bug(opencode): orchestration-created OpenCode agents fail to start (opencode serve spawn timeout) under load #1355

Observed (2026-09-26, manager fleet)

`orchestration_create_agent` with kind `opencode` failed three separate times in about an hour, from three different workers, each time with `opencode serve` failing or timing out on spawn:

• W4, the Workflow runs and workflow codex-home sessions are never pruned (1.3 GB) #1275/wfm#65 reviewer slot: "opencode serve timed out on spawn"
• W3, the perf(storage): hold the boot prune until the workspace has recovered (#775) #1351 reviewer c: "opencode serve failed to start"
• W4, the test: remove machine- and time-dependence from five tests and personal data from a fixture #1353 reviewer c: "opencode serve timed out on spawn", its second time today

Each fell back to Pi. The machine's load average was 40–60 at the time: about 7 parallel `tsc -b` runs, many Codex reviewers, and roughly 60 agents open.

Why it matters

The reviewer rotation (codex/codex/opencode, which keeps Claude and Codex quotas in reserve) depends on OpenCode starting reliably. A spawn deadline that assumes an idle machine turns load into a hard failure.

Expected

• Either the serve readiness deadline scales, or it keeps waiting while the process is alive and making progress, instead of a fixed timeout (never widen blindly; find what it waits on).
• The failure names its cause (timeout vs exit code vs port) through a curated message.
• Check bug(testing): opencode-terminal live suite times out waiting for the permission condition on every version #1009, whose opencode-terminal live suite also times out waiting for the provider, for a shared root cause.

Evidence

Worker status files in `temp/manager/status/w{3,4}.md`, 2026-09-27 03:10–03:37Z.