chore(ci): trim darwin test profile

[marius-kilocode]

Problem

On CLI-touching PRs the unit pipeline gates at 11-14 minutes, set by the single unit (macos) job. Linux (2 shards) and Windows (4 shards) usually finish in 5-8 minutes.

The darwin profile drifted far beyond macOS-specific coverage. It ran 100 files containing ~1300s of test time, dominated by platform-neutral application logic that the full Linux and Windows suites already cover. From a recent main junit artifact:

File	macOS	Linux	Coverage
session/prompt.test.ts	213s	98s	Session loop logic, no darwin API
kilocode/daemon.test.ts	93s	29s	Daemon lifecycle over HTTP, no darwin API
cli/acp/lifecycle.test.ts	85s	26s	ACP protocol logic
cli/smokes/read-only.test.ts	83s	24s	Covered by the retained serve-process smoke
cli/run/run-process.test.ts	81s	22s	Run command logic
snapshot/snapshot.test.ts	78s	26s	Git snapshot logic
kilocode/server/config-overlay.test.ts	77s	28s	Server config logic

Every file also runs roughly 3x slower on the 3-vCPU macos-15 runner, making macOS the most expensive place to duplicate platform-neutral coverage.

Change

Trim the darwin profile to 24 files whose failure would indicate a macOS-specific problem:

• Seatbelt sandbox coverage (kilocode/sandbox/*, including tests that no-op on Linux)
• node-pty and HTTP PTY coverage
• FSEvents watching (core-watcher)
• Process and terminal behavior (background-process, interactive-terminal, tool/shell, util/process)
• Filesystem boundary behavior
• One kilo serve boot smoke and the darwin install artifact

Every removed test continues to run in the full Linux and Windows suites. Existing Linux and Windows sharding is unchanged. Non-CLI package tests also remain on macOS because kilo-sandbox contains genuine seatbelt coverage.

The profile's recorded test workload drops from ~1300s to ~218s without committing generated test inventories or timing snapshots.

Full list of the 76 files removed from the darwin profile (all still run on Linux and Windows)

• Session loops and snapshots: session/prompt.test.ts, snapshot/snapshot.test.ts, kilocode/snapshot-{cache,freeze-repro,revert-move,seed}.test.ts, kilocode/session-diff-restore.test.ts
• Daemon and CLI runtime: kilocode/daemon.test.ts, cli/acp/lifecycle.test.ts, cli/run/run-process.test.ts, cli/smokes/read-only.test.ts, cli/tui/{plugin-lifecycle,plugin-loader-entrypoint,thread}.test.ts, kilocode/cli/cmd/serve.test.ts
• HTTP API and server routing: server/{experimental-session-list,httpapi-experimental,httpapi-file,httpapi-listen,httpapi-reference,httpapi-workspace-routing,project-init-git,worktree-endpoint-repro}.test.ts, kilocode/server/{config-overlay,listener-runtime,tui-config,worktree-list}.test.ts, kilocode/mcp-oauth-callback.test.ts
• Config and project: config/{config,tui}.test.ts, control-plane/workspace.test.ts, kilocode/config/config.test.ts, project/{instance,instance-bootstrap,migrate-global,project,project-directory,vcs,worktree,worktree-remove}.test.ts
• Session export: kilocode/session-export/{e2e,sequence,worker,workspace-provider}.test.ts, kilocode/session-export/worker/{storage,zstd}.test.ts
• Worktrees: kilocode/{primary-worktree,worktree-diff,worktree-diff-summary,worktree-family,worktree-family-submodule,worktree-project-skills,worktree-remove-lock}.test.ts
• Tools and plugins: tool/{apply_patch,edit,glob,grep,read,recall,registry,skill,truncation,write}.test.ts, plugin/{install-concurrency,loader-shared}.test.ts, kilocode/tool/repo_clone.test.ts
• Misc neutral: kilocode/{diff-full,indexing-worker,indexing-worktree,project-id,task-nesting}.test.ts, fixture/fixture.test.ts, git/git.test.ts, image/image.test.ts, util/{glob,module}.test.ts

Kept (24 files): kilocode/sandbox/* (11, seatbelt is macOS-only), pty/* + server/httpapi-pty.test.ts, kilocode/{background-process,core-watcher,external-directory-boundary,interactive-terminal,read-directory}.test.ts, kilocode/cli/install-artifact.test.ts, cli/serve/serve-process.test.ts, tool/shell.test.ts, filesystem/filesystem.test.ts, util/{filesystem,process,which}.test.ts.

