Dispatch Execution Checklist

Every dispatch runs a mandatory protocol. Step 0 applies only to code-producing dispatches; non-code dispatches begin at step 1.

0. Gate 0 — Fuda scoping (code-producing only) — before any code is written, the change is scoped via a Fuda: read the dependency map, draft the Fuda, post it for review, and capture the tracking id. See Clean Code Pipeline.
1. Decompose + tag — identify the agent(s) from the roster's semantic triggers and resolve the mission namespace.
2. Report dispatch — register with the gateway (agent name, task description, namespace, a 3–5 item milestone array, optional Fuda id / retry-of / plan slug). Capture the returned dispatch_id — it threads into every subsequent cache write.
3. Seed the psychic cache — write all mission-relevant context the agent needs (constraints, decisions, prior findings, instructions), each entry carrying the dispatch_id.
4. Dispatch — launch the agent with mandatory clauses: the dispatch id and namespace, an instruction to assemble context, a progress-reporting schedule, and an activity-logging instruction.
5. Confirm — acknowledge the dispatch to the operator: agent, task summary, milestone count.
6. Merge + re-seed — on return, review and contextualise the output, write key findings back to the cache as intel, record any decisions or state changes, and present the synthesised result.

Reporting

Agents report progress at each milestone transition and write an output cache entry — a deep link to the vault-filed deliverable — before signalling completion. LISA always reviews and contextualises an agent's output before presenting it; nothing is passed through raw.