Skip to content

revert: restore subagent persistence and codex contract tests broken by #11960 - #12088

Merged
bolichen97 merged 1 commit into
mainfrom
revert/11960-stale-codex-continuation
Sep 19, 2026
Merged

bolichen97 merged 1 commit into
mainfrom
revert/11960-stale-codex-continuation

Conversation

@bolichen97

@bolichen97 bolichen97 commented Sep 19, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

main is red since 6c21f0d4d9 (#11960), merged one second before #11947 (whose own breakage #12091 has already reverted). #11960 was authored against a base that predates both #11780 (bc0d2b0cfc) and #11946 (bf3fccf7d5), and its merge re-applied that stale state:

Why it matters

Every PR now inherits a red CI workflow (Backend Tests, Coverage Gate, Real Adapter Contract Tests), so PR Readiness cannot pass for anyone until main is repaired. With #11947 already reverted on main (#12091), this is what still keeps main red.

What changed (motivation → approach → change)

A clean git revert of 6c21f0d4d9, no other edits. The feature (continuing a codex subagent from its persisted thread) is worth re-landing, but on the current subagent_persistence API and with #11946's contract-test markers kept; that is a fresh PR against green main.

Backwards compatibility

Compatible: restores the codex subagent continuation code, tests and docs to the state before #11960; nothing that worked before that merge changes.

Tests

No new tests. On this branch test/test_codex_session_mcp.py carries the three real_adapter markers again. On this branch (rebased onto main after #12091) test_subagent_continuable.py, test_codex_session_mcp.py, test_member_memory_runtime.py and test_subagent_persistence*.py pass locally: 469 passed, 3 skipped; mypy is clean on context.py, dashboard/state.py and subagent_persistence.py.

Manual verification

N/A — the failures are a missing test-only attribute and missing pytest markers; the revert restores both verbatim.

Related Issues

no linked issue: reverts #11960; follows #12091 (revert of #11947).

Pattern harvest

Rule candidate: review-prompt
Pattern: two PRs merged within the same second on stale bases each re-applied files a refactor merged hours earlier had rewritten; PR Readiness should not report passed for a head whose last CI ran against a base that has since moved in the PR's own files (green_age.py exit 30 measures exactly this locally).

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

@bolichen97
bolichen97 requested review from a team and chenmingwei23 September 19, 2026 20:38
@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Sep 19, 2026
@github-actions

github-actions Bot commented Sep 19, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Fable 5) — ✅ PASS

Design-level review of 15a4514898212726297d8af0c1004efd162701f8 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

Confirm the three red lanes (Backend Tests, Coverage Gate, Real Adapter Contract Tests) go green on this head before merging.

[DESIGN-REVIEWED] 15a4514

@github-actions

github-actions Bot commented Sep 19, 2026 •

Copy link
Copy Markdown
Contributor

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of 15a4514898212726297d8af0c1004efd162701f8 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 15a4514

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 15a4514898212726297d8af0c1004efd162701f8: <one-sentence reason>

@github-actions

github-actions Bot commented Sep 19, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Fable 5) — ✅ PASS

Premise-level review of 15a4514898212726297d8af0c1004efd162701f8 — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

First-Principles-Verdict: PASS

Confirm this diff is byte-identical to git revert 6c21f0d4d9 — the reviewer cannot run git, and one stray edit would ride under the revert's blanket justification.

What this change ships

Inventory (8 items) — 8 justified

Intent: repair red main by reverting #11960, which merged stale-base files over #11780's API and #11946's test markers — a FIX.

  1. Backend Tests / Coverage / Real Adapter lanes go green — 25 tests no longer call a removed helper — justified
  2. Three codex live tests carry real_adapter markers and the fail-closed gate again — justified
  3. Codex subagent continuation removed; spawn_continue answers conversation_gone again — justified
  4. Codex subagents get dedicated sessions again (out of ACP_BACKENDS_SESSION_SHARING) — justified
  5. Live load-after-close measurement test and its driver deleted — justified
  6. Shared-bind cleanup identity reverts to the constant kiro label — justified
  7. Specs and RFC §2.7/§5 text revert to pre-feat(acp): continue a codex subagent from its persisted thread #11960 state, same commit as the code — justified
  8. Restricted-run state-lifetime tests restored — justified

All eight are one atomic git revert; the deleted pins are the reverted commit's own, and the description names the reversal and the re-land plan explicitly.

[FIRST-PRINCIPLES-REVIEWED] 15a4514

@github-actions

github-actions Bot commented Sep 19, 2026 •

Copy link
Copy Markdown
Contributor

Opus 4.8 Review — ✅ no blocking findings

Reviewed 15a4514898212726297d8af0c1004efd162701f8 — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 15a4514

Verdict parsed from the review's SHA-scoped output markers for commit 15a4514898212726297d8af0c1004efd162701f8.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 15a4514898212726297d8af0c1004efd162701f8: <one-sentence reason>

@bolichen97
bolichen97 force-pushed the revert/11960-stale-codex-continuation branch from 144c367 to 15a4514 Compare September 19, 2026 21:13
@bolichen97 bolichen97 changed the title revert: undo the two stale-base merges (#11960, #11947) that broke main revert: restore subagent persistence and codex contract tests broken by #11960 Sep 19, 2026
@github-actions github-actions Bot added readiness: checking Automated validation is still running and removed readiness: passed Eligible automated validation passed for the current revision labels Sep 19, 2026
@bolichen97
bolichen97 merged commit ddea8d7 into main Sep 19, 2026
87 of 94 checks passed
@bolichen97
bolichen97 deleted the revert/11960-stale-codex-continuation branch September 19, 2026 21:27
@github-actions github-actions Bot removed the readiness: checking Automated validation is still running label Sep 19, 2026
@iamwhatever

Copy link
Copy Markdown
Contributor

Re-landed as #12101, rebased onto main after this merged.

One addition to the diagnosis here, recorded because the next reader of this revert will want it: the stale merge clobbered three upstream commits, not two. Alongside #11780 and #11946 it also collapsed test/test_acp_capability_sets_leaf.py's ACP_BACKENDS_COMPACT pin from the three members #11944 (d92b8c4162) gave it back to two, so test_membership_is_unchanged_by_the_move was red on the pre-revert main as well. This revert fixed it along with the rest; I only mention it because the body names two causes and the count matters if anyone re-derives this later.

It surfaced from diffing every file the merged commit touched against the revert base and confirming each changed line belonged to the feature, rather than repairing only the two reported failures. #12101 is built the same way: for each contested file it takes main's copy and appends only #11960's own additions, so no file is carried over from the old commit.

Two numbers from #12101, for the lane specifically: pytest -m real_adapter over test_codex_session_mcp.py + test_opencode_session_mcp.py collects 4, the floor's value — the three markers and the require_real_adapter guard are main's, untouched. The new prompting test deliberately carries no marker: the lane asserts on its junit report that no selected case was skipped, and it holds no model credential, so marking it would trip that assertion rather than widen coverage.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants