Repository navigation
test(stream): replay 3 real codex runs + acceptance (M14 slice 2, partial) - #188
Merged
Merged
Conversation
- Add 3 verbatim codex --json recordings (gocc, codex-cli 0.166.0, ChatGPT login, 2026-11-12): 2 completed runs (ls, cat) + 1 run killed mid-call (sleep 45, SIGTERM/SIGKILL, exit 143). - New tests: clean replay of all 3 (checkPairing 0 violations, every tool_start paired exactly once), killed run closes as abandoned, acceptance 2 (every call terminal after replay + close). - README provenance updated with recording method + flag note (codex 0.166 dropped --full-auto; sandbox danger-full-access + --dangerously-bypass-approvals-and-sandbox is its equivalent). Partial slice: no real claude stream-json recording — the claude CLI on gocc reports 'Not logged in' over SSH and from the GUI-session bridge alike (OAuth token in macOS keychain, unreachable from any non-interactive context); MacBook has no claude CLI. The claude fixtures stay labelled reconstructed. Refs #187.
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Refs #187 (M14 slice 2). Partial: the codex half is done with verbatim recordings; the claude half is blocked (see below). Do not merge until the main round decides whether to hold for the claude recording.
What this changes
--jsonrecordings (raw stdout, nothing edited) as new fixtures intest/fixtures/agent-stream/, recorded 2026-11-12 on gocc with codex-cli 0.166.0 (ChatGPT login) in an empty scratch dir:codex-real-normal-ls.ndjson— 9 lines, 2 shell calls (ls,find … | wc -l), both paired,turn.completedpresent, exit 0.codex-real-normal-read.ndjson— 7 lines, 1 shell call (cat sample.txt), paired,turn.completedpresent, exit 0.codex-real-killed-sleep.ndjson— 4 lines:sleep 45call hasitem.startedand noitem.completed, noturn.completed; process SIGTERM'd onceitem.startedwas observed (+5 s), then SIGKILL; exit 143.test/agent-stream.test.js): clean replay of all 3 throughcheckPairing+ the sink (0 violations, everytool_startpaired exactly once); killed run closes asabandoned; acceptance 2 (after replay +close(), every call has a terminal state — completed/completed/abandoned, 0 left running).--full-auto;--sandbox danger-full-access --dangerously-bypass-approvals-and-sandboxis its documented equivalent and the--jsonline shapes are unchanged.Honest note — claude half still open
No real claude stream-json recording is in this PR. The claude CLI on gocc (2.1.302) reports "Not logged in · Please run /login" both over SSH and when run from the GUI-session bridge (login shell included); its OAuth token lives in the macOS keychain and is not reachable from any non-interactive context available to this worker, and the MacBook has no claude CLI at all. The reconstructed claude fixtures stay labelled reconstructed, and no test claims a claude fixture is a verbatim recording. A real claude recording needs the user to run it in their own terminal — exact commands are in the evidence comment on #187.
No verdict/threshold changes, no check/comment ledger changes, no new adapter.
Tests
node --test test/agent-stream.test.js→ 33 pass, 0 fail (3 new).npm test→ 295 pass / 296, 1 fail:test/i6d-acceptance-demo.test.js(M3 acceptance demo, realxflow auditCLI). Reproduced identically on unmodifiedmain(8081e71, separate worktree, same test 0/1) — pre-existing environment failure, not a regression.