Repository navigation
feat(slack): code channels run the default agent; add "Always use conductor mode" - #38240
Conversation
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS No design-level blocker: the zero-config default changes only when a stored value is read, and the new switch is opt-in, so both can be undone. Notes
[DESIGN-REVIEWED] 5596950 Description read[DESCRIPTION-READ] 8ee83a78538954f8d11043edc6bc575067345900c4f2956bb95871e5f887b066 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 |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Nothing blocks: the base RFC ( What this change shipsInventory (9 items) — 9 justifiedIntent: let a Slack code channel with no settings touched do real work on the owner's default agent, and offer one switch that runs every new channel in conductor mode instead. This is an ADDITION, plus a changed default.
[FIRST-PRINCIPLES-REVIEWED] 5596950 Description read[DESCRIPTION-READ] fc48a9c4449663c871e7e58146035a471b28e5e0a8b866f1b9795fceca7460e3 That sha256 names the PR description this verdict read. Recompute it with |
UX Review (Opus 5.5) — ✅ PASSUX-level review of UX-Verdict: PASS Nothing blocks. The seven screenshots show every new state, and the blind reader read the switch correctly. The remaining confusion is about wording. Notes
[UX-REVIEWED] 5596950 Description read[DESCRIPTION-READ] 253ce59163a30febbdb8b7a55e73354d9b461945f81577e7e16ad65efda22136 That sha256 names everything this verdict read: the PR description, plus the 10 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 |
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: |
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: |
Code change approver recommendation: A person should decideWhy: the decider estimates a 12% chance that this change introduces a defect, within the 12% limit for approving on its own, but it changes 1659 lines, more than the 400 it may approve on its own. I recommend approving. The PR builds what the RFC on main already decided (rfc-slack-code-channels.md §3: "Always use conductor mode", the resolution order, the auto folder and the legacy Resolver ( Binding:
Settings save:
Tier: 2, held for a person: 40 files, more than 20. An automated recommendation, not a review: a maintainer still decides. It approves on its own only when the estimated chance of a defect is at most 12% and the change has at most 400 changed lines. Decider v3c (217fe607) · commit |
KiroCrewCaptain
left a comment
There was a problem hiding this comment.
Thanks for this. The code review found no defects on this head. One item blocks merge: evidence for two new UI states is missing.
website/src/pages/settings/SlackCodeChannelDefaults.tsx:315and:320. This PR adds two lock states to "Always use conductor mode". Pinned: config.local.json sets the switch, so it is disabled and shows the "set in {{path}}" note. Unreadable: config.local.json cannot be read, so the switch is forced off and an inline error appears. The unreadable state also adds new copy. The UX review on this head says it cannot evaluate these two states. The PR body has four screenshots: off, on, on with helpers off, and the LTS case. Neither lock state is among them, so nobody has seen either one rendered. Smallest fix: add two screenshots of this head's card to the PR body, one pinned and one unreadable, captured the same way as the other four. Then re-run the UX review. No code change is needed.
Optional:
src/kiro_crew/dashboard/slack_code_settings.py:579. The overlay-merged agent check also runs on a plain folder-only save. With an overlayagentset, a save that main refuses with 400 now returns 200, and the PR body does not mention this. You could gate that block on a conductor_mode change, or describe the behaviour and add a test for a folder-only save.- This PR and #38112 both change
slack_code_settings.pyand conflict there. Naming the intended merge order in one of the two bodies would help. src/kiro_crew/slack/code_channels/binding.py:112._live_agent_configwraps thelive.snapshot() or cfgread that other modules write inline. You could inline it, or say why the try/except is needed here.
114318e to
0960c15
Compare
|
Thanks. All four items are addressed in 0960c15:
|
|
This changes behaviour users already rely on, so it needs a design sign-off from a maintainer before it can merge. A maintainer will post the decision here. To propose a design, follow the request-for-change process in CONTRIBUTING.md (docs/request-for-change/). The RFC amendment merged in #38124 records the decision this PR builds. A maintainer will confirm here that it covers this change. The review found no blocking defect on head 0960c15. What it checked:
The five changed backend test files pass at this head. One thing done well: old installs keep working without rewriting any data, and a test pins each case. Earlier change request, item by item:
Optional:
|
Withdrawn: every item in this review is fixed or answered on the current head (see the follow-up comment). Other holds on this PR are listed there.
