This repository was archived by the owner on Nov 10, 2026. It is now read-only.
Repository navigation
Match Chinese, Japanese and Korean strategy steps by character pairs - #54
Closed
forrestlinfeng wants to merge 1 commit into
Closed
forrestlinfeng wants to merge 1 commit into
forrestlinfeng wants to merge 1 commit into
Conversation
The applied verdict split text into words of four or more letters. Chinese and Japanese have no spaces, so each clause of a step became one long "word" that an answer never repeats verbatim: a turn that followed every Chinese step was judged `unknown`, or `unused` once it ran a command, which tells the Hub the strategy was ignored. CJK runs are now compared by overlapping character pairs; other text keeps its word matching and stopwords. On a real Codex run with a four-step Chinese strategy, the answer that followed it covers 32-69% of each step's pairs and unrelated Chinese answers at most 8%, so the existing 50% step threshold reads it as `used` (3 of 4 steps). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Sentinel 已审核此变更，未发现需要报告的问题。 开发 agent 查询详情： 调用 MCP 工具 由 Sentinel 审核 · 目标提交 fa77856e8a36 |
Contributor
Author
|
Superseded by #55: instead of matching Chinese/Japanese/Korean steps, an injected strategy now counts as used, so no per-language matching is needed. |
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 subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Problem
applied(did the turn follow the injected strategy?) matched a prose step by its words of 4+ letters. Chinese and Japanese have no spaces, so each clause of a step became one long "word", e.g.先定位一个细分赛道和目标人群, that no answer repeats verbatim.Found on a real Codex run (plugin from EvoMap/evolver-codex-plugin#10, same
appliedcode as here) with the prompt我想在小红书里面种草…and a 4-step Chinese strategy: the answer clearly followed all four steps, yet the reuse report saidapplied: unknown(tier: none). With any command in the turn it would have saidunused, a soft negative on the Hub for a strategy that was used.Change
src/applied.js: CJK runs (Han, Hiragana, Katakana, Hangul) are split into overlapping character pairs; other text keeps the existing word matching and stopwords. Threshold (50% of a step's terms) andMIN_STEP_TERMSare unchanged.used; an unrelated Chinese answer on the same platform →unused; a step mixing English words and Chinese.Evidence
Per-step pair coverage on the real run: followed answer 0.61 / 0.60 / 0.69 / 0.32; four unrelated Chinese texts (debugging, a recipe, exam prep, a generic Xiaohongshu sentence) ≤ 0.08. Verdict goes from
unknowntoused(3/4 steps).npm test: 185 pass. The new "follows Chinese steps" test fails on main.Once merged, the same change goes to the Codex, Claude Code and WorkBuddy ports.
🤖 Generated with Claude Code