TEST-ONLY: #1250 launch-window private config fixture correction
Date: 2026-10-14

Source PR: https://github.com/sleep2agi/agent-network/pull/2754
Source head: 851614f27185e8427c84c428b9e35dc4bc323137
Ordinary source main: 8dbe728be8dbe0f52584a79ceab4aedb98e15e43
Merged: 2026-10-14T15:19:27Z; fresh main ancestry passed.
Source and ordinary-main tree: 745019f4547aac4edda66aad693de82dfb4110d3
All ten protected exact-head checks passed. Wider Actions QA/full regression
were still pending at source merge; this report does not claim their PASS.

Qualified Docker inputs
-----------------------
agent-node/src tree: 87545525bbf964543f89545c8d40e2e5f73831f9
tests/test894-launch-window tree: b90a6bfeaf25674409e912c212410c75b19ac66d
Image: sha256:dadda5881dd043718f94d6bf54d086246451c8f00a1faaa1d0c82bd01c5775dd
Its image revision label stamps the exact source head above; other OCI labels
are inherited from the pinned Bun base and are not repository provenance.
Bun1.3.14 base digest is pinned in the independent Dockerfile. Non-root runtime,
--init/--network none, no external models/logins/history. Build+existing run.sh
completed with exit0 inside one 110-second hard local cap. Only this created
container ID was removed; no shared containers or services were interrupted.
No new tests, guard changes, assertion changes, global package or production ops.

Result and interpretation
-------------------------
Existing launcher-health unit:12 pass/0fail. Existing process probe:five PASS.
The bypassed-wait negative exits1 at the exact original post-exec readiness
assertion; runner verifies that reason and emits final PASS (whole suite exit0).
It is an expected RED, not an uncorrected baseline test failure.
The selected and foreign config paths now each have an exclusive0600 synthetic
JSON file inside a fresh private test directory. Existing health inspection
reads config before process validation. Without both readable configs, the
probe stopped at generic missing/unsafe record instead of its original pre-exec
or bridge-config mismatch assertion. Production validation remains unchanged.

Previous actual RED
-------------------
Run38583114561/job115918923788 at43c1f9efc failed at original probe44:
expected TUI session mismatch, actual missing or unsafe launch/attach record.
Artifact11904224309 SHA2564864b06869897c996993a2f46169eda568d25306661ba5d504eea3f82b81b60d.
Base and candidate guard/probe/bridge were identical then. No separate base
Docker reproduction was performed; source inspection plus the corrected
actual probe supports the fixture diagnosis, not a claim of old base PASS.
The older report-test1245-v1-migration-plan.txt retains its historical RED.

Six reconstruction answers and scope
-----------------------------------
1. This fixture starts through tests/test894-launch-window/Dockerfile and
   run.sh, spawning only synthetic bridge/serve/TUI child processes. It is not
   a new persistent service or production startup change.
2. No proxy/tunnel/port changes; the qualified process suite runs network none.
3. Selected/foreign configs are fresh synthetic empty JSON; no credentials,
   production environment or original histories are needed/read/copied.
4. Rebuild the exact selected Git revision with the pinned Dockerfile and run
   the existing suite. Verify unit/probe exit0 and the negative's exact reason.
   This does not verify an installed CLI/native binary on an original machine.
5. Roll back the fixture source by selecting its prior software revision; no
   runtime binding/data changed to restore. Keep the original RED evidence.
6. Git reproduces software and this disposable fixture, not original native
   history, identities/login, encrypted backup retention or restore authority.
   Default SAFE migration, atomic journal and final lossless recovery remain
   independent #1246/#1238/#1239 gates. No npm/installer/deployment or adoption.

Full actual baseline output
---------------------------
bun test v1.3.14 (0d9b296a)

agent-node/src/runtime/opencode-copresence/launcher-health.test.ts:
(pass) daemon OpenCode live launch generation > matching bridge, child serve and same-session TUI [0.78ms]
(pass) daemon OpenCode live launch generation > missing/dead and PID-reused processes all fail closed [0.37ms]
(pass) daemon OpenCode live launch generation > old record and wrong TUI generation rejected [0.09ms]
(pass) daemon OpenCode live launch generation > unrelated live serve/config/TUI rejected [0.18ms]
(pass) daemon OpenCode live launch generation > exit zero required independently of health [0.12ms]
(pass) daemon OpenCode live launch generation > malformed records, unknown process identity and missing proof refuse [0.49ms]
(pass) daemon OpenCode live launch generation > real child proof private, replaced generation survives old cleanup [11.77ms]
(pass) daemon OpenCode live launch generation > daemon removes only the exited PID and records verified bridge [1.12ms]
(pass) bounded attach exec readiness > argv class separates pre-exec shell, wrong session, and the two-arg contract [0.09ms]
(pass) bounded attach exec readiness > rechecks the unchanged full guard until exec makes the same session ready [0.88ms]
(pass) bounded attach exec readiness > wrong session and a deadline already reached do not pass or gain budget [0.57ms]
(pass) bounded attach exec readiness > linux pre-exec cmdline is TUI session mismatch and the same pid matches after exec [96.49ms]

 12 pass
 0 fail
 50 expect() calls
Ran 12 tests across 1 file. [246.00ms]
PASS: real launcher exit0 + live same-generation records BEFORE exec precisely rejects TUI session mismatch
PASS: AFTER exec same PID/start ticks/generation passes original health guard
PASS: repeated probes never accept a foreign generation
PASS: reused PID/start ticks rejected
PASS: foreign config and nonzero launcher remain rejected; restored exact identity healthy

Full actual bypassed-wait negative output
----------------------------------------
PASS: real launcher exit0 + live same-generation records BEFORE exec precisely rejects TUI session mismatch
52 |     ? Promise.resolve(check()) // witnessed old one-shot behavior; test only
53 |     : waitForLaunchHealth(temp, config, launchedAt, Date.now() + 2000);
54 |   await Bun.sleep(100);
55 |   assert.deepEqual(check(), {ok: false, reason: 'TUI session mismatch'});
56 |   writeFileSync(gate, 'release');
57 |   assert.equal((await pending).ok, true, 'bounded full-identity wait did not observe exec');
              ^
AssertionError: bounded full-identity wait did not observe exec

false !== true

 generatedMessage: false,
     actual: false,
   expected: true,
   operator: "strictEqual",
       code: "ERR_ASSERTION"

      at /fixture/probe.ts:57:10

Bun v1.3.14 (Linux x64 baseline)

Runner final verdict
--------------------
PASS: bypassed wait rejected at exact post-exec readiness assertion
Suite exit0; negative exit1 matched the exact readiness assertion.
