Skip to content

fix: correct improvement report accuracy (reactiveCompact + speculation)#35

Closed
wenshao wants to merge 1 commit intomainfrom
fix/improvement-report-accuracy
Closed

fix: correct improvement report accuracy (reactiveCompact + speculation)#35
wenshao wants to merge 1 commit intomainfrom
fix/improvement-report-accuracy

Conversation

@wenshao
Copy link
Copy Markdown
Owner

@wenshao wenshao commented Apr 3, 2026

Summary

Fixes 2 factual errors in qwen-code-improvement-report.md (written by qwen3.6-plus in PR #34):

1. reactiveCompact.ts does not exist

  • Before: "4 层:microCompact, autoCompact, reactiveCompact, sessionMemoryCompact"
  • After: "4 层:microCompact, apiMicrocompact, autoCompact, sessionMemoryCompact"
  • Evidence: ls services/compact/ shows no reactiveCompact.ts; apiMicrocompact.ts (154 lines) exists

2. Qwen Code speculation is complete, not "有骨架但功能不完整"

  • Before: "564 行,有骨架但功能不完整"
  • After: "v0.15.0 已实现完整系统(563 行 + overlayFs + speculationToolGate),默认关闭"
  • Evidence: PR #2525 (2026-04-03) merged full followup system; enableSpeculation: false is the only gap

Test plan

  • wc -l verified: speculation.ts=991, apiMicrocompact.ts=154, reactiveCompact.ts=DNE

🤖 Generated with Claude Code

1. reactiveCompact.ts does NOT exist — changed to apiMicrocompact.ts
   (API-native context management: clear_tool_uses / clear_thinking)

2. Qwen Code speculation is NOT incomplete — v0.15.0 (PR #2525) has
   full system: 563-line speculation.ts + overlayFs + speculationToolGate
   + acceptSpeculation + pipelined suggestions. Only default-disabled.

3. Claude Code speculation.ts: 992 → 991 lines (wc -l verified)

4. Updated summary table: speculation gap from "中等差距" to "小差距(仅默认值)"

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@wenshao
Copy link
Copy Markdown
Owner Author

wenshao commented Apr 3, 2026

修正已合入 PR #34 分支(commit 2cd9e92),此 PR 不再需要。

@wenshao wenshao closed this Apr 3, 2026
wenshao added a commit that referenced this pull request Apr 4, 2026
p2-stability #35: System prompt behavioral guidance for dangerous
operations — 4 categories of risky actions, behavioral principles
(investigate before deleting, resolve conflicts not discard), and
approval scope limitation (one-time approval != permanent).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
wenshao added a commit that referenced this pull request Apr 4, 2026
p0-p1 #47: System prompt content gaps — OWASP security types, prompt
injection detection guidance, code style constraints (no extra features/
abstractions/docs), output format rules (file_path:line_number,
owner/repo#123).

Also added p2-stability #35: dangerous action behavioral guidance.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
wenshao added a commit that referenced this pull request Apr 5, 2026
Closed PR #75 — 10 of 16 items duplicated existing 187. Source code
references all verified correct (glm-5.1 self-corrected 8 errors).

Extracted 4 new items to p2-tools:
- #35 preserveQuoteStyle (quote style preservation in edits)
- #36 areFileEditsInputsEquivalent (skip duplicate edits)
- #37 MCP channel plugin allowlist
- #38 SDK message type enrichment (11 → 30+ types)

Total: 191 items.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
wenshao added a commit that referenced this pull request Apr 5, 2026
PR #105: 0/5 new (all duplicated).
PR #108: 5/5 new (useSSHSession, useSwarmPermissionPoller, useTasksV2,
  useArrowKeyHistory, useCanUseTool).
PR #109: 3/5 new (useSearchInput, usePasteHandler, useScheduledTasks).

New P3 items: #35-#42. Total: 225 items.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
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