Mission Specification: Nightly reds B (2026-10-04)
Mission Branch: kitty/nightly-reds-b-2026-10-04 Created: 2026-10-04 Status: Draft Input: Operator brief: make the stress (#5610), integration (#5611) and integration-slice (#5612) nightly suites green for a reason that can be named, or prove a red is intentional. Evidence run: https://github.com/spec-kitty/spec-kitty/actions/runs/37177460494 (head b2c466d7d1, no newer nightly as of 2026-10-04).
Root-cause evidence for every row lives in research/nightly-red-memo.md.
User Scenarios & Testing (mandatory)
User Story 1 - The nightly stress suite proves concurrent status emits again (Priority: P1)
A maintainer reading the nightly sees the stress suite exercise 20 concurrent bookkeeping transactions against a coordination Mission shaped the way mission create shapes one, and pass.
Why this priority: the stress suite is the only proof that the per-Mission status lock serializes real OS processes; while its fixture is malformed it proves nothing.
Independent Test: run tests/stress/test_concurrent_emits.py; both tests pass, and the fixture's mid8 equals mission_id[:8] with a meta.json that declares the coordination branch.
Acceptance Scenarios:
1. Given a repository with a coordination branch and a Mission meta.json declaring it, When 20 processes each run acquire → append_event → commit, Then all 20 succeed and the event log holds 20 unique valid events.
User Story 2 - Integration oracles track the coordination single-home contract (Priority: P1)
A maintainer sees the integration suite pin the coordination single-home behavior that commit 5b5699e50 shipped on purpose (create seeds the coordination status log; the executor resolves its STATUS directory through the write accessor; the shared read-handle canonicalizer adds two get_main_repo_root reads), instead of the pre-change behavior.
Why this priority: these three tests went red on the nightly the day the change landed; until they are re-pinned they hide any real regression in the same seams.
Independent Test: run the three named integration node ids; they pass and each still asserts the guarantee it protected (surface placement, coordination-aware STATUS directory, exact post-mint ledger).
Acceptance Scenarios:
1. Given a COORD Mission created by create_mission_core, When the golden-path test inspects the coordination status log, Then it finds the seeded log already committed rather than committing an empty one itself. 2. Given a coordination Mission with a husk worktree, When the executor resolves its STATUS directory, Then the spy on the write accessor captures the coordination directory and read_dir(STATUS_STATE) is not called. 3. Given an owned finalize, When post-mint get_main_repo_root reads are counted, Then they equal the re-pinned exact ledger.
User Story 3 - agent config sync --create-missing restores a missing Mistral Vibe pointer (Priority: P1)
An operator whose .vibe/config.toml pointer is missing runs the recovery command that init recommends and gets the pointer back, without the pinned command-skill manifest being rewritten.
Why this priority: it is a user-visible product regression; the documented recovery command silently does nothing.
Independent Test: tests/init/test_init_idempotent.py::test_initialized_clone_vibe_pointer_recovery_and_repeat passes, and the manifest-pinning tests in tests/specify_cli/cli/commands/test_agent_config.py still pass.
Acceptance Scenarios:
1. Given an initialized clone with the vibe skills installed and .vibe/config.toml deleted, When the operator runs spec-kitty agent config sync --create-missing --keep-orphaned, Then .vibe/config.toml exists again and a repeat run reports no changes.
User Story 4 - The nightly integration slice runs in the environment it was written for (Priority: P2)
A maintainer sees the integration slice pass on the runner: the corpus-recovery suite can read its pinned historical commit, the repo-root status-guard self-test parses pytest's CI-mode summary lines, and the implement --recover --json characterization runs against a materialized coordination worktree.
Why this priority: these reds are harness defects first surfaced by the new slice; they do not indicate product breakage but they mask any real signal from the slice.
Independent Test: run the named node ids with CI=true and from a --depth 1 clone where relevant.
Acceptance Scenarios:
1. Given the nightly integration-slice job, When it checks out the repository, Then it fetches full history, like every other job that runs the corpus-recovery suite. 2. Given CI=true in the environment, When the status-guard self-test parses the nested pytest summary, Then an ERROR <nodeid> - <message> line counts as an error for that node id. 3. Given a coordination Mission whose coordination worktree is materialized, When implement --recover --json runs with no crashed sessions, Then it exits 0 with the success envelope.
