feat(review): drop the default reviewer timeout — reviewers run until done

[peyton-alt]

https://entire.io/gh/entireio/cli/trails/784

Problem

Reviewers were killed at a wall-clock default (10m, later raised to 20m) that conflates "slow because working" with "hung". Observed live: a reviewer at ~90% completion killed with everything discarded. Plain agent invocation has no such cliff — the user is the liveness check.

Two findings shaped the fix:

• No inactivity watchdog can replace the wall clock. Legitimate review subagents run 12+ minutes while the reviewer's parent process emits zero output (measured on a real run), so a headless child offers no liveness signal that distinguishes a quiet working subagent from a hang. Any watchdog on available signals re-creates the kill-real-work bug.
• The duration is the task, not a hang risk. A controlled A/B on the same branch with the same two skills: plain claude session 9m57s vs entire review 9m08s. Reviews take as long as they take, wherever they run.

Change

• --timeout default: 20m → 0 (none). Reviewers run until they finish, exactly like a directly-invoked skill. The flag remains as an explicit opt-in hard cap for CI/cost-bounded runs.
• The judge keeps its own 5m default regardless — it's a single text-generation call; a new judgeTimeoutArg mapping prevents the reviewer's no-cap sentinel leaking through as "judge unbounded".
• A stuck reviewer in an interactive run is Ctrl+C (process-group kill already handles it, fix(cli): close keyring-interrupt race in Ctrl-C signal-abort #1625-era fix); unattended liveness is the background-mode/durable-run design's job, not a kill timer's.

Testing

• TestReviewerTimeout_NoDefaultCap, TestJudgeTimeoutArg, and the flag-resolution test updated to pin the new contract
• mise run test (7,604) green, lint clean

🤖 Generated with Claude Code

---

Note

Medium Risk
Default behavior change means long or stuck reviewer processes are no longer auto-cancelled unless users pass --timeout; multi-agent reviews may run much longer, though the judge consolidation step remains bounded.

Overview
entire review no longer applies a wall-clock cap to reviewers by default (previously ~20m). Reviewers run until they finish, matching a skill invoked in a normal session; --timeout stays an opt-in hard cap for CI or cost limits, with updated flag help (default 0).

reviewerTimeout drops the implicit default: only a positive ReviewerTimeout enables context.WithTimeout; unset, zero, or the no-cap sentinel means no reviewer deadline.

The consolidating judge is handled separately via judgeTimeoutArg: an explicit positive --timeout still bounds the judge; when reviewers are uncapped, the reviewer sentinel must not disable the judge—the judge keeps its existing ~5m synthesis default.

Tests pin the no-default-cap contract, judge mapping, and flag → resolver behavior.

^{Reviewed by Cursor Bugbot for commit fbe6fe6. Configure here.}

[Copilot]

Pull request overview

This PR updates the entire review command’s timeout semantics so reviewers no longer have a default wall-clock cap. The reviewer timeout flag remains available as an explicit hard limit, while the consolidating judge keeps its own bounded default unless a positive --timeout is provided.

Changes:

• Remove the default per-reviewer timeout so reviewers run until completion unless explicitly capped via --timeout.
• Add a judgeTimeoutArg mapping so the reviewer “no-cap” sentinel does not disable the judge timeout (judge keeps its default unless explicitly bounded).
• Update tests and flag-resolution expectations to pin the new contract.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
cmd/entire/cli/review/run.go	Drops the default reviewer timeout and treats non-positive values as “no cap”.
cmd/entire/cli/review/run_test.go	Updates and adds tests to lock in the new reviewer/judge timeout contract and flag resolution.
cmd/entire/cli/review/cmd.go	Changes --timeout default to 0, introduces judgeTimeoutArg, and wires judge timeout behavior accordingly.