技能 2026-07-21 17:51 UTC
distributed-system-testingv4.8
版本【4.8】由提交人 shenli 提交,主要更新类型为:功能特性新增与扩展。提交说明:Fix cross-skill contract drift and repo-hygiene issues from review (#10)。为项目引入了新的能力与模块支持。
Commit提交信息原文
Fix cross-skill contract drift and repo-hygiene issues from review (#10) High priority: - Trim both SKILL.md descriptions under the 1024-char frontmatter limit - Replace the abolished "exit criteria" field with budget-tier language in the executing halt condition, designing field lists, green-but-broken reference, and findings-report checkbox - Make §7.M.S arms always serious with a mandatory per-arm §7.M block, so boundary/fairness scenarios are no longer both required and forbidden a §7.M and the verdict tree can score them - Add the missing "## 2. SUT model" heading (two §2 cross-refs now resolve) - Rewrite README Status: drop dead links into gitignored verification/ and the false per-run-README claim; describe runs in prose - Move plugin.json to .claude-plugin/plugin.json with a corrected schema, add marketplace.json, document /plugin install in README Medium priority: - Add NOT-RUN to the two nine-of-ten verdict lists - Route long-tail scenarios to template §9, not the §7 scenario list - Fix stale oracle-patterns.md §3 (serializability) pointers to §1 - Qualify two cross-skill paths as "the executing skill's" - Add Boundary and Fairness to step 1b's category list - Extend session-verdict rules to cover PARTIAL-* and NOT-RUN - Bring findings-report checklists to full parity (10 red-flag + 14 weak-oracle items, including boundary/fairness) - Add env-capability matrix and budget-tier/verdict table to session log - Relabel evals as manual regression prompts, add evals/README.md - Standardize plan/session output paths on docs/testing-plans/ - Fix a claim×fault -> claim×hypothesis axis error and mixed numbering Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>