Repository navigation
fix(website): drop "whole" from the crewmate saved-agent card line - #21863
Conversation
Claude, Codex and DeepSeek now confirm a crewmate's saved spec by projection and report hooks, toolsSettings, excludedTools, native tool limits, per-tool mounts and auto-approval as gaps, so "Crewmates load their whole saved agent" overclaims on those cards. The label is keyed per capability, so it now reads "Crewmates load their saved agent" on every backend, and the card's existing "will not take effect" list names the fields a backend does not apply. Every locale drops its "whole" equivalent; en-XA is regenerated. Closes #19066
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Nothing to check. What this change shipsInventory (4 items) — 4 justifiedIntent: stop the Agent Harness card from telling users crewmates load their "whole" saved agent on backends that only project it. FIX.
[FIRST-PRINCIPLES-REVIEWED] 446d100 Description read[DESCRIPTION-READ] d48101a579b183d8fc2d66ac67bfaa23af4d83e4c1f36bce0e03f641a145d798 That sha256 names the PR description this verdict read. Recompute it with |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing to check. [DESIGN-REVIEWED] 446d100 Description read[DESCRIPTION-READ] 3e68dc0cb21e9310b08acbe3359add69762075a0f718510da845227361b9c0b4 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) — ✅ PASSUX-level review of UX-Verdict: PASS Check that the green tick on "Crewmates load their saved agent" still reads as "everything applies". The "will not take effect" list, which says otherwise, sits two rows lower. [UX-REVIEWED] 446d100 Description read[DESCRIPTION-READ] fc442b4acdf1285310570c9e599a22006fc766a731e77845baa10c0d9da941e4 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 |
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: |
KiroCrewCaptain
left a comment
There was a problem hiding this comment.
Tier 1 auto-approve: fix (16 files). Criteria: no conflict, no requested changes, security path denylist clean, design-doc gate clean, security checklist all-NO, no reviewer-lane findings, readiness signal fresh for this head SHA, declared category matches the diff. Category: the crewmate saved-agent card line drops "whole" in every locale, with the matching backend comment, spec line and test string.
Problem / Motivation
Goal: The Agent Harness card stops telling users that crewmates load their "whole" saved agent on backends that only confirm part of it.
Since #16847, Claude Code, codex and deepseek confirm an enrolled crewmate's saved spec by projection. Those backends report some saved fields as projection gaps instead of applying them (hooks, toolsSettings, excludedTools, native tool limits, per-tool mounts, auto-approval; see
docs/system-specs/modules/harness-parity.md, H6 paragraph). The card's detail line still read "Crewmates load their whole saved agent" on every backend that has the capability, so on those three it promised more than ships.Why it matters
A user picking a harness for crewmates reads "whole saved agent" as "everything in my spec applies here". On Claude Code, codex and deepseek it does not, so hooks and auto-approval rules a user relies on are silently not applied.
Not a goal
AgentBackendTab.tsxdoes for every line.crewCapabilities.newRuntime. feat(members): confirm saved member specs on Claude, Codex and DeepSeek #16847 already made it backend-neutral on main ("Saving leaves active sessions unchanged. A new runtime loads the saved version.").ACP_BACKENDS_MEMBER_CAPABILITIESor to how gaps are computed.What changed (motivation → approach → change)
member_saved_agentdetail line says "whole" on Claude Code, codex and deepseek, which report projection gaps.CAPABILITY_LABELinwebsite/src/pages/developer/AgentBackendTab.tsx), so it cannot say "whole" for kiro-cli alone without a second line id fromagent_sdk/backend_cards.py.en-XA.jsonis regenerated withnpm run i18n:pseudo.LINE_MEMBER_SAVED_AGENTentry insrc/kiro_crew/agent_sdk/backend_cards.pyand theACP_BACKENDS_MEMBER_CAPABILITIESrow indocs/system-specs/modules/harness-onboarding.mdstill said "only kiro-cli" / "full saved agent spec"; both now matchagent_sdk/backends_members.py: kiro-cli and KAS load the saved spec, Claude Code, codex and deepseek confirm the projection they consumed.Backwards compatibility
Compatible: copy and a comment only; no key renamed, no API or data shape touched.
Removes nothing: the line, its key and its id are unchanged; only the label text changes.
Tests
website/src/test/AgentBackendTab.test.tsx("folds the detail lines behind a disclosure and keeps them out of the count") now looks up the new label. Revertingen.jsonto the old label turns that test red (1 failed, run with-t "detail"); with the change the file passes 117/117.npm run i18n:checkpasses (key parity, placeholders, pseudolocale).scripts/check_comment_history.pywithCOMMENT_HISTORY_BASE_REF=origin/mainpasses.Manual verification
Booted a pod from this worktree, opened Settings → Agent Harness, opened the Claude Code, codex, deepseek and Kiro CLI rows and expanded "Other differences (not counted)". Each shows "Crewmates load their saved agent"; the old string is absent. Claude Code, codex and deepseek also show the "will not take effect" list; Kiro CLI shows none. Translations were reviewed by reading only, not by a native speaker.
Screenshots / video
Claude Code (projecting backend): the new line, with the fields it does not apply listed under it.
codex, deepseek, Kiro CLI
Related Issues
Closes #19066
Refs #16847, #12451
Pattern harvest
Not generalizable: a single label written before a capability set grew; the card already sources per-backend gaps from data, which is what this relies on.
Decision for the owner
The decision is whether the projecting backends get their own card wording (a second line id) or one neutral label covers every backend. This draft picks the neutral label without "whole" and relies on the card's existing "will not take effect" list to name the gaps, because a per-backend label would need a new line id in
backend_cards.pyand a new key in every locale for the same information.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; see the Licensing section of CONTRIBUTING.md.