fix(slow_skip): documented 60s per-script cap is wrong — real cap is 300s/1800s

[Jammy2211]

Overview

The SLOW-skip convention documents a "60s per-script timeout cap". No such cap is enforced anywhere. The real values are 300s (smoke/default) and 1800s (mode=release).

This is not cosmetic. A 5× understatement of the cap biases every "is this script too slow to un-skip?" judgement in the same direction — toward parking scripts that would in fact pass. It did exactly that during the triage of autolens_workspace_test#193 on 2026-07-22: a clean run of database/scrape/general measured 214s, which reads as a clear SLOW-skip against 60s but has ~86s of headroom against the real 300s cap. The correct call was to leave it un-skipped.

Where the real cap lives

• 300s — autobuild/build_util.py:12, TIMEOUT_SECS = int(os.environ.get("BUILD_SCRIPT_TIMEOUT", "300")), matching autobuild/run_all.py:49 DEFAULT_TIMEOUT_SECS = 300.
• 1800s — PyAutoHeart/.github/workflows/workspace-validation.yml:308 sets BUILD_SCRIPT_TIMEOUT: "1800" for the release run.

Plan

• Correct the three sites in autobuild/slow_skip_check.py, sourcing the number from build_util.TIMEOUT_SECS instead of hardcoding a second copy that can drift again — and let run_all.py pass the timeout actually in effect for the run.
• Correct the identical stale line-9 header comment in all seven workspaces' config/build/no_run.yaml.

Detailed implementation plan

Affected Repositories

• PyAutoHands (primary — the code and the report text)
• autolens_workspace, autolens_workspace_test, autogalaxy_workspace, autofit_workspace, autofit_workspace_test, HowToLens, HowToGalaxy (comment-only, one line each)

Suggested branch: feature/slow-skip-timeout-cap-doc

The three stale sites in autobuild/slow_skip_check.py

1. Line ~18, module docstring — "SLOW entries indicate scripts that exceed the 60s per-script timeout cap".
2. Line ~140, _BANNER_CONFIG["slow"]["footer"] — "These scripts are skipped because they exceed the 60s per-script cap". Printed to stdout by run_all.py at lines 174 and 237.
3. Line ~159, _REPORT_CONFIG["slow"]["intro"] — "{n} script(s) are being skipped because they exceed the 60s per-script timeout cap". This one lands in the slow-skip section of every mega-run report.

Implementation Steps

1. autobuild/slow_skip_check.py: add import build_util (the established flat-import idiom in this package — generate.py, run.py, run_python.py, generate_autofit.py, generate_markdown.py all do it). Replace the hardcoded 60s in all three sites with the value from build_util.TIMEOUT_SECS, noting the release override.
2. Thread an optional timeout_secs argument through format_warning_banner and format_report_section, defaulting to build_util.TIMEOUT_SECS, so the text reflects the cap actually in force.
3. autobuild/run_all.py: pass timeout_secs (already in scope from args.timeout_secs) at the four banner/report call sites, so a --timeout-secs override is reported accurately rather than as the default.
4. Seven workspaces: fix line 9 of config/build/no_run.yaml.

Key Files

• autobuild/slow_skip_check.py — the three stale sites.
• autobuild/run_all.py — banner call sites (174, 237) and report section.
• autobuild/build_util.py — TIMEOUT_SECS, the single source of truth.
• <workspace>/config/build/no_run.yaml — line 9 header, × 7.

Follow-up worth considering (separate task)

The four database/scrape/* siblings (multi_analysis, slam_general, slam_multi_one_by_one, slam_pix) are still tagged # SLOW 2026-04-10 - exceeds 60s timeout. Those markers were written against the wrong cap; some may clear the real 300s cap and be un-parkable. Worth re-timing once this lands.

Original Prompt

Click to expand starting prompt

PyAutoMind/draft/bug/pyautobuild/slow_skip_60s_timeout_cap_is_wrong.md

[Jammy2211]

8 PRs open.

Code: #173 — slow_skip_check.py now sources the cap from build_util.TIMEOUT_SECS across all three stale sites (module docstring, _BANNER_CONFIG["slow"]["footer"], _REPORT_CONFIG["slow"]["intro"]); format_warning_banner / format_report_section take an optional timeout_secs, and run_all passes the timeout actually in force. 132 tests pass; verified rendering 300s by default and 1800s when overridden.

Header comments (one line each): autolens_workspace#313 · autolens_workspace_test#194 · autogalaxy_workspace#144 · autofit_workspace#105 · autofit_workspace_test#62 · HowToLens#43 · HowToGalaxy#33

Two things deliberately left alone, both detailed in #173: the five remaining 60s strings inside individual entries' own reason text (rewriting them without re-timing the scripts would assert something unverified — the re-timing is the filed follow-up), and the fact that format_report_section is imported at run_all.py:164 but never called, so the corrected report intro has no live call site.