permission-slip-openclaw-skill-slack

name	permission-slip-openclaw-skill-slack
description	Check unread Slack, read channels and threads, search, and post messages through a Slack workspace connected to Permission Slip. Use when the user says things like "check my Slack", "any unread messages?", "what's in

Slack (via Permission Slip)

This skill lets you act on the user's Slack workspace by driving the permission-slip CLI. You never talk to the Slack API directly — every action goes through Permission Slip, which enforces the user's approval policy.

This skill contains no code of its own: it is a thin shim over the permission-slip CLI. All authentication, approval enforcement, and default-workspace selection live in Permission Slip and its built-in slack connector.

Defaults

• "Check my Slack" → list unread conversations. Run slack.list_unread (no parameters). It returns conversations with unread messages, each with an unread count and a latest-message preview. Coverage caveat: Slack only exposes unread counts for DMs and group DMs — public/private channels are not included (Slack's API doesn't surface channel unread counts). The response's notes field spells this out. To check a specific channel, call slack.read_channel_messages with oldest set to that channel's last_read_ts. Relay this DM/channel limitation to the user rather than implying "nothing unread" means fully caught up.
• Workspace selection: omit --instance. Permission Slip auto-selects the user's default Slack instance. Only pass --instance if the user explicitly names a second workspace.

Preflight (run once per session, before the first action)

1. permission-slip whoami — confirm this agent is registered. If not, tell the user to register (permission-slip register ...) and stop.
2. permission-slip connectors — confirm slack is available. If it's missing, the user hasn't connected a Slack workspace yet; point them at the connector setup docs and stop.

Intent -> action mapping

User says	Action	Params
"check my Slack", "any unread?"	slack.list_unread	none
"list my channels"	slack.list_channels	—
"what's in #channel?"	slack.read_channel_messages	{"channel":<id>,"limit":<n>}
"read that thread"	slack.read_thread	{"channel":<id>,"thread_ts":<ts>}
"search Slack for X"	slack.search_messages	{"query":"<text>"}
"who is @user?"	slack.get_user_profile	{...}
"message #channel"	slack.send_message	message params
"DM so-and-so"	slack.send_dm	DM params
"react with :emoji:"	slack.add_reaction	reaction params
"delete that message"	slack.delete_message	HIGH risk
"archive #channel"	slack.archive_channel	HIGH risk

Risk tiers (governs the approval gate)

• Low (often auto-approve via standing approvals): list_unread, list_channels, read_channel_messages, read_thread, search_messages, list_users, get_user_profile, send_message, send_dm, add_reaction, remove_reaction, pin_message, mark_read, upload_file, schedule_message.
• Medium: create_channel, set_topic, rename_channel, invite_to_channel, remove_from_channel, update_message, unpin_message.
• High (destructive — always confirm with the user first): delete_message, archive_channel.

Note: posting (send_message / send_dm) is rated low by the connector, so it may auto-approve. Even so, always confirm the exact target and wording with the user before sending — a posted message can't be quietly taken back.

How to run an action

permission-slip request \
  --action slack.list_unread \
  --params '{}'

The CLI prints JSON. Two outcomes — rely on what the CLI returns, don't assume from the action name:

• Auto-approved: the result includes the action output. Unread/listing results include channel_id (and ts / thread_ts for messages) — use those for follow-up reads or replies.
• Pending approval: the CLI returns a request id in a pending state. Tell the user plainly: "That needs your approval — I've sent the request to Permission Slip; I'll know once you approve it." Then poll with permission-slip request-status <id> and report the outcome. Never claim a message was sent, edited, or deleted until the status is approved and executed.

Presenting results

• Summarize unread briefly (channel/DM, count, latest preview). Don't dump raw JSON unless asked.
• Use emojis to make the list easier to scan — e.g. 💬 channel, 👤 DM, 🧵 thread, 🔵 unread count, 📎 attachment. Keep it tasteful: one or two icons per line, not a wall of emoji.
• If there's nothing unread, say the workspace is all caught up — don't retry.
• Deletes and channel archives are destructive. Always confirm the exact target with the user before submitting, and make clear it requires approval.
• On error, surface the connector's message verbatim and suggest the fix rather than guessing.

Constraints

• One concern per request. For several messages, make one request each.
• Treat medium/high-risk actions as approval-gated and communicate the wait; for low-risk posts, still confirm content and target before sending.