[napetrov]

Part of #413 (A10).

Summary

_flag_description
  letters outside covered scripts > covered letters -> None (not checked)
  WHEN_RE per language: en, es, fr, de, pt, ru, zh, ja, ko
    Latin cues: word boundaries
    CJK cues: substrings, no word boundaries
  no cue -> description-no-when

• Bare 使用 and 用于 are not cues (they mean "use").
• Negative tests cover the common false passes: 当天, 相当, 同时, 实际, 国際, 交際, 一時に, 平時は, 用户说明.

Evidence

• Before: corpus eval (description-no-when flag rate) CJK 86.6%, Latin 40.0%, total 718,264 flagged.
  After: CJK 48.9%, Latin 37.5%, total 629,620 flagged.
• Precision sample of rows the check now clears: 39/40 CJK and 30/30 Latin are genuine when-less descriptions.
• tests/test_description_when_cues.py: 95 passed; ruff, mypy, and make ci-local static and typecheck pass.

Merge Danger

Door: two-way (the check is a warning; reverting restores the English-only check)

Blast Radius: catalog validation warnings only; no change to routing or grading. Some CJK skills that used to warn now pass, and some Latin-script skills with no when cue still warn.

Declined, recorded in the PR: Polish, Dutch, Turkish and Vietnamese cue lists; descriptions in uncovered scripts stay unchecked (documented in docs/external-content.md); zh, ko, es, pt and ru recall gaps; 時には and 平時に treated as scope rather than a trigger.

🤖 Generated with Claude Code

[Copilot]

🟡 Changes recommended

Multilingual when-clause detection incorrectly accepts occasional-action phrases, capability descriptions, and compound words, suppressing valid description-no-when warnings.

4 open issues

• Exclude occasional-report phrases from when-clause cues · New
• Exclude 充当 from the Japanese 当…時 cue · New
• Avoid matching Japanese when cues inside compound words · New
• Require invocation context for the Japanese trigger cue · New

What changed in this PR

4 new findings.

This PR broadens catalog validation’s description-no-when check beyond English, reducing warnings for descriptions with native-language invocation cues.

Changes:

• Adds multilingual cue patterns and skips descriptions predominantly written in uncovered scripts.
• Adds positive, negative, and end-to-end regression tests.
• Documents language coverage, limitations, and changed warning behavior.

File	Description
dokimi/​tests/​test_description_when_cues.py	Tests native cues, misleading expressions, script exclusions, and catalog warnings.
dokimi/​dokimi/​external/​validate.py	Adds multilingual cue detection and script-coverage handling.
dokimi/​docs/​external-content.md	Explains supported languages and uncovered-script behavior.
dokimi/​changelog.d/​20261106_120000_napetrov_description-when-multilingual.md	Records the validation change and recommends rerunning catalog checks.

_{Review effort: Balanced}

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.14%. Comparing base (15d840d) to head (60d7c76).
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #454   +/-   ##
=======================================
  Coverage   95.14%   95.14%           
=======================================
  Files         373      373           
  Lines       49271    49281   +10     
  Branches     8467     8468    +1     
=======================================
+ Hits        46879    46889   +10     
  Misses       1515     1515           
  Partials      877      877           

Files with missing lines	Coverage Δ
dokimi/dokimi/external/validate.py	96.33% <100.00%> (+0.13%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[napetrov]

Its four open issues are the threads above, each fixed in the commit named there.

This is the Codecov report. It makes no claim, so nothing changes.