…ductor mode"
A zero-config code channel filed into the auto "Code channels" folder, which was
created with kirocrew-conductor as its default agent, so the channel ran the
conductor and refused work without helper agents. The auto folder is now created
with no default agent, so the owner's default agent runs. An auto folder still
holding the legacy kirocrew-conductor reads as never edited (read-time only; the
folder is not rewritten). Any other folder agent is kept.
New slack.code_channel_defaults.conductor_mode ("Always use conductor mode"), off
by default, shown and honoured wherever conductor mode is offered (every build
but LTS). When on, every new code channel runs agent.crew_mode_agent in conductor
mode, over the code-channel default agent and the folder's agent; its helpers
stay filed in the channel's folder. The default-agent picker is
disabled with a note while it is on and keeps its stored value, and the card
warns (display only) while helper agents are off. A missing or malformed value
reads as off. Helper dispatch stays independent of the agent.
0960c15 to
5596950
Compare
|
@iamwhatever this needed a rebase after #38112 merged, since both touch the same settings card, and the push dismissed your approval on 0960c15. The head is now 5596950. Besides the rebase it adds the three non-blocking UX Review notes and KiroCrewCaptain's optional notes 2 and 3. CI is green (88 pass, 0 fail). Could you re-approve? |
|
The design is signed off. A maintainer approved the RFC amendment in #38124 and then approved this PR on 0960c15. The rebase push dismissed that approval, so the PR now waits for a maintainer approval on this head. The review found no blocking defect on head 5596950. What it checked:
One thing done well: old installs keep working without any stored folder being rewritten, and tests pin both cases: the legacy auto folder, and another folder that names the conductor. Earlier change request, item by item:
Optional:
|
Problem / Motivation
Goal: A Slack code channel with no settings touched runs the owner's default agent, and the owner can choose to run every new code channel in conductor mode instead.
A code channel with nothing configured is filed into the auto "Code channels" sidebar folder, which was created with
kirocrew-conductoras its default agent. The channel therefore ran the conductor, which only plans and hands off work, and with helper agents off it refused with "This code channel isn't allowed to open helper chats". The owner asked for an option to always use conductor mode, shown where that feature is offered, and otherwise for the default agent rather than the conductor.Why it matters
Every install that turns Code channels on gets a channel that cannot do the work it was opened for until the owner finds and turns on helper agents. Owners who want a conductor in every channel have no single switch for it.
Not a goal
slack.code_channel_dispatch), its gate or its refusal text. It stays independent of the agent, and the new switch never turns it on./kirocrew agent.What changed (motivation → approach → change)
Governed by
docs/request-for-change/rfc-slack-code-channels.mdon main, which already records this decision (#38124); this PR builds it and does not edit the RFC. The decision is in §3 Code channels ("Always use conductor mode", "Resolution order", "The auto folder", and "Work board" for filing helper agents) and in Backward compatibility. It follows two maintainer decisions already on main:2026-11-14-conductor-mode-graduates-from-feature-previews.md(conductor mode is on in every build but LTS) and2026-11-15-conductor-mode-runs-on-the-conductor-by-default.md(agent.crew_mode_agentdefaults tokirocrew-conductor). Its wording follows2026-11-13-conductor-mode-keeps-no-ultra-names.md. The RFC index row is re-audited in a docs-only PR once this amendment's PRs are on main.Default agent (the switch off, the default). The auto "Code channels" folder is now created with no
default_agent, so resolution falls through toagent.default_agent(elsekirocrew). Existing installs already have that folder withkirocrew-conductor. A folder record keeps no edit history, so the value is the only signal: a top-level folder named "Code channels" that still holds exactlykirocrew-conductor(features.LEGACY_AUTO_FOLDER_AGENT) is read as never edited and counts as no folder agent, however it is reached (by path or by id). That is read-time only; nothing on disk is rewritten. Any other value, or the conductor on any other folder, is the owner's and is kept. Rejected: clearing the value on disk, because a later owner-set conductor would then look the same, and it rewrites user data with no undo.Always use conductor mode. New
