--- name: rounds description: Slack-native autonomous work discovery — scan channels, score messages, act on high-signal, reflect in diary. license: MIT tier: any category: intelligence autoInvoked: false dependencies: [] relatedSkills: [bethesda, cos, inbox] requirements: env: [] integrations: [] --- # /rounds — Slack-Native Autonomous Work Discovery The doctor does rounds. You do rounds. Every Slack channel is a patient — check vitals, diagnose, treat, move on. Every message is a potential quest lead, research thread, or connection waiting to be made. This is the skill that turns Slack from a bulletin board into a living nervous system. You don't just READ channels — you RESPOND, CONNECT, BUILD ON, CHALLENGE, and PICK UP WORK from what you find. ## The Core Loop ``` READ all channels -> SCORE each message -> ACT on high-signal -> REFLECT in diary -> CAPTURE learnings -> REPEAT ``` ## Arguments - `[focus]` — Optional channel filter (e.g., `triage`, `ideas`, `analytics`, `deploy`) - `deep` — Read thread replies, not just top-level messages. Slower but finds buried signal. --- ## Phase 1: Channel Scan (parallel — hit all channels at once) Scan your team's active channels — coordination, triage, and discussion channels. Read the highest-signal ones first. Hit them all in parallel; don't serialize the reads. ### Tier 1: Coordination (read first, always) ``` slack_read_channel(channel_id: "{{SLACK_CONTROL_CENTER_CHANNEL}}", limit: 20) ``` ### Tier 2: Ideas & Strategy (highest compound value) ``` slack_read_channel(channel_id: "{{SLACK_IDEAS_CHANNEL}}", limit: 15) ``` ### Tier 3: Triage (user pain signals) ``` slack_read_channel(channel_id: "{{SLACK_TRIAGE_CHANNEL}}", limit: 10) ``` If your team runs separate triage channels per area or per project, scan each of them — recent error reports and user bug signals are the highest-value ACTIONABLE source. ### Tier 4: Pulse (analytics, deploys, marketing) Search for analytics/deploy channels as needed: ``` slack_search_channels(query: "analytics") slack_search_channels(query: "deployments") ``` Read the most active ones for anomalies. --- ## Phase 2: Message Scoring For EVERY message from Phase 1, classify into one of 5 categories: ### ACTIONABLE (score: 5) — Needs work NOW - Cron failures, deploy failures, error alerts - User bug reports with reproduction steps - Direct requests or steering messages from a human - Unresolved questions from other agents - **Action:** Fix it, file it, or assign it ### ENGAGEABLE (score: 4) — Can add genuine value in thread - Agent discoveries that connect to something you know - Hypotheses you can challenge or validate - Strategic posts where your session findings add context - Questions from agents or a human that you can answer - **Action:** Reply in thread with SUBSTANCE, not "nice find" ### CONNECTABLE (score: 3) — Links two things nobody connected - A triage alert that relates to a strategy post in the ideas channel - A deploy event that explains an analytics anomaly - A piece of user feedback that validates or challenges a research finding - **Action:** Post the CONNECTION to the coordination channel or reply in both threads ### CAPTURABLE (score: 2) — Worth remembering but no immediate action - New patterns you haven't seen before - Domain-specific details (file paths, API patterns, user names) - Timeline events (when things happened, in what order) - **Action:** capture the learning (see Phase 3) ### NOISE (score: 1) — Skip - Repeated churn alerts you've already analyzed - Automated messages with no new information - Messages you've already engaged with this session - **Action:** Nothing. Move on. --- ## Phase 3: Act (the work) Process messages in score order (5 -> 4 -> 3 -> 2). For each: ### For ACTIONABLE messages: 1. Announce in the coordination channel: `[SlotName] Picking up from #: ` 2. Do the work (fix, investigate, file a ticket) 3. Post result back in the ORIGINAL thread (not a new message) {{#IF_MEMORY}} 4. Capture to memory if non-obvious {{/IF_MEMORY}} ### For ENGAGEABLE messages: 1. Read the full thread (use `slack_read_thread`) {{#IF_MEMORY}} 2. Check memory for related context: `memory_search(query: "")` {{/IF_MEMORY}} 3. Reply in thread with one of: - **ADD:** "This connects to X because Y" (your finding enriches theirs) - **CHALLENGE:** "I disagree because Z" (constructive pushback with evidence) - **BUILD:** "What if we also did W?" (extend the idea) - **ANSWER:** "The answer is X, found in file Y" (resolve their question) 4. If the thread discussion reveals new work: cascade into it ### For CONNECTABLE messages: 1. Post the connection in the coordination channel or reply in both threads 2. Format: "[SlotName] CONNECTION: + = " 3. If the connection reveals a gap or opportunity: create a quest lead {{#IF_MEMORY}} ### For CAPTURABLE messages: 1. `memory_store` or batch via `memory_pulse` 2. Include channel context: "From #: " {{/IF_MEMORY}} {{#IF_ISSUES}} ### Filing tickets from high-signal items When an ACTIONABLE message describes work that won't get done this session, file it as a ticket so it isn't dropped. Include the channel, the reporter, and the reproduction details in the description. {{/IF_ISSUES}} --- ## Phase 4: Reflect (diary update) {{#IF_OBSIDIAN}} After processing all channels, update your diary: ```bash VAULT_PATH="${VAULT_PATH:-$HOME/Documents/Obsidian Vault}" SLOT_NAME=$(basename $(pwd)) # append to: ${VAULT_PATH}/Work/Agents/${SLOT_NAME}/diary.md ``` ```markdown ### /rounds: