Skip to content

Audit fixes from the 2026-11 manual test of v0.1.4 - #10

Merged
jehrr merged 3 commits into
mainfrom
audit-2026-11-16
Nov 16, 2026
Merged

jehrr merged 3 commits into
mainfrom
audit-2026-11-16

Conversation

@jehrr

@jehrr jehrr commented Nov 16, 2026 •

Copy link
Copy Markdown
Collaborator

Fixes the defects from the 2026-11-12..16 manual test of v0.1.4 that reproduce on 67be0b3. Each was checked against the code first; the numbers are the report's.

Fixed

  • 1 --url listing links: experience-level= / employment-type= become endpoint filters; any other query parameter is refused with exit 2 naming it (it used to be dropped silently).
  • 2 Dockerfile installs libatomic1 before pip install (Playwright's node driver needs it; the build failed on arm64 and on amd64 with Playwright 1.65).
  • 3 Ctrl+C between pages ends the run as interrupted (status: partial, exit 6) and writes the pages already fetched, in all three engines.
  • 4, 5, 25 diff_runs.py: warns on a listings run that took <100% of its query; refuses api against ssr; "no such file" for a missing input.
  • 6, 12, 22 scraper_api_client.py: a listing page is parsed as the rendered page; an offer page (no parser exists), a foreign host, --timeout outside 1-120 and mode mismatches are refused before the paid request; --mode facets has a default URL; stale help text from another scraper removed.
  • 7, 9, 10 --out directory, --pages < 1 and a missing --slugs-file are refused with exit 2 before any request.
  • 13, 14, 16 help/README wording (--fp-tags, ssr ~50 rows, concurrency only in Playwright).
  • 15 a lone - is not escaped in CSV. 23 --allow-empty + finished empty run is complete. 24 queue_exhausted for a finished offer queue. 34 --concurrency works for an offer queue.
  • 17a a mistyped --proxy is quoted, not ?://?. 21 an explicit empty flag beats .env.
  • 26-28 --fingerprint: Selenium/Puppeteer apply the fingerprint's user agent, screen and timezone; Puppeteer's init script was a syntax error (((() => {...})();)(), reproduced with node); WebGL strings are read from glParameters.
  • 33 the per-page solve allowance is spent when a task is created (captcha_solver.solve_attempts()), not when the check ran.

Second commit: the ones first left out

  • 8 the html category is out of the hint list and the README loop; a misspelt --category gets a warning with the close match.
  • 11 a pyppeteer BrowserError prints a --chromium-path hint (Apple Silicon).
  • 17b a dead proxy ends as proxy_unreachable.
  • 18 sample_output.* regenerated from a live run (7 rows, two with several contract types).
  • 19 Puppeteer answers the proxy's 407 over CDP Fetch. Reproduced first: Network.setRequestInterception wasn't found on Chromium 154. After: against a local authenticating proxy the proxy sees CONNECT justjoin.it:443 ... AUTH, and the browser no longer dies at launch.
  • 20 a proxy that answers 407 to CONNECT is named up front and exits 2 (checked against the same local proxy; the password is not echoed). The 60 s board-count timeout is unchanged.
  • 29 a refused sitemap (403/429/challenge) rotates the exit and ends as exit 3 (blocked_sitemap), in all three engines.
  • 30/31 on the endpoint route a challenge page is recognised from the markup and a refused page is saved as markup. Only partly reproducible from the code (the report's exact text was already a marker), so this covers the other texts.
  • 32 no reload after a reCAPTCHA token: the page's callback runs, or the form is submitted. Checked on a local page in Chromium (token stays, form submits, callback fires). No paid solve was bought, so it is not confirmed against a real site.

Still not done

  • 10: only the missing---slugs-file row of the exit-code table (now exit 2); a nonexistent offer is still exit 5 from the browser engines and 4 from the Scraper API client.
  • The 60 s board-count timeout (part of 20), and --concurrency for Selenium/Puppeteer (README now says so).
  • 5: apply_url is not made route-dependent; the two routes are refused instead.

Evidence

  • Offline suite: 1115 (Playwright env), 1090 (Selenium), 1092 (Puppeteer), 980 (core only), 0 failed. Baseline on 67be0b3 was 1021 / 1004 / 917.
  • 30 new checks (check_audit_*). Run against the old code, they fail (exit codes, stop reasons, - escaping, status, --proxy "", concurrency warning), and the Ctrl+C one kills the suite with a real KeyboardInterrupt.
  • Exercised for real: Chromium 154 with a local authenticating proxy (19, 20), a local page with a form (32), a live fetch of the free endpoint (18). Not exercised: Docker (the CI docker job is green), Selenium over a real browser, paid APIs, the live site through the engines.

🤖 Generated with Claude Code

Petr and others added 3 commits November 16, 2026 11:00
Fixes the bugs from the report that reproduce on 67be0b3: listing-URL
filters (1), Docker libatomic1 (2), Ctrl+C keeps fetched pages (3),
diff_runs route/slice/missing-file guards (4, 5, 25), Scraper API client
refusals and help (6, 12, 22), argument validation (7, 9, 10), help text
(13, 14), CSV lone dash (15), README concurrency (16), proxy error text
(17a), empty flag beats .env (21), --allow-empty status (23), queue
stop reason (24), fingerprint in Selenium/Puppeteer (26-28), solve
allowance counted on purchase (33), offer-queue concurrency (34).

Not fixed here: 8 (hint only, html key kept as a dated snapshot), 11, 17b,
18, 19, 20, 29, 30, 31, 32 - they need a live browser, a paid call or a
design decision; see the PR description.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
- 8: drop the stale `html` category hint, warn on a misspelt --category
- 11: BrowserError in Puppeteer gets a --chromium-path hint
- 17b: a dead proxy ends as proxy_unreachable, not page_load_timeout
- 18: sample_output regenerated from a live run (multi-contract rows)
- 19: Puppeteer proxy login over CDP Fetch (verified on Chromium 154
  against a local authenticating proxy; page.authenticate is dead)
- 20: a 407 from the proxy is named up front, exit 2
- 29: a refused sitemap rotates the exit and ends as exit 3
- 30/31: challenge pages on the endpoint route are recognised from the
  markup, and saved as markup
- 32: no reload after a reCAPTCHA token; callback or form submit instead
  (checked on a local page, no paid solve)

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
@jehrr
jehrr marked this pull request as ready for review November 16, 2026 15:55
@jehrr
jehrr merged commit 2092650 into main Nov 16, 2026
9 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant