fix(tests): use isCompressSuccessText in export-cmd setupSession (stale /1 block/ after #376/#377)

[ranxianglei]

Context

Issue #389: 7 subtests in tests/export-cmd.test.ts fail on master. The shared helper setupSession() asserted /1 block/ against the compress tool result, but since #376/#377 compress returns the span-form panel (blocks: b1=m00002) instead of the legacy count form (1 block).

Fix

Replace the regex assertion with the production success predicate isCompressSuccessText() from src/compress-tool.ts — the same predicate the compress retry loop uses and that compress-retry / compress-loop-breaker / compress-tool tests already assert against. It encodes "completed run that created ≥ 1 block" and accepts BOTH panel forms (legacy count + #376 span), so the fixture no longer couples to panel wording.

Verification

• Before (origin/master 2bdc831): 3 pass / 7 fail, error verbatim matches the issue report.
• After: tests/export-cmd.test.ts 10/10 pass; full suite npm test → 722 pass / 0 fail / 3 skip; npm run typecheck clean.

Fixes #389

[github-actions]

📦 Built Extension Artifact

Branch: 2026-09-12_fix-export-cmd-stale-assertion (816969e)

Option A — Install from npm PR tag (recommended)

pi install npm:billion-context-pi@pr-390

Each push to this PR publishes a new version under the pr-390 npm tag.

Option B — Download artifact

1. Download the artifact from the Actions run
2. Extract the tarball and install:

tar xzf billion-context-pi-pr390.tgz
pi install ./package

---

This comment is automatically updated on each push.

[ranxianglei]

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 ## PR review — 已验证,可以合并(请人工合并)

Diff 整洁度:单 commit(1d49203),单文件 tests/export-cmd.test.ts,+2/-1(import + 断言替换),无无关改动。✅

谓词正确性:isCompressSuccessText(src/compress-tool.ts:174,exported)= compressPanelBlocks(text) > 0;compressPanelBlocks 同时接受旧计数形式(… N blocks))与 #376 span 形式(blocks: bN=mXXXXX),语义为「completed run that created ≥ 1 block」——正是 compress 重试循环的生产判定,compress-retry / compress-loop-breaker / compress-tool 三个测试文件也已在用。✅

无覆盖损失:helper 断言从「恰好 1 block」弱化为「≥1 block」,但各子测试独立断言精确数量(assert.equal(summaries[0]!.blocks, 1)、assert.match(text, /blocks=1/)),该属性仍被覆盖。✅

实证验证(干净 worktree,非引用自述):

• Before(origin/master 2bdc831):tests/export-cmd.test.ts → 3 pass / 7 fail,报错逐字一致(actual: '▣ ACP | 8.0K → 6.2K tokens (~1.8K reclaimed, blocks: b1=m00002)')
• After(PR 分支):export-cmd 10/10;全量 npm test → 722 pass / 0 fail / 3 skip;npm run typecheck clean

⚠️ 重复筛查发现:存在竞争修复 PR #386

查重时发现本 PR 与以下条目修的是同一行同一缺陷:

条目	方案	状态
#385(首个报告,来源 #375 分析)	—	open
PR #386(fixes #385):/1 block/ → /blocks: b1=/	正则,仍耦合面板措辞且耦合具体块标签 b1=	open
PR #388(更早一次尝试,/blocks: b\d+=/)	—	closed(未合并)
#389(再报告,来源 #380 rebase 分析)	—	open
PR #390(本 PR):改用生产谓词 isCompressSuccessText()	语义判定,兼容新旧两种面板格式	open

#386 我也实测过:同样 10/10 通过——两个 PR 都是有效修复,但改同一行,不能同时合并。差异点:#386 的正则把 fixture 重新绑回面板措辞(下次面板再改版还会漂),#390 用生产谓词彻底解耦,方案更优;issue 层面 #389 与 #385 也是同一问题的两次报告。

建议:合并本 PR(#390),关闭 #386 及 #385 为重复(#389 随本 PR 合并自动关闭);或反向操作,但我不推荐保留措辞耦合版。取舍由你拍板。

一句话中文摘要:审查并实证了本 PR(master 上 7 个 export 子测试因 #376/#377 面板改版而红的修复,改用生产语义谓词而非正则,全量 722/0/3、typecheck 干净,diff 收敛可合并);另查出竞争修复 PR #386 与本 PR 修同一行,建议二选一优先保留本 PR。