slack.code_channel_defaults.conductor_mode, off by default and stored only when on. A two-state rule:features.conductor_mode_offered(), the edition lock the dashboard'screwModeEnabled()reads: every build but LTS): every new code channel runsagent.crew_mode_agentas a template, over the code-channel default agent and any folder agent (agent_source"conductor").code_channel_defaults.agent> the folder'sdefault_agent(legacy rule above) >agent.default_agent>kirocrew, as before.A missing or malformed value, or the LTS edition, reads as off. The default-agent fallback (
agent.default_agent, now the zero-config path) binds as a template, resolved the way a new dashboard chat resolves it (default_template, so a removed spec falls back to the managed default, thendispatch_kiro_agentfor the declared name), and the settings preview names the same agent. A Bot keyed by the same name never takes a code channel. Theagent.*values a new channel reads (agent.default_agent,agent.crew_mode_agent) come from the live config snapshot, not the gateway's config object, which a reload refreshes only for Slack fields. The session is a conductor mode root by its agent. It does not get thecrew_mode_agentstamp a dashboard New chat born on the conductor gets: that stamp moves a root's workers out of folders (crew_mode.slot_in_turned_on_tree), and a code channel files its helpers in the channel's own folder, as conductor code channels do today. Bound sessions keep their agent, and a pick in the kickoff picker or/kirocrew agentstill applies, within the existing lock.Settings. Settings → Channels → Slack → Code channels → defaults card: an Always use conductor mode switch above Default agent, rendered only where conductor mode is offered (
useCrewMode(); settings search hides it the same way). Its text says it overrides the default agent and the folder's agent; its info tip names Let code channels start helper agents. While it is on, Default agent is disabled with a one-line note and keeps its stored value; the preview row shows the conductor mode agent with the source "Conductor mode"; and while helper agents are off in the draft, an inline warning under the switch names that setting (display only: it saves nothing). The auto-folder row no longer names the conductor. ThePUTaccepts only a boolean or null forconductor_mode. While conductor mode decides the agent, the stored default agent is dormant: a folder change does not re-judge it against the new folder's project, and the card hides its stale-agent warnings while the draft has the switch on. The GET still reportsagent_missing, so turning the switch off in the draft shows the warning at once, and a save that turns the switch off (alone or with a folder change) re-judges the stored agent, as an explicit agent edit is. That save judges what the runtime will run after it,config.jsonwithconfig.local.jsonmerged over it, so an agentconfig.local.jsonsets never blocks turning the switch off. A plain folder change with the switch off judges the stored agent against the new folder exactly as main does, overlay or not (test_a_folder_only_save_judges_the_stored_agent_as_main_does). Whenconfig.local.jsonsetsslack.code_channel_defaults.conductor_mode(or cannot be read), the switch is disabled with a note naming that file, and thePUTrefuses a change with 409, as for the helper-agents switch, so a save never reports a change the overlay overrides. An overlay that setscode_channel_defaultsto a non-object replaces the whole section in the merge, which the reader takes as all-unset, so it reads as pinning the switch off (disabled, 409 on a change), not as unreadable. Whileconfig.local.jsoncannot be read at all, whether it turns the switch off is unknown, so every reader counts the switch as off, read from the file at the moment of the decision (conductor_overlay_unreadable): the bind, the settings preview, the save and the card. This fails closed, as the helper-agents switch does, whether the loader dropped the file or the live watcher kept an earlier copy. A danglingconfig.local.jsonsymlink is no file, as the loader reads it, so it locks nothing (test_a_dangling_overlay_symlink_reads_as_no_overlay).Files:
