alfred-connected-apps

name: alfred-connected-apps description: Conversational management of Sir's Composio-connected third-party apps — list, connect (OAuth + API key), reconnect on expiry, disconnect, and inspect capabilities. Covers the full OAuth lifecycle including the 1h grace window after reconnect. triggers: connected apps, reconnect, oauth expired, disconnect, integrations, what apps am I connected to, connect my, add my api key version: "1.0" metadata: openclaw: emoji: "🔌"

Alfred Connected Apps

Sir manages his third-party app connections (Gmail, Calendar, Notion, GitHub, Slack, Clockify, …) through Composio. This skill teaches you to handle every conversational verb on that surface — list, connect, reconnect, disconnect, inspect — without fumbling the OAuth handshake or claiming an app is connected before Sir actually completes it.

All endpoints below live on this tenant's ctrl-api. Call them through the self MCP tool: self({endpoint, method?, body?, query?}). Never invent a ctrl_* tool name — those don't exist.

1. List connected apps

Sir says: "What apps am I connected to?" / "What's hooked up?" / "Show me my integrations."

self({ endpoint: "/api/v1/integrations" })

Response is a list of {connection_id, toolkit, status, …}. Group your reply by status so Sir sees the picture at a glance:

• ACTIVE — working. Speak by display name ("Gmail", "Google Calendar"), not the slug ("gmail", "googlecalendar").
• INITIATED — OAuth started but Sir hasn't completed it yet. Surface these and offer the link again if you still have it.
• FAILED / EXPIRED — needs reconnect. Offer to do it.

Hard rule: never list more than 10 connections in chat. If Sir has 30 apps connected, group by status and summarize counts.

2. Browse the catalog (1000+ available apps)

Sir says: "What integrations are available?" / "Can you add Linear?" / "Do you support Airtable?"

self({ endpoint: "/api/v1/integrations/catalog", query: { search: "linear" } })

The catalog has 1000+ entries. ALWAYS pass a search filter; never list everything. Cap conversational replies at 10 hits per turn.

3. OAuth connect

Sir says: "Connect my Notion." / "Hook up Linear."

self({
  endpoint: "/api/v1/integrations/connect",
  method: "POST",
  body: { toolkit_slug: "notion" },
})

Response includes connect_url (the OAuth link) + connection_id + status: "INITIATED".

Mandatory rules (these matter — get them wrong once and Sir loses trust):

1. Surface the URL verbatim. Paste it into the reply exactly as Composio returned it. Do not shorten, hyperlink-wrap, or paraphrase it.
2. Tell Sir to click it and complete the handshake. "Click here, sign in to Notion, and approve. I'll see it land as ACTIVE."
3. DO NOT claim it's connected. Status is INITIATED until Sir finishes the OAuth dance. Saying "your Notion is now connected" before that is a lie. After Sir confirms (or after a polling pass, see §6), you may say so.

4. API-key connect (non-OAuth)

Sir says: "Add my Clockify API key: " / "Here's my Tavily token: "

self({
  endpoint: "/api/v1/integrations/connect-api-key",
  method: "POST",
  body: { toolkit_slug: "clockify", credential: "<key>", auth_scheme: "API_KEY" },
})

auth_scheme defaults to API_KEY; pass BEARER_TOKEN if the toolkit needs it (PostHog, some Slack flavors).

This goes straight to ACTIVE — there's no browser handshake. Confirm immediately: "Clockify connected, Sir. Anything you'd like me to pull right away?"

Hard rules:

• Never echo the key back to Sir in your reply. Audit logs capture chat. "Got it — Clockify is connected" is correct. "Got it, your key sk-abc123 is connected" is not.
• Never log the key in a vault note, session summary, or commit message. Composio holds the credential; you must not redistribute it.

5. Reconnect on OAuth expiry — the one-call flow (#635)

Sir says: "Reconnect Calendar — the OAuth expired." / "Notion looks dead, fix it." / "Gmail keeps 401-ing."

This is the most common breakage on the fleet (Gmail re-auth happens roughly every few weeks). Use the dedicated reconnect endpoint — do NOT chain /connect + DELETE /:id manually.

Step A — find the connection_id

self({ endpoint: "/api/v1/integrations" })

Look for the toolkit Sir named (gmail, googlecalendar, notion, …) and grab its connection_id. If Sir didn't name the app but said "X isn't working", check recent logs/events:

self({ endpoint: "/api/v1/streams/events", query: { status: "error", limit: 20 } })

If you see repeated 401/403 against a Composio-backed stream, that toolkit is the culprit.

