlabsELEVENLABS_API_KEYElevenLabs profile settings, API keyspython skills/add-mouth/setup.py --engine elevenlabs, then speak the line "testing my new voice" with the quoted block in skills/add-mouth/SKILL.md step 3 (never inside quotes on the command line), then ask "did you hear it?"
    connect geminiGEMINI_API_KEY (a second free key, GEMINI_API_KEY2, is optional headroom)https://aistudio.google.com/apikey, in their own Google account, freeif the realtime voice is installed, python voice/live_server.py --models (a free call that lists the models the key opens); otherwise the key is checked the first time they start it

    The founder types the key into the file themselves. It never goes through the chat or a command. Claude Code's tool cannot read anything typed into a terminal prompt, so a "paste it on stdin" step ends with the key pasted into the chat instead. Do this:

    1. Add the line, with a placeholder that is not a secret: python scripts/connect.py add-key <tool> (for the optional second Gemini key: add-key GEMINI_API_KEY2). It adds KEY=paste-your-key-here only when the line is missing. A line already there is left exactly as it is, so running "connect" again never undoes a working key. Read its output:
      • "Added the line" or "still holds the placeholder": go on to step 2.
      • "already filled in": ask once, "You already have a key saved for this. Keep it, or put in a new one?" Keep: skip to the check. New one: go on to step 2, and in step 3 say "replace the old key" instead of "replace paste-your-key-here".
      • "REFUSED": stop and show the founder the refusal. Do not ask for the key. The writer refuses any target that is not on the allowlist (.env, .mcp.local.json) and not proven ignored, so the line cannot land in a tracked file. Where the install has version history, git supplies the proof. On a ZIP install with no git, the shipped .gitignore is read directly. Connectors work on both - a missing git is never a reason to refuse.
    2. Open the file for them: notepad .env on Windows, open -e .env on a Mac. If it does not open, tell them where it is: the file named .env in their Founder OS folder (on a Mac, Command, Shift and the full stop show hidden files in Finder).
    3. Tell them, in these plain words, with the line name from the table: "In the file that just opened, find the line that starts with GEMINI_API_KEY=. Replace paste-your-key-here with your key, keep everything before the = sign, save the file, close it, and type done here. Please do not paste the key into this chat."
    4. Wait for "done". Do not read .env back and do not print its contents. Run python scripts/connect.py key-status <tool>. It says "filled in", "still the placeholder" or "missing" and never shows the key. If it is not filled in, ask them to open .env again and check that line, without reading it out.
    5. Run the check from the table. An API accepting a call does NOT prove the founder saw or heard it. For Telegram and ElevenLabs, ask the question in the table and only mark connected after a yes. If Telegram rejects the token, the usual cause is that the placeholder is still there or the line lost its TELEGRAM_BOT_TOKEN= start.
    • If the founder pastes a key into the chat anyway: do not repeat it, do not write it anywhere yourself, and say once, kindly: "That key is now in this chat. To be safe, cancel it (the revoke step below), make a new one, and put the new one in the .env file." Then continue with the steps above using the new key.
    • Record status: python scripts/connect.py status set <tool> "connected (key in .env)".
  3. guide-only class (e.g. calendar, email):
    • This is reasoning only - no local write of a secret. Give the exact Claude Code MCP-add steps (see docs/tools-and-mcps.md). State plainly that the OS cannot store the credential or set the OAuth scopes; that lives at the account level.
    • Record status: python scripts/connect.py status set calendar "guide-only - add the MCP at the Claude Code account level".
  4. manual-link class:
    • Ask for the share URL, store it where the relevant skill expects it (or in stack.json as a plain reference - never a secret), and record status.
  5. Skipped connector: if the user declines or it fails, record the skip so it stays visible: python scripts/connect.py status set <name> "not connected - say 'connect <name>' to set up". The SessionStart brief surfaces it day to day.

How this differs from reconnect-prompt

One posture repo-wide, no conflict:

  • connect = first-time local-env setup. It MAY store a secret, but only in a gitignored local file (.env / .mcp.local.json) the OS never commits.
  • reconnect-prompt = manual re-auth after a 401 / expired token. It never stores a token and never persists to the repo.

If a connected tool later throws a 401, route to reconnect-prompt, not back here.

Revoke

Every connector documents its revoke path in the registry. Telegram: /revoke in @BotFather. ElevenLabs: delete the key in the dashboard. Gemini: delete or regenerate the key in Google AI Studio (aistudio.google.com/apikey). MCP tools: remove the MCP in Claude Code settings or revoke at the provider's security page. Surface the revoke line whenever you connect a tool.

Runtime honesty

env-key connectors need a local runtime (they run scripts/connect.py and write .env). On a web-only agent that cannot run a script or write local files, do NOT claim a connection. Guide the user through the steps and say the actual key storage and reachability check have to run in Claude Code (or any local-runtime agent pointed at the folder). guide-only and manual-link routing is pure reasoning and works on any surface.

Rules

  • A secret goes ONLY to .env or .mcp.local.json (both gitignored). Never to stack.json, a company or project .mcp.json, or any tracked file. The writer enforces this; do not work around it.
  • Never report "connected" off an API 200 the user cannot confirm. Ask "did it arrive?" for anything the API cannot prove was delivered.
  • Never echo a token back to the user or into a file you write. Never ask for a token in the chat. The founder types it into .env themselves. A token only stays out of the transcript when they follow that path.
  • No em dashes or en dashes in anything you write. Hyphens only.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.