Feature/tracer #113
Feature/tracer #113
Conversation
…tures-from-events.jsonl Guard planner_input dependency when emitting replay_point requires
…com:AlexanderOnischenko/fetchgraph into codex/reorganize-fixtures-and-migrate-paths
…com:AlexanderOnischenko/fetchgraph into codex/reorganize-fixtures-and-migrate-paths
…com:AlexanderOnischenko/fetchgraph into codex/reorganize-fixtures-and-migrate-paths
…tures-and-migrate-paths Clarify fixture help defaults in Makefile
…com:AlexanderOnischenko/fetchgraph into codex/reorganize-fixtures-and-migrate-paths
…tures-and-migrate-paths Codex/reorganize fixtures and migrate paths
|
@codex, please review |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d9076ec05a
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
tests/test_replay_fixtures.py
Outdated
| bucket_dir = FIXTURES_ROOT / bucket | ||
| if not bucket_dir.exists(): | ||
| continue | ||
| for path in bucket_dir.rglob("*.json"): | ||
| paths.append((bucket, path)) |
There was a problem hiding this comment.
When a replay fixture is exported with --with-requires, export_replay_fixture/write_bundle copies dependent resource files into the same output directory, and those resources can be .json files. _iter_fixture_paths currently glob/rglob’s every *.json under tests/fixtures/replay_points/{fixed,known_bad}, so resource payloads get parameterized as fixtures and then test_replay_fixture fails at assert event.get("type") == "replay_point" because resource files don’t contain a replay payload. This breaks tests as soon as any fixture includes JSON resources; consider filtering by type in _iter_fixture_paths or storing resources in a subdirectory that isn’t globbed.
Useful? React with 👍 / 👎.
…er-v2-features-and-export Tracer: observed-first `replay_case` v2, bundle export/runtime, CLI, and remove generated fixtures
|
@codex, please review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: be0ea0f3e2
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
examples/demo_qa/runner.py
Outdated
| events_path = run_dir / "events.jsonl" | ||
| case_logger = event_logger.for_case(case.id, events_path) if event_logger else EventLogger(events_path, run_id, case.id) |
There was a problem hiding this comment.
Because run_one always instantiates an EventLogger when event_logger is None, per-case events.jsonl emission becomes unconditional. That means batch.handle_batch’s --events off path (which intentionally passes event_logger=None) no longer disables event logging, and the agent still records events to disk; this can unexpectedly write large trace files or capture data that users explicitly opted out of logging. Consider only creating a case logger when event_logger is provided (or add an explicit flag) so the CLI semantics for --events still work as advertised.
Useful? React with 👍 / 👎.
…ing-with-events-off Respect events flag when creating case loggers
This reverts commit 2685c7dc10e1586321b13073d8a586b631aab156.
AlexanderOnischenko
force-pushed
the
feature/tracer
branch
from
764aa87 to
be0ea0f
Compare
…een-to-ensure-validation fixture-green: validate by default, add --expected-from, rollback on mismatch with improved diffs
* Allow explicit disabling of event logging * Refine event logger sentinel typing
* Fix relational selector validation * Add replay validator checks for fixtures * Use shared replay validators in known_bad tests * Require replay validators in fixture-green and fixed tests
No description provided.