Finish the legacy Claude TeamCreate retirement in the binary and skills - #736
Merged
clkao merged 2 commits intoAug 20, 2026
Merged
Conversation
The contract already stated the post-retirement shape; the binary and skill
prose still carried the legacy TeamCreate-registry surface. Remove it and
bind the one unbound break-glass template slot that root-caused the live
selected-team red.
Binary (internal/dispatch, internal/claudeteam):
- Drop --team-name/TeamName/teamNamePattern from dispatch build; a stdin
team_name key now degrades to the ignore-unknown-keys path (byte-identical
envelope with or without it). --team-name is a refused usage error (exit 2).
- Every non-bare host=claude dispatch is now unconditionally the merged
.178+ shape (name + run_in_background, no team_name); the legacy envelope
and its stderr advisory (LegacyTeamNameAdvisory) are gone.
- Remove spawn-standing (singular, hard-required a TeamCreate name) and the
legacy --team branch of spawn-standing-all, plus the now-orphaned
MemberExists dedup probe. spawn-standing-all always emits the merged
background shape.
- Retire the dead standing-teammate auto-injection branch in dispatch build:
it only ever fired on the deleted legacy path (merged/bare dispatches
always omitted it, per existing documented behavior).
- keeps, unchanged: reconcile --team-name (auto-team identity, not the
TeamCreate envelope), leadSessionId discovery, contextbudget,
BareModeAdvisory, RenderStandingTeammatesSection.
Root-cause fix (skills/fo-dispatch-recovery): bind the break-glass manual
template's {dispatch_agent_id} slot in both arms with an inline comment at
the fill point (the stage's agent: field; spacedock:ensign default). This
was the one unbound placeholder behind the nondeterministic
TestLiveBreakGlashShimRecovery/selected-team red — the oracle already
asserted the correct post-retirement shape and needs no change.
Skills prose: rewrite commission's Step 3 from a TeamCreate probe to the
shipped SendMessage-availability boot probe; drop stale --team-name mentions
from claude-fo-dispatch.md, fo-dispatch-core.md, codex-first-officer-runtime.md,
and fo-gate-lifecycle.md. Add a contractlint invariant (AC-4) that fails on
any shipped skill instructing a TeamCreate(...) call.
Tests: delete the legacy-only vehicle files (build_team_name_advisory_test.go,
build_teamname_path_test.go, spawn_standing_all_test.go, plus the
now-impossible TestBuildLegacyModeUnchanged and
TestBuildMaxLegalDispatchFilenamesRemainExact scenarios); port the
mode-neutral spawn-standing validation cases (fable model, non-ASCII prompt
escaping, missing-model/bad-enum/trailing-heading) onto spawn-standing-all;
drop team_name from ~20 files' base stdin fixtures and regenerate the
affected goldens. Add a TestMain to internal/dispatch that clears
CLAUDE_CODE_SESSION_ID for the whole test binary: with legacy team_name gone,
every non-bare claude fixture now takes the merged session-token filename
path, which would otherwise leak a real Claude Code session id from a
developer's shell into golden fixtures.
selected-team stays hard-gating and byte-unchanged (no xfail, no registry
edit, no live-test edit) per the settled gate decision.
Net LOC vs the approved ~-500/~47-file estimate: net -832 across 76 files
(product -188/5 files, skills docs -1/6 files, tests -536/29 files, golden
fixtures -107/36 files). Product and docs landed close to estimate; the
tests+fixtures overage (net -643 across 65 files vs estimated -320/~37) is
real, not padding: team_name turned out to be pervasive base-stdin decoration
across ~24 test files (not just the crossproduct family ideation grepped),
and each one's dispatch_file_path is captured byte-for-byte in its own
golden, so removing the field's effect forced far more golden regeneration
than the cycle-1 estimate anticipated. This exceeds the declared
[-250,-750]/<=60-files tolerance; flagging for the gate rather than trimming
correct, necessary changes to fit the number.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Cycle-2 correction: the captain rejected AC-5's mechanism from cycle 1. The
observed live red was a DROPPED subagent_type parameter, not a miscomputed
one — the FO resolved {worker_key} to spacedock-ensign correctly in the same
Agent() call where it omitted subagent_type entirely. A comment next to an
unbound placeholder cannot stop a drop, and is only read by an FO that
already decided to fill the line.
skills/fo-dispatch-recovery/SKILL.md: both break-glass Agent() arms now read
subagent_type="spacedock:ensign" (literal, matching build.go rule 6's
default) with an inline override comment, instead of
subagent_type="{dispatch_agent_id}" plus an explanatory comment. Break-glass
is the only path where a placeholder ever reached a tool call unresolved;
every normal dispatch already gets this filled by dispatch build itself.
Maintenance cost, recorded per the gate: the literal now duplicates rule 6
(build.go:536, subagentType := "spacedock:ensign") and the string is also
hardcoded a third time in the shape oracle
(dispatch_recovery_assert_test.go:176, unchanged, out of scope). No sync
mechanism added — that trade was weighed at the gate and the literal won.
Two additional cheap polish items from validation's list:
- EnumerateDeclaredStandingTeammates(workflowDir, teamName) dropped the
vestigial teamName parameter: its only caller (runShowStanding) passed the
sentinel "_show_standing_" solely to defeat a teamName == "" early return
that predates the merged floor. Signature is now
EnumerateDeclaredStandingTeammates(workflowDir); the stale doc comment
describing bare mode by empty teamName is gone too.
- spawn-standing-all now refuses a legacy --team flag (exit 2), matching
dispatch build's --team-name refusal, instead of silently ignoring it
(exit 0). Added TestSpawnStandingAllTeamFlagRefused.
Not touched, per the gate's explicit scope: the oracle's hardcoded
subagent_type, the unreachable isFile branch, the codex/pi disambiguator —
all recorded findings with promote conditions, deferred on purpose.
go test ./... and -race clean (the same pre-existing, unrelated
TestCodexResolveManifestAgainstInstalledHost failure, present on main too).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Collaborator
Author
Corroborating validation notesA second validator ran an independent pass on this branch after a provider disconnect, before seeing the landed report. It reached the same conclusions on every point and found no contradictions. It correctly declined to append a duplicate stage report — the entity is behind a frozen briefing digest — so its two additional findings are recorded here instead. 1. The show-standing removal is better guarded than the landed report credits.
Re-introducing the injection in a throwaway checkout reds that inverted test. So the removal is protected by a falsifying test, not only by the enumerator parity check the landed report cites. 2. Broader race coverage than reported. Full Neither finding changes the verdict. Both raise confidence in it. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
A First Officer told only "team mode" reached for the advertised
--team-nameflag and invented a team name. The flag selects a dispatch path no supported host can consume. This removes it.What changed
--team-name,TeamName,teamNamePattern,LegacyTeamNameAdvisory, andMemberExists.spawn-standing(singular) andspawn-standing-all --team; both now refuse.subagent_typeto a literal, not a placeholder.TeamCreate(.Evidence
go test ./...and-race: green, one pre-existing environment-local failure reproduced identically onmain.main's own commission text.Net −724 across 78 files, inside the declared tolerance band. The literal was confirmed by running the binary across absent, named, and empty
agent:READMEs.Two deferred risks are recorded in the entity: an oracle that hardcodes the same default, and
{worker_key}as the template's remaining unbound slot.nr
🤖 Generated with Claude Code