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Files Reviewed (2 files)

• packages/opencode/script/kilocode/test-profile.ts
• packages/opencode/test/kilocode/test-profile.test.ts

The current diff against main reduces to these two files: the darwin profile is reorganized into filesystem, pty, runtime, and sandbox groups per the PR description, and the test assertions (file-count threshold, contains/not-contains checks) are updated consistently with the new group contents. No behavioral inconsistencies found between the profile definition and its test coverage.

Previous Review Summaries (2 snapshots, latest commit dafb3c2)

Current summary above is authoritative. Previous snapshots are kept for context only.

Previous review (commit dafb3c2)

Status: No Issues Found | Recommendation: Merge

Files Reviewed (9 files)

• packages/opencode/script/kilocode/test-durations.darwin.json (deleted)
• packages/opencode/script/kilocode/test-durations.linux.json
• packages/opencode/script/kilocode/test-durations.ts
• packages/opencode/script/kilocode/test-durations.win32.json
• packages/opencode/script/kilocode/test-profile.ts
• packages/opencode/script/kilocode/test-shard.ts
• packages/opencode/script/test-runner.ts
• packages/opencode/test/kilocode/test-profile.test.ts
• packages/opencode/test/kilocode/test-shard.test.ts

Incremental review: the latest commit only removes test-durations.darwin.json. TestShard.durations() already falls back from test-durations.<platform>.json to test-durations.linux.json (see test-shard.ts), so this deletion is safe and matches the documented fallback behavior.

Previous review (commit 58cf6ff)

Status: No Issues Found | Recommendation: Merge

Files Reviewed (9 files)

• packages/opencode/script/kilocode/test-durations.darwin.json
• packages/opencode/script/kilocode/test-durations.linux.json
• packages/opencode/script/kilocode/test-durations.ts
• packages/opencode/script/kilocode/test-durations.win32.json
• packages/opencode/script/kilocode/test-profile.ts
• packages/opencode/script/kilocode/test-shard.ts
• packages/opencode/script/test-runner.ts
• packages/opencode/test/kilocode/test-profile.test.ts
• packages/opencode/test/kilocode/test-shard.test.ts

---

_{Reviewed by claude-sonnet-5 · Input: 30 · Output: 7.3K · Cached: 624.3K}

_{Review guidance: REVIEW.md from base branch main}

[marius-kilocode]

Final profile-only measurement from run 29815173247 (commit 4e1655da8c):

Job	main #12294 (29731286534)	main #12401 (29734453216)	profile-only PR
unit (macos)	13m57s	7m20s	2m51s
unit (linux, 1/2)	5m16s	5m55s	5m23s
unit (linux, 2/2)	6m37s	5m39s	6m03s
unit (windows, 1/4)	7m58s	7m03s	7m21s
unit (windows, 2/4)	7m02s	6m51s	6m25s
unit (windows, 3/4)	5m49s	6m16s	6m04s
unit (windows, 4/4)	5m18s	5m42s	4m56s
HttpApi exerciser	4m16s	4m18s	6m01s
workflow wall	~14m38s	~7m58s	~8m22s

The macOS profile trim has a clear effect:

• macOS job: 7m20s-13m57s before, 2m51s now.
• macOS CLI test step: 11m13s in the slow baseline, 1m38s now (100 profile files reduced to 24).
• Linux and Windows sharding is unchanged and their timings remain within their previous run-to-run range.

The full workflow is now gated by Windows 1/4 at 7m21s, not macOS. This clean run is 6m16s faster than the slow baseline but 24s slower than the fastest recent baseline because Windows and HttpApi were slower in this sample. One run does not support a claim of reliably sub-8-minute total wall time. The durable result is that the variable 7-14 minute macOS tail is removed; further reliable wall-time reduction must target Windows 1/4, HttpApi variance, or workflow startup/tail overhead.

A pre-existing kilocode/session-prompt-queue.test.ts flake occurred only on the discarded first revision and passed on rerun. The final profile-only run was green without retries.