src/kiro_crew/slack/features.py:LEGACY_AUTO_FOLDER_AGENT,conductor_modeonCodeChannelDefaults,conductor_mode_offered,code_channel_conductor_agent,conductor_agentinresolve_code_channel_session.src/kiro_crew/slack/code_channels/binding.py: auto folder created with no agent; the conductor agent passed to the resolver;agent.default_agentbound as a template.src/kiro_crew/dashboard/slack_code_settings.py: GET reportsconductor_mode,code_channel_conductor_agent,legacy_agent; PUT validatesconductor_mode.src/kiro_crew/config/sections_slack.py,section_builders.py: the config field and its reader.website/src/pages/settings/SlackCodeChannelDefaults.tsx,slackCodeDraft.ts,api/client/messaging.ts,commandPalette/settingsSearchCore.ts: the switch, the disabled picker, the warning and the preview.config-baseline.json.docs/guides/slack-setup.mdandsrc/kiro_crew/docs/slack-integration.md(including that a dashboard New chat in a legacy auto folder still getskirocrew-conductor), one row indocs/system-specs/modules/crew-mode.md, and the code-channel sentences indocs/system-specs/modules/slack-gateway.md(binding resolution order, work board) andsession-control.md(the second exemption) that assumed the conductor was the default. The RFC is unchanged here (main has the decision from docs(rfc): record the Slack Code channel decisions the next pull requests build on #38124).No Sensitive Change Review watched path is touched.
Backwards compatibility
Merge order: #38112 (helper agents turn on with a first enable of Code channels) has merged, and this PR is rebased onto it. Both change
slack_code_settings.py, the defaults card's draft and payload, and the locale catalogs; the rebase conflicted only in one comment block, where both sides add lines.Compatible:
conductor_modeis a new optional key that reads as off when absent, and every existing default keeps its meaning. The PUT accepts one more key and refuses nothing it accepted before.The diff takes something away: a zero-config code channel (and one filed into an auto folder that still holds the legacy
kirocrew-conductor) no longer runs the conductor, and the GET'scode_channel_auto_folder.agentfield is replaced bylegacy_agent.Reader: src/kiro_crew/slack/features.py:resolve_code_channel_session -- chat -- the one resolver;
test_the_legacy_conductor_on_the_auto_folder_reads_as_unset,test_another_folder_naming_the_conductor_keeps_itReader: src/kiro_crew/slack/code_channels/binding.py:resolve_session_spec -- chat --
test_a_fresh_home_resolves_to_the_default_agent_in_code_channels,test_an_existing_legacy_auto_folder_runs_the_default_agentReader: src/kiro_crew/dashboard/slack_code_settings.py:code_settings_view -- crew page --
test_unset_folder_reports_the_auto_folder_and_no_folder_agentReader: website/src/pages/settings/SlackCodeChannelDefaults.tsx:autoLabel -- crew page -- the only frontend reader of
code_channel_auto_folder.agent, moved tolegacy_agent;reads the legacy conductor on an existing Code channels folder as no folder agentReader: src/kiro_crew/slack/code_channels/agent_picker.py:kickoff_blocks -- chat -- preselects the agent the session runs;
test_the_kickoff_carries_a_picker_preselected_with_the_default_agentA dashboard New chat inside the auto folder still reads the folder's stored
default_agentunchanged; only code-channel resolution applies the legacy rule.Tests
test/test_slack_features.py: resolution order (conductor mode over default and folder agent, legacy conductor on the auto folder by path and by id, conductor on another folder kept, nothing configured falls to the default agent);conductor_modereads on only for a realtrue;code_channel_conductor_agentrunskirocrew-conductorby default, keeps a configuredagent.crew_mode_agent, and is off on LTS, off when unreadable.test/test_slack_code_settings.py: GET defaults includeconductor_mode; PUT savestrue, clears onfalse, refuses non-booleans; while it is on, a folder change does not re-judge the dormant default agent, while an explicit agent edit, or turning it off (alone or with a folder change), still is judged;agent_missingis still reported while it is on; a non-objectcode_channel_defaultsoverlay