teamlisted

# /team — two-agent task execution Read `~/clawd/agent-comms/team/PROTOCOL.md` FIRST and follow it exactly, then `~/clawd/agent-comms/team/LESSONS.md`. This skill is the Claude Code entry point; the protocol file is the source of truth for classifier, transport, ledger, tiers, and failure rules. Paths here are home-anchored (`~/…`) on purpose. `agent-comms/team/...` is relative and does not resolve from the usual cwd, and `~/agent-comms` is a DIFFERENT repo with no `team/` dir. A relative path silently skips the protocol and the lessons file. ## Procedure 0. **Recon before you design.** Before proposing any adapter, wrapper, bot, or browser automation, enumerate what the target already exposes — list its API routes, CLI commands, and exported functions FROM SOURCE. Never accept "there is no X" from a README, a report, or the other agent; a negative capability claim rules out whole designs, so verify it directly. Two agents agreeing on an inherited premise is one check, not two. (See LESSONS.md 2026-08-10.) 1. **Classify** the task with the PROTOCOL.md Classifier. Announce the lead in one sentence. 2. **Create the task** (you are the receiving agent): `node ~/clawd/agent-comms/team/bin/team-ledger.mjs create --slug <slug> --instruction "<task>" --origin claude-code --lead <claude|openclaw>` then `update --task <id> --status in_progress`. 3. **If YOU lead:** - `dashclaw_session_start` for the task. - Decompose. For each piece that belongs t