Step B — call reconnect

self({
  endpoint: "/api/v1/integrations/ca_OldConnId/reconnect",
  method: "POST",
})

Response shape:

{
  "old_connection_id": "ca_OldConnId",
  "new_connection_id": "ca_NewConnId",
  "new_connection_link": "https://composio.dev/oauth/...",
  "app": "googlecalendar",
  "expires_at": "2026-04-01T00:00:00Z",
  "cleanup_after_ms": 1745692800000,
  "grace_window_seconds": 3600,
  "instructions": "Send this link to Sir. After he completes the OAuth handshake..."
}

Step C — deliver the link to Sir on his most-immediate channel

Use the alfred-channel-delivery skill (PR #642 / KNOWN_CONTACTS.md) to pick Sir's best reachable channel right now (active Slack DM thread, ongoing voice call, the email thread you're already in, etc.). Inline the URL verbatim — same rule as a fresh OAuth connect.

If you're already in a chat turn with Sir, just paste the URL into your reply. Don't fan out across channels.

Step D — confirm ACTIVE before declaring success

After Sir completes the handshake (he'll usually say "done" or "fixed it"), verify:

self({ endpoint: "/api/v1/integrations/ca_NewConnId/capabilities" })

If status reads ACTIVE and you see a sensible action list, the new connection is healthy. Tell Sir: "Confirmed — Calendar's back. The old connection will be cleaned up in about an hour."

Step E — DO NOT preemptively disconnect the old one

The old connection stays alive during the 1-hour grace window so:

• In-flight composio_execute calls keyed off the old id keep working until streams naturally migrate.
• If Sir abandons the OAuth handshake, he isn't left with NO connection at all.

A background reaper deletes the old connection 1 hour after the new one becomes ACTIVE. You don't need to do anything. If Sir explicitly asks "kill the old one now," you may call:

self({ endpoint: "/api/v1/integrations/reconnect-cleanup?force=1", method: "POST" })

Otherwise, leave it. The reaper handles it.

Worked example

Sir: "Reconnect my Gmail — it stopped working this morning."

// 1. Find Gmail
self({ endpoint: "/api/v1/integrations" })
// → finds {connection_id: "ca_oldGmail", toolkit: "gmail", status: "EXPIRED"}

// 2. Reconnect
self({ endpoint: "/api/v1/integrations/ca_oldGmail/reconnect", method: "POST" })
// → {new_connection_link: "https://composio.dev/oauth/abc...", new_connection_id: "ca_newGmail", ...}

Reply to Sir:

Sir — your Gmail OAuth has expired. Open this link, sign in to Google, and approve:

https://composio.dev/oauth/abc...

Once you're back, tell me "done" and I'll confirm. The old connection stays live for an hour as a fallback, then gets cleaned up automatically.

After Sir confirms:

self({ endpoint: "/api/v1/integrations/ca_newGmail/capabilities" })
// → {status: "ACTIVE", stream_actions: [...], tool_actions: [...]}

Reply to Sir:

Confirmed — Gmail is back, Sir. I see the daily fetch is queued for the next pull. Nothing to do on your end.

6. Disconnect

Sir says: "Disconnect GitHub." / "Drop my Notion connection."

Before calling DELETE, walk Sir through the blast radius. The disconnect cascade removes:

• The Composio connected_account (credential gone — he'll need to re-OAuth to reconnect later).
• All stream configs backed by this connection (Gmail digests, Calendar pulls, GitHub PR streams).
• The associated Temporal schedules (no more periodic pulls).
• The skill file in ~/.openclaw/workspace/skills/alfred-composio-<toolkit>/ (you lose the per-app conversational hints) — only if this is the LAST connection of that toolkit. Two Gmail accounts? Disconnecting one keeps the skill alive for the survivor.
• The toolkit's tool-prefix from gateway.tools.allow (no more GMAIL_* / GOOGLECALENDAR_* action calls) — also only if this is the LAST connection of that toolkit (issue #658).

If Sir has a daily chore reading from this stream, name it in your warning: "Disconnecting Gmail will stop your daily 6pm digest. Sure?"

self({
  endpoint: "/api/v1/integrations/ca_OldConnId",
  method: "DELETE",
})

Response lists what was actually cleaned up. If the response shows gateway_restart_triggered: true, openclaw will restart for ~40s. Tell Sir to expect a brief unresponsive window if he was about to ask something.

A single-connection DELETE NEVER removes composio_execute from the gateway, even if it happens to be the last Composio connection on the tenant. That global teardown is intentionally a separate operation — see §6b.

6b. Disconnect EVERYTHING (global Composio reset)

Sir says: "Disconnect all my integrations." / "Wipe my Composio setup." / "Start fresh on my apps."

This is the explicit nuclear option. Confirm clearly before calling it — it deletes EVERY Composio connection, EVERY auto-configured stream + schedule, removes the entire composio_execute tool surface, and deletes every alfred-composio-* skill dir. Sir will need to reconnect each app from scratch afterwards.

self({
  endpoint: "/api/v1/integrations/disconnect-all",
  method: "POST",
})

Response shape: { disconnected_count, disconnected_ids[], failed_ids[], toolkits[], cleaned_streams[], deleted_schedules[], removed_tools[], removed_skill_dirs[], gateway_restart_triggered }. Read the failed_ids carefully — anything in there will need a manual retry. Expect a ~40s gateway restart if gateway_restart_triggered: true.

Use this only when Sir explicitly asks to wipe everything. Per-app disconnects should always go through §6.

7. Inspect a connection

Sir says: "What can my Notion integration do?" / "Why isn't my Calendar pulling?"

self({ endpoint: "/api/v1/integrations/ca_ConnId/capabilities" })

Returns:

• stream_actions[] — read-side actions (FETCH/LIST/SEARCH). Each has enabled, stream_id, schedule_interval_seconds, last_event_at, last_pull_status. If last_pull_status is a 4xx, that's likely the OAuth-expiry signal — offer reconnect.
• tool_actions[] — write-side actions (SEND/CREATE/UPDATE). Always available once composio_execute is in the gateway allowlist; no per-action toggles to read.
• stale_streams[] — stream configs whose action slug is NOT in Composio's current catalog for this toolkit. Broken on every pull. Offer the migrate-stream flow.
• deprecated — flagged true if Composio has marked the toolkit deprecated.

Use this to answer "why is X broken" precisely, not by guessing.

Anti-patterns — never do these

1. Never silently swap accounts on the same toolkit. Composio allows two Gmail connections per tenant. If Sir says "connect Gmail" and one already exists, ASK: "You already have Gmail connected as <address>. Is this for a different account, or did you mean reconnect?"
2. Never paste an API key back to Sir in chat or vault notes. (See §4.)
3. After reconnect, the old connection_id is dead in 1h — update any prose, vault notes, or stream configs that reference it. Don't tell Sir "your Gmail is ca_OldId" if it's been reconnected to ca_NewId.
4. Never claim ACTIVE status before verifying via /capabilities or /integrations. OAuth is asynchronous; trust the API, not your inference.
5. Never manually DELETE the old connection after a reconnect. The reaper does it after 1h. Forcing it early breaks in-flight tool calls and any user who isn't watching gets a worse outcome.
6. Don't try to write tool calls outside composio_execute. Toolkit actions all dispatch through that one gateway tool — there is no standalone gmail_send or notion_update_page MCP tool, despite what older skill files might suggest.

When something goes sideways

• /reconnect returns 422 "no auth_config_id" — the old connection was created via a flow that didn't persist its auth_config (rare, mostly old API-key migrations). Fall back to /connect with the toolkit_slug, then disconnect the old one manually after Sir confirms the new one works.
• /reconnect returns 404 — the connection_id is unknown to Composio (already deleted, never existed, or belongs to a different tenant). Don't retry — re-list /integrations to find the live id.
• OAuth handshake stalls — Composio's INITIATED status persists. If Sir says "I clicked it, it's done" but /capabilities still reads INITIATED, ask him to check the popup wasn't blocked. Last resort: call /reconnect again (the new entry supersedes the prior one and the reaper handles cleanup).
• composio_execute calls 401 right after reconnect — the gateway may not have picked up the new token yet. Wait 30s for the openclaw debounce + readiness probe (GET /api/v1/openclaw/ready), then retry.

Relationship to other skills

• alfred-email-channel, alfred-voice — channel-specific output. This skill is the upstream OAuth lifecycle; those skills consume the resulting capability.
• alfred-channel-delivery (PR #642 / KNOWN_CONTACTS.md) — how to pick the right channel to deliver the OAuth URL to Sir.
• The auto-generated alfred-composio-<toolkit> skills — per-app action references. Created on first connect, refreshed on /regenerate-skills. Refer to those for "what does Notion's NOTION_UPDATE_PAGE need as arguments".

Sir owns his integrations. Your job is to make the lifecycle frictionless — one chat turn for connect, one chat turn for reconnect, no juggling between Composio's dashboard and Alfred's chat.