Repository navigation
Conversation
…pers The digest, the refresh loop and the tests import a job store, a location classifier, the collector's coverage defaults and two Gmail helpers that live in the private system. These stand-ins provide only that surface. Each module's docstring states how it differs from the private version: no rescore on open, no private default settings, a classifier that matches configured city names and the word London only, and a Gmail client that refreshes a stored token and never starts a consent flow.
pytest.ini puts src on the import path and points collection at tests, so a clean clone needs no environment variables. requirements-dev.txt pins the pytest version the suite was run with.
The self-only recipient guard failed open. mailbox_address() returns an empty string when the Gmail profile lookup fails, and resolve_recipient() then falls back to the token file's account. That field is often empty, in which case the comparison was skipped and the digest went to whatever address was configured, even one that is not the authenticated mailbox. deliver() now refuses a real send when Gmail has not reported the mailbox address, unless allow_other_recipient has been set to true. Dry runs and empty digests are unaffected, as nothing is sent in either case. The new test reproduces the failure: before the fix it did not raise and the message was handed to the sender.
The digest already counted roles excluded by the location gate, but neither the command-line counts line nor the empty-digest text showed that number, so considered roles could not be reconciled with the rest. Both now include it, and a test checks that every considered role falls into exactly one of the reported buckets.
'python scripts/demo.py' seeds a temporary store with seven fictional roles (every employer marked synthetic, every link on example.invalid), prints a SYNTHETIC banner, and runs the real ingestion census and the real digest command as a dry run. It needs only the standard library, never builds a Gmail client, and refuses to write over an existing store. The tests run it with network access and the Gmail client both blocked, and check the banner, the census and digest sections, that every link and employer is invented, and that the counts line accounts for every role.
Several docstrings pointed to SESSION_NOTES.md or AGENTS.md, neither of which is in this repository; the rules they cited are now stated in place. Figures measured on the private system's store (in digest.py, signals.py and refresh.py) are kept but now say where they were measured and that they cannot be reproduced here. The delivery docstring also notes that only the private Gmail helper has a consent-flow fallback, and that a send needs Gmail to confirm the mailbox.
signals-tests.txt is now the output of 'python -m pytest -v' on a clean clone of this branch, with the interpreter path shortened to .venv/bin/python and rootdir to '.'. signals-tests.png is rendered from that text.
…de this repository
…ecord its render config
…ild a bandit, and document the offline run
On a fresh chain scheduled_budget wrote timeout_seconds as the slice (600s), below the project default of 900s, into the highest-precedence budget slot. The timeout is now raised to the default when consumed plus slice is smaller. The existing widen-only test missed this because it pre-seeded 5000s; a new test runs on a fresh store with no scope budgets. The docstrings now state exactly what is guaranteed and what is not compared.
reap_stale_runs cleared any status other than running or cancelling, so an unrecognised status such as queued was treated as finished. It now clears only completed, failed, cancelled, time_limit, limit_reached and interrupted, as the docstring already said, and leaves every other status alone.
render() cut every link at 400 characters, which broke long apply links with tracking parameters. The limit is now 2,048 and control characters are still removed. Tests cover a 1,000-character link and a link with a CR LF in it.
digest.select built its set of enabled countries from the raw location keys, while the classifier reports codes in upper case. A location saved as "fr" therefore matched nothing and its roles were counted as outside configured locations. The keys are now upper-cased before the comparison.
A real send now asks users.getProfile whose mailbox this is, and that call needs gmail.readonly, not gmail.send. A send-only token passed preflight and was then refused on every send with a message that did not mention scopes. REQUIRED_SCOPES now lists gmail.readonly as well, so preflight rejects such a token with the missing-scope reason.
requirements-dev.txt pinned pytest 9.1.1, which needs Python 3.10 or later, so installing it with the python3 that ships with macOS (3.9) failed. The suite passes on pytest 8.4 as well, so the requirement is now a range.
The digest docstring and a test docstring described the search that was offered Shanghai in terms of one person's own location preferences. They now say London plus a few configured countries, and a search limited to configured locations.
Seven docstrings and comments had their citations of AGENTS.md reworded as general project rules. The original wording is restored in each place, with "(not included in this repository)" after the file name so the reference no longer reads as dangling.
The default fixture title and one set of searched terms matched a real search field. They now use a logistics planner and logistics terms; each affected test still checks the same property.
The capture now shows the 85 tests on this branch, taken from a real 'python -m pytest -v' run in a fresh clone with the interpreter path shortened to .venv/bin/python and rootdir shown as '.'. The PNG is rendered from it.
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
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
1 participant
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.
Summary
Repositions the repository around explainable selection with deterministic signals and a measured decision about when not to use machine learning; the decision not to build a contextual bandit is now the centrepiece. It also makes the extract run. Before this change, a clean clone collected no tests at all, because the private store, classifier and Gmail helpers were missing. With minimal stand-ins added, all 85 tests pass offline with only
pytestinstalled, on Python 3.9 or later, and a synthetic demo runs on the standard library alone. Along the way the review found, and this change closes, a path where the recipient guard failed open.What changed
Offline runnability
pytest.iniandrequirements-dev.txtare added.careerops.store,careerops.inventory,careerops.discoveryand the twojob_cv_agentGmail helpers. Each docstring says how it differs from the private version.scripts/demo.pyseeds a temporary store with invented vacancies. It runs the real census and a dry run of the real digest, and never contacts Gmail.docs/images/is regenerated from a clean-clone run.Fixes
Docs
AGENTS.mdkeep their original wording, with a note that the file is not included in this repository.Tests
python -m pytestPYTHONPATH=srcpython scripts/demo.pyEach behavioural fix has a regression test that was confirmed to fail with the fix reverted.
Decisions for review