Broaden Slack agent prompt for multi-tool workflows

[fbraza]

Summary

The current Slack agent prompt (app/pompts/slack.md) is tightly scoped to slack.chat.postMessage. We are preparing to expose additional Slack tools (DMs, threads, history lookup, summaries, etc.), so the prompt needs to be generalized to cover the broader toolset and decision policies.

What needs to change

• Reframe the "Role" and "Core objectives" sections so the agent understands it can handle multiple Slack workflows, not just posting in a channel.
• Introduce guidance on choosing between tools (e.g., when to post publicly vs DM, when to start a thread, how to handle multi-user PMs, when to read conversation history before acting).
• Document how to confirm sensitive actions, clarify when to ask for extra context, and set expectations about summarizing or referencing prior conversations.
• Add concise usage notes for upcoming tools (e.g. slack.chat.postMessage, slack.conversations.history, slack.chat.postDM, slack.chat.postThreadReply, slack.conversations.summary). These can be placeholders until we settle on exact names, but the structure should encourage adding more tools later.
• Expand the examples section to include at least:
  • sending a direct message to one user and multiple users,
  • posting follow-ups inside a thread,
  • reading recent conversation context before composing a reply,
  • summarizing a channel discussion for posting elsewhere.

Guidance / acceptance ideas

• Provide a clear "Tool selection checklist" the agent can run through before taking action.
• Capture a reusable template for confirmations (e.g., acknowledge action, channel/recipients, whether history was consulted, references to message timestamps when available).
• Ensure the prompt reminds the agent to avoid leaking tokens or user secrets when handling retrieved history.
• Organize the prompt so future tools can be slotted in under consistent headings (Tool catalog, Decision rules, Safety & etiquette, Confirmation policy, Output style, Examples).

Examples to model

• Add a "Thread response" example that shows: gather latest thread context → compose reply → call thread-reply tool → confirm outcome.
• Add a "Conversation summary" example that shows: read history for channel X → synthesize summary → publish summary in target channel or DM.

Deliverable: An updated app/pompts/slack.md capturing this broader scope and structure.