pins the switch off; an overlay that sets the switch, or an unreadable overlay, refuses a change with 409 and nothing is staged, while an overlay that sets only helper agents does not; effective agent isconductorwhen on and the usual order on LTS; on LTS a stored on never makes the default agent dormant (a folder-only change to a project without it is refused, 400); a danglingconfig.local.jsonsymlink locks nothing; the legacy auto-folder agent is not reported.test/test_slack_code_channels_binding.py: a fresh home creates the folder with no agent and binds the default agent as a template (so a Bot keyed by the same name never takes it); a legacy folder runs the default agent and is not rewritten; conductor mode wins over default and folder agents; inert on LTS; a conductor-born session carries nocrew_mode_agentstamp, so its helpers keep the channel's folder;agent.crew_mode_agentandagent.default_agentare read from the live config snapshot; a removedagent.default_agentfalls back to the managed default.test/test_slack_code_channels_routing.py,test/test_slack_code_channel_agent_picker.py: zero-config channels and the kickoff picker land on the default agent.website/src/test/SettingsSlackCodePanel.test.tsx: hidden on LTS, and a stored on is inert there; on, it disables the default-agent picker with the note, keeps the stored pick, previews the conductor mode agent and savesconductor_mode: truewithout touching dispatch; the helpers-off warning shows only while helpers are off and saves nothing; the stale-agent warnings are hidden while it decides the agent; it is disabled with the pinned or unreadableconfig.local.jsonnote; a saved on reads back and turning it off sendsnull; the legacy auto-folder agent is not shown.Per-file coverage of changed
srcfiles under the tests that exercise them:slack/features.py96%,code_channels/binding.py81%,dashboard/slack_code_settings.py94%,code_channels/agent_picker.py88%,config/section_builders.py85%,config/sections_slack.py100%.Manual verification
N/A — unit coverage of the resolver, the binding, the settings API and the settings card covers every branch.
Screenshots / video
Settings → Channels → Slack → Code channels, the code-channel defaults card (built SPA, stubbed API).
Switch off (default): the default agent and folder resolve as before.
Switch on: Default agent disabled with its note, the stored pick kept; the preview names
kirocrew-conductor, source "Conductor mode".Switch on with helper agents off: the display-only warning under the switch.
Conductor mode not offered: LTS builds don't offer conductor mode anywhere, so the switch is not rendered and Default agent stays usable.
Set in
config.local.json: the switch shows the value in force, disabled, with a note naming the file.config.local.jsoncan't be read: the switch reads off, disabled, with the error naming the file. The error opens with "Off:", so the position does not depend on colour.No Code channels folder yet: the default-folder row and the preview name the one created on first use.
Review notes on 114318e and how each was handled
kirocrew-conductor. Said so inslack-setup.mdandslack-integration.md._live_agent_configduplicated the inlinelive.snapshot() or cfgidiom. Deleted;binding.pynow uses the idiom (live.snapshot()only reads a module attribute, so there was no failure for the try/except to guard)._live_agent_configis inlined (above).config.local.json) and 6 (config.local.jsonunreadable).UX Review notes on 0960c15 and how each was handled
SettingsSlackCodePanel.test.tsx).KiroCrewCaptain's optional notes on 0960c15 and how each was handled
rfc-slack-code-channels.md:52: not edited here; the RFC index row is re-audited in a docs-only PR once this amendment's PRs are on main.test_on_lts_a_stored_on_never_makes_the_default_agent_dormant(a stored on, a folder-only change to a project without the agent: 400, as on main). It passed before any change.dangling_unreadablebranch is deleted, so all three readers take a danglingconfig.local.jsonas no file, as the loader does;test_a_dangling_overlay_symlink_reads_as_no_overlayfailed before and passes now. A file that exists but cannot be read still locks the switch off (shot 6).Related Issues
N/A
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement
Contributions are accepted under the project's Apache-2.0 license.