Repository navigation
feat(settings): move Subagent Settings and Warm Pool to Settings > Agent Harness - #18048
Conversation
First Principles Review (Opus 5.5) — 🟡 CONCERNSPremise-level review of First-Principles-Verdict: CONCERNS The move rests on #8940, but the relabel that rides along has no named user who misread "Warm Pool", and the card now says the same thing twice. Not justified as shipped
What this change shipsInventory (9 items) — 7 justifiedIntent: let ordinary users tune subagent limits and pre-started sessions without turning on Developer Mode. This is an ADDITION (a move).
[FIRST-PRINCIPLES-REVIEWED] dd2fbd9 Description read[DESCRIPTION-READ] ef618adbcc92d112559f9ebbee520fe0703a937f3c581b4febcf8769063a8096 That sha256 names the PR description this verdict read. Recompute it with |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Check that the shipped docs and guides that still say "drain the warm pool" make sense now that the dashboard calls the card "Pre-started Sessions". Suggestions
[DESIGN-REVIEWED] dd2fbd9 Description read[DESCRIPTION-READ] 2aaa3a8cc3b7269a34b3f4bef66f4c2ca38ca3bfd59946f2f9ab7fe35cba8168 That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
UX Review (Opus 5.5) — 🟡 CONCERNSUX-level review of UX-Verdict: CONCERNS "agent" means two things on one page: the harness picker ("New sessions use this agent") and the new Pre-started Sessions copy ("Keeps agents started"). Watch
Suggestions
[UX-REVIEWED] dd2fbd9 Description read[DESCRIPTION-READ] 2f437fc390aade1cdcd90da7e0fec95c4b71429d0989e392cd5073628ee16895 That sha256 names everything this verdict read: the PR description, plus the 7 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
838c66c to
286425a
Compare
|
|
286425a to
f224e85
Compare
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
89755f7 to
dd2fbd9
Compare
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
This PR moves the Subagent Settings and Warm Pool cards from Developer > Config to Settings > Agent Harness. It also renames user-visible labels ("Pool Size/Pool TTL/Pool Agent" become "Kept Ready/Retire After/Uses Crewmate", and the card becomes "Pre-started Sessions") and rewrites copy in all 14 locales. I read the description and the lane history. The draft-resync fix for the GPT finding (cc7e3de6ed48: resync only when the card's own stored keys change) is a real fix and is pinned by a test. Every Design, UX and First Principles item on head dd2fbd9 has a posted disposition. I am holding because this is a product/UX decision: it changes where two run-time knobs live, takes them out from behind Developer Mode, and renames them. Several rounds of UX BLOCKs drove that copy, and the PR does not show a maintainer signing off on the placement or the new names. Auto-merge is already armed, so a maintainer should confirm the placement and the labels before it lands. Checks are green and base is main. — crew-pr-reviewer |
cruisercohen
left a comment
There was a problem hiding this comment.
Approved after review: relocates Subagent Settings and Warm Pool to Settings > Agent Harness with no change to config keys or endpoints, so persistence is unchanged. The two shared-cache regression gaps (unsaved draft surviving an unrelated config write; leave guard against the picker hand-off) are each closed with a dedicated test.
Problem / Motivation
Goal: Subagent Settings and Warm Pool are edited from Settings > Agent Harness, next to the coding-agent switch, instead of from the Developer page.
The two cards that change how agents run (how many subagents, how many turns each, how many pre-started processes) only show on Developer > Config. That page is for internals and sits behind Developer Mode, so an ordinary user has no visible place to tune them. The Agent Harness tab already holds the other "how agents run" knob, the backend picker, since #13888.
Why it matters
A user who wants faster session starts (Warm Pool) or more parallel subagents has to find a hidden internals page. Putting both cards on the Settings tab that already governs agents makes them findable without Developer Mode.
Not a goal
/developeror moving any of its other tabs: the internals views stay where they are.What changed (motivation → approach → change)
The cards keep their controls and config keys. Settings is read by people who never open the internals page, so the cards speak plainly there: "Kept Ready", "Retire After" and "Uses Crewmate" replace "Pool Size", "Pool TTL" and "Pool Agent"; the empty Uses Crewmate option reads "Default crewmate ()"; each card shows one line saying what it does and how it saves. The Max Turns hint says a turn is one tool call (default 1000).
KiroCrewCfgTab.tsxexportsAgentRunSettings, which renders both cards on the same['kirocrewConfig']cache, so an edit on one page shows on the other; a refused warm-pool save shows throughErrorNotice, and Subagent Settings keeps an unsaved draft safe on the shared tab: it resyncs only when its own stored values change, so a harness switch or warm-pool edit leaves the draft alone, and it stakes a leave guard so a hand-off from the picker's error notices asks before dropping it.SettingsPage.tsxlazy-loads it under the backend picker on theagenttab, in its own Suspense boundary. Developer > Config no longer renders the cards, and its subtitle drops the "Subagent Settings has its own Save" clause. Every locale carries the new copy;en-XA.jsonis regenerated. Theagentrow ofdocs/feature-map/README.mdnames the two cards.Backwards compatibility
Compatible: the same controls write the same config keys through the same endpoints; only the page that hosts them changes.
Reader:
website/src/pages/SettingsPage.tsx:AgentRunSettings-- entry: crew page -- the cards now render on /settings/agent (SettingsPage.agentHarness.test.tsx "hosts the Subagent Settings and Warm Pool cards beside the switch")Reader:
KiroCrewCfgTab.tsx:KiroCrewCfgTab(in website/src/pages/overview) -- entry: crew page -- Developer > Config keeps every other card (CfgTabCoverage.test.tsx "leaves Subagent Settings and Warm Pool to Settings > Agent Harness")Tests
SettingsPage.agentHarness.test.tsx: the Agent Harness tab renders the run cards beside the switch, waiting for each lazy panel on its own. Red on origin/main, green here.CfgTabCoverage.test.tsx: the Developer config tab no longer shows either card; every existing Subagent Settings and Warm Pool test rendersAgentRunSettingsunder the new labels; cross-card cases render both on one cache; a failed warm-pool patch shows as an alert in its own card; each card's visible line renders; the empty Uses Crewmate option names the default crewmate.CfgTabCoverage.test.tsx"stakes a leave guard while a subagent draft is unsaved": the guard stakes nothing when clean and asks before leaving a dirty draft. Red without the guard.CfgTabCoverage.test.tsx"keeps an unsaved subagent draft when an unrelated config write lands": red when the card resyncs on any config object.SettingsPage.agentHarnessSignIn.test.tsx: stubs the run cards, which read a config shape that suite does not seed.--sequence.shuffleand random seeds, all passing.Manual verification
Rendered the real SettingsPage and DeveloperPage through the capture harness (
capture/settings-agent-harness.html) with fixture API answers, before and after, and checked both pages by eye. Gates run locally: tsc, eslint, i18n:check, i18n:render, phantom-classes, restyle-ratchet, jscpd, scoped frontend tests, vite build with bundle-size and chunk-cycle checks.Screenshots / video
Settings > Agent Harness, after:
Developer > Config, after (no Subagent Settings or Warm Pool):
Before (origin/main)
Related Issues
Closes #8940
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)