arjitj2/open-pstack/swarm

Swarm

Dispatch contract. Resolve each worker descriptor through `provider-dispatch.md`. The parent starts native and external lanes; workers never route themselves. On Codex, resolve remaining Claude tool names via `codex-tools.md`.

Start

Open a todolist with one entry per phase before launching anything.

1. Frame
2. Fan out
3. Aggregate
4. Report

Phase A: Frame

1. State the done predicate and the artifact or report the swarm must return.
2. Choose the shape. Partition into slices, race N workers on identical briefs, or mix both. For a race or mixed shape, declare first pass, rank all, or best-of before spawning.
3. Pick the worker descriptor from the swarm workers line in the current harness's pstack model sheet. If the sheet or that line is missing, use grok:grok-4.7@xhigh. For auto or inherit-parent, the workers run on the parent model through its native lane. Each worker seat follows provider-dispatch's saved-chain policy before every attempt. A terminal outcome the sheet's # fallback policy authorizes makes a read-only worker eligible for its next saved attempt (usage-exhausted only when no policy is saved); a writer advances only when the helper proves the workload never started or a recorded parent inspection clears it, and otherwise stops for inspection. Every other route failure is a dropout, not a substitution. For a model race, name each arm's descriptor up front.
4. Give each worker its own writable output when it writes. When workers verify or measure commits, each brief names the exact SHAs. A measurement brief also names the method (sample count, what one sample is, order). The worker records both in its result.

Phase B: Fan out

Start all N workers in one fan-out phase through provider dispatch. Native lanes use background subagents; external lanes invoke the launcher as background work with retained task/session handles. Never use Claude's foreground Bash path for a long worker. Every writer runs in its assigned worktree or output directory. Isolation comes from those paths, not the provider.

When a worker must start from a non-default branch, check that branch out in the worker's own worktree and name the worktree path in its brief.

Every brief stands alone. Include the goal, scope, exact slice or race arm, how to verify, and what to report. Reports use PASS, ISSUES, or BLOCKED with evidence. A worker that can prove a defect reports ISSUES and lists every issue it can prove, not only the first.

If the helper stops a worker or its chain is exhausted, proceed with N-1 and note the provider, model, and receipt failure. If it returns inspect, preserve and inspect that writer before aggregation. Report the actual provider distribution across completed workers; a fallback keeps the seat but may reduce diversity — say so. Never substitute another provider silently.

Phase C: Aggregate

Read the terminal results. Drop a result that does not record the SHAs and method its brief names, and respawn that worker once. After a second miss, record a gap. A gap does not count as a pass. For coverage, every required slice needs a result. For a race, apply the selection rule declared up front. Use first pass, rank all, or best-of. Do not paste raw worker dumps.

Phase D: Report

Return one consolidated in-chat report with the table, issue one-liners, gaps or dropouts, and the race rule when used.