Edge Cases
- A coordination Mission whose PRIMARY dir is a bare slug (no
-<mid8>suffix): out of scope here; see C-004 and the memo. - A summary line without the CI-mode message suffix (local runs) must still parse.
- A repeat
agent config syncafter the pointer is restored must not report a change.
Requirements (mandatory)
Functional Requirements
| ID | Title | User Story | Priority | Status | Delivery | No-op passable? |
|---|---|---|---|---|---|---|
| FR-001 | Stress fixture is a well-formed coordination Mission | As a maintainer, I want the concurrent-emit stress fixture to declare its coordination branch in meta.json and use mid8 == mission_id[:8] so that the stress suite exercises the real coordination write path. | High | Open | [ratchet] | no |
| FR-002 | Golden-path oracle accepts the create-time coordination seed | As a maintainer, I want the placement golden-path COORD case to assert the seeded coordination status log instead of committing an empty one so that it pins the intended create-time seed. | High | Open | [ratchet] | no |
| FR-003 | Executor STATUS directory is observed through the write accessor | As a maintainer, I want the coordination-read residuals proof to spy on write_dir(STATUS_STATE) and assert read_dir(STATUS_STATE) is not used so that it still proves the STATUS leg stays coordination-aware. | High | Open | [ratchet] | no |
| FR-004 | Owned-finalize ledger re-pinned to the shared canonicalizer | As a maintainer, I want the exact post-mint get_main_repo_root ledger to carry the two reads the shared read-handle canonicalizer added, with a dated rationale, so that the ledger stays exact. | High | Open | [ratchet] | no |
| FR-005 | Vibe pointer recovery | As an operator, I want agent config sync --create-missing to restore a missing .vibe/config.toml for an already-installed vibe agent so that the recovery init recommends works. | High | Open | [build] | no |
| FR-006 | Recover characterization runs on a materialized coordination worktree | As a maintainer, I want the implement --recover --json coordination characterization to use a materialized coordination worktree, as mission create produces, so that it characterizes the success envelope rather than the intended fail-closed refusal. | Medium | Open | [ratchet] | no |
| FR-007 | Status-guard self-test parses CI-mode summary lines | As a maintainer, I want the nested-pytest summary parser to accept the - <message> suffix pytest appends when CI is set so that the self-test holds on CI runners. | Medium | Open | [ratchet] | no |
| FR-008 | Integration slice checks out full history | As a maintainer, I want the nightly integration-slice job to fetch full history so that the corpus-recovery suite can archive its pinned historical commit. | Medium | Open | [ratchet] | no |
Non-Functional Requirements
| ID | Title | Requirement | Category | Priority | Status |
|---|---|---|---|---|---|
| NFR-001 | No weakened checks | Zero added retries, skips, xfails, deselections, timeout or budget changes; every pre-existing assertion is kept (only stale literals change, each with a rationale). | Reliability | High | Open |
| NFR-002 | Red-first evidence | Every repaired node id is shown red on skupstream/main (bc8d53d09) and green on the branch, with commands and counts recorded. | Reliability | High | Open |
Constraints
| ID | Title | Constraint | Category | Priority | Status |
|---|---|---|---|---|---|
| C-001 | Do not touch PR #5617's files | No change to any file PR #5617 modifies (notably tests/_support/shared_build_artifacts.py). | Technical | High | Open |
| C-002 | Targeted runs only | Tests run by named node id or file, foreground; no directory-wide, nightly-wide or make test-full runs. | Technical | High | Open |
| C-003 | No suite-wide harness rework | Anything #5353-sized is reported, not done. | Technical | High | Open |
| C-004 | Bare-slug coordination consolidate is reported, not changed | The test_explicit_delete_override_still_reachable red (a coordination Mission whose primary dir lacks the -<mid8> suffix cannot consolidate onto a protected target since 5b5699e50) needs a design decision across the commit router and the reconciliation gate; it stays red and is reported. | Technical | High | Open |
Success Criteria (mandatory)
Measurable Outcomes
- SC-001:
tests/stress/test_concurrent_emits.pypasses 2/2 (red 1/2 on base). — [ratchet] · no-op passable: no - SC-002: The three coordination single-home integration node ids pass (red on base). — [ratchet] · no-op passable: no
- SC-003: All 12 integration-slice node ids from run 37177460494 pass locally under the nightly's conditions (
CI=true; full history for the corpus suite). — [ratchet] · no-op passable: no