§ 5.2 Slack agent that does real work

What you’ll have at the end

A Slack bot that reads your team’s standup channel, posts a daily summary every morning at 9am, and can answer ad-hoc questions about what was discussed. About 60 minutes once you’ve got hello-world (§5.1) working.

This is the recipe that takes the agent from “useful for me alone” to “useful for my team.”

Why this recipe matters

Standup summaries are the tractable version of the bigger question — “can the agent be a useful team mate?” If you can answer that with “yes, here’s the daily summary that everyone reads,” you’ve earned the agent’s place in the team’s workflow.

It also exercises three OpenClaw mechanisms together: channels (Slack), automation (cron), and memory (the agent remembers what was said). That’s a useful capability cluster to have working.

What you need

Hello-world working (§5.1)
A Slack workspace where you have admin rights or can get an admin to approve a bot
About 60 minutes (most of it is the OAuth dance + initial config)

Step 1 — Pair the Slack channel

If you didn’t pair Slack during onboarding, do it now:

openclaw channels add slack

This walks you through:

Creating a Slack app in the workspace (you’ll need admin or admin-blessing)
OAuth scopes (the onboarder asks for the minimum — see §6.2 signal #4)
Pairing — the bot gets added to channels you allow

Once paired, verify:

openclaw channels list
# expected: slack: paired (workspace: yourworkspace)

Step 2 — Configure the standup channel

In your ~/.openclaw/openclaw.json:

{
  "channels": {
    "slack": {
      "dmPolicy": "pairing",
      "allowFrom": ["@your-handle"],
      "allowChannels": ["#standup"]
    }
  }
}

allowChannels is your way of scoping the agent: only #standup will reach it (other channels’ messages don’t even get processed). This is the §6.1 hygienic shape.

Step 3 — Edit AGENTS.md to give it standup context

Your AGENTS.md is where you put operating instructions. Add:

## Slack standup behaviour

You read the #standup channel. People post:
- "Yesterday I worked on X"
- "Today I'll work on Y"
- "Blocker: Z"
- Sometimes ad-hoc updates throughout the day

When summarising:
- Group by team member.
- Surface blockers prominently.
- Note who's heads-down on what.
- Don't editorialise — quote where useful.
- Skip social chatter (lunch plans, etc.).

Daily summary format:
\`\`\`
# Standup summary — YYYY-MM-DD

## Yesterday's progress
- @person: [headline + 1-line context]

## Today's plan
- @person: [headline]

## Blockers (live)
- @person: [blocker — what's needed to unblock]

## Notes
- [anything significant that didn't fit above]
\`\`\`

Save AGENTS.md. The agent reads this every session.

Step 4 — Set up the daily cron

OpenClaw’s cron lets you schedule runs. Add a cron job:

openclaw cron add \
  --name "daily-standup-summary" \
  --schedule "0 9 * * 1-5" \
  --task "Read the last 24 hours of #standup and post a summary to #standup. Use the standup format from AGENTS.md."

This runs at 9am Mon–Fri (assuming your daemon’s time zone matches). Verify:

openclaw cron list

Step 5 — Manual test before letting it loose

Don’t wait until 9am to find out it’s broken. Trigger the cron manually:

openclaw cron run daily-standup-summary

The agent should:

Read recent #standup messages (using its Slack tools)
Group by person, extract blockers
Format per AGENTS.md
Post to #standup

Read what it posts. Edit AGENTS.md if the summary’s shape isn’t quite right — that’s how you tune behaviour. Re-run the cron after each edit.

Step 6 — Add the “ask the bot” capability

Beyond the daily summary, your team should be able to DM the bot questions like:

“What did Sarah say about the deploy yesterday?”

For this to work well, the agent needs memory. The default builtin engine works for short windows (a few days); for longer recall, set up §3.5 QMD memory so the agent has semantic search over historical messages.

Configuration sketch:

{
  "agents": {
    "defaults": {
      "memory": {
        "engine": "qmd",
        "indexChannels": ["slack#standup"]
      }
    }
  }
}

(Verify exact keys against memory-qmd docs — the field is in flux.)

After enabling, the agent will index #standup messages it sees. Ask it about past conversations and it’ll find the relevant turns.

What it looks like in practice

Day 1 — you set this up, manually trigger the cron once. Verify the summary lands.

Day 2 — 9am, the summary posts automatically. Your team reacts to it (“good summary,” “you missed the bit about the Postgres migration”). You note the gap; tweak AGENTS.md.

Day 7 — the summaries are good. People ask the bot questions throughout the day. The bot answers from memory. Standup-as-meeting becomes lighter because the standup-as-async-channel works.

Day 30 — you forget the bot exists because it just runs. That’s success.

Things to try beyond the basics

Cross-channel summaries. Add #deploys, #incidents. Summarise across all three.
Per-person digests. “DM me my own summary every morning” — different cron, different format.
Trigger on incident. Hooks instead of cron — when a message contains ”🔥 incident”, the bot pages the on-call.
Sentiment / blocker alerting. If 3+ people post a blocker on the same topic in a week, the bot DMs the team lead.

What we are NOT going to claim

This recipe is sourced from official Slack channel + cron docs and standard agent patterns. Sush hasn’t run it; the exact config keys above (especially allowChannels, memory.indexChannels) need verification against the live runtime. Once Sush runs it, this page updates with corrections + actual session transcripts.

Common pitfalls

Symptom	Likely cause	Fix
Cron runs but doesn’t read messages	Slack scopes too narrow	Re-pair with `channels:history` scope
Summary is generic / unhelpful	AGENTS.md guidance too vague	Be more specific in the format spec
Bot summarises social chat	”Skip social chatter” not strong enough	Add explicit examples of what to skip
Memory recall is fuzzy	Built-in engine, not QMD	Switch to QMD for cross-day recall
9am summary fires at wrong time	Daemon TZ vs Slack TZ mismatch	Set TZ explicitly in cron schedule

Slack agent that does real work