Phase A + B: bug fixes + perspective structure consistency#3
Merged
Conversation
A1: wuying-browser.md hardcoded ~/mba/ paths broke the ${SKILL_DIR}
abstraction documented in SKILL.md L75-95. Three sites (helper script
path + 2 ssh commands) now use ${MAC_MBA_ROOT:-~/mba}; added the
symbol to the path resolution table.
A7: SKILL.md L451 said "Default Dimensions (7)" but references/dimensions.md
defines 9 (1-7 core + 8-9 advanced). Phase 2F builds filenames
dimension_{n}_{slug}.md so N can exceed 7. Header copy and the
references catalog now both state 7 core + 2 advanced and the 1..9
filename range.
A3 (jobs): default panel's jobs-perspective treats Steve Jobs Archive (2022) and *Make Something Wonderful* (2023) as highest-weight sources, but those are post-mortem IP curated by Laurene Powell Jobs / Tim Cook / Jony Ive. Audits of Apple-family products or Archive offerings now declare a structural conflict and recommend --panel-drop jobs. A4 (leijun): perspective had a "Xiaomi Conflict Rule" section but the panels/auto.yaml self-conflict declaration uses the name "Self-Conflict Rule" and lists a broader brand set (Xiaomi / Redmi / Xiaomi Auto / SU7 / 玄戒 / Shunwei portfolio). Rename + align brand list. Both sections follow the pattern in zhangmingzheng-perspective L111-118.
A5: three scripts used to swallow failures, hiding real problems. * download_subtitles.sh: yt-dlp stderr was unconditionally piped to /dev/null and "no subtitle found" was indistinguishable from "network/cookie/region failure". Added a run_ytdlp wrapper that captures stderr to a temp file, prints a [warn] on non-zero exit, and re-emits stderr when DEBUG=1. * sync_mba_desc.sh: inline JSON parsing assumed the BotLearn API response shape; if it changed, the script silently 500-ed. Wrap json.JSONDecoder.raw_decode in try/except, add a defensive guard around obj['data']['skill'] field access, and route diagnostics to stderr. * wuying/smoke_test.py: cleanup-failed branch printed to stderr but still returned exit 0, so a leaked (paid) browser session looked like success. Track cleanup_ok and return exit 3 when teardown fails so CI / babysitters notice.
A6: grow_metric_brand_auditor.sh OUT_DIR defaulted to /private/tmp
which is macOS-specific. Use ${TMPDIR:-/tmp} so the growth helper
also runs on Linux CI.
A2: published/reports/lenovo/report.html referenced _raw/synthesis.md,
but the report directory ships only report.html / report.pdf / _assets/
— _raw/ is gitignored (.gitignore L18: reports/ excluded). The link
was a permanent 404 for anyone reading the published report.
Replace the dangling pointer with a self-contained note that the full
100+ source index lives in the internal report-generation tree and is
not part of the public report.
Also: __pycache__ / *.pyc were not gitignored; py_compile or pytest
runs were leaving cache dirs untracked. Add to .gitignore.
B0 + B4: codify the H2 layout every <slug>-perspective/SKILL.md must have. Three patterns had drifted across the 15 perspectives — default panel (zh, no Anti-Fab section), auto panel (mixed, no Honest Boundary section), security panel (full English, the target shape). The spec accepts Chinese / English aliases so existing perspectives don't have to be retranslated. Required logical sections (any listed alias passes): - Core Mental Models : ## Core Mental Models | ## 核心心智模型 - Honest Boundary : ## Honest Boundary | ## 诚实边界 - Anti-Fabrication : ## Anti-Fabrication Red Lines / 红线 / Guard - Self-Conflict Rule : ## Self-Conflict Rule - Sources : ## Sources | ## 附录：调研来源 check_structure.py walks all <slug>-perspective/ dirs and reports which logical sections are missing per perspective. CI now runs it alongside validate_panels.py — both block on PR. No-deps regex parser, same pattern as scripts/validate_panels.py.
B3 batch 1: the 5 default-panel perspectives had fabrication rules inline in 角色扮演规则 but no formal H2 — check_structure.py FAIL'd them. Add the spec-required sections so the validator passes and future automated audit has a stable hook. - fusheng : Anti-Fab + Self-Conflict (猎豹/OpenClaw/猎户星空/紫牛) - jobs : Anti-Fab (Self-Conflict was added in A3) - likejia : Anti-Fab + Self-Conflict (BotLearn/Aibrary/字节智慧教育) - wu-jundong : Anti-Fab + Self-Conflict (Aibrary/BotLearn/TAL/极课大数据) - zhang-yiming: Anti-Fab + Self-Conflict (ByteDance全家 + 字节系) Conflict brand families derived from each judge's research/01-06.md + public reporting; Self-Conflict block follows the zhangmingzheng L111-118 template (--panel-drop + founder-canon-self-check fallback).
B3 batch 2: the 5 auto-panel perspectives had Self-Conflict + Anti-Fabrication sections but no formal Honest Boundary H2 — limits were folded into Anti-Fab. Split them so the validator can find both sections separately, and add a Sources appendix pointing at references/research/01-06.md + quotes.md. - leijun : zh Honest Boundary + Sources (11 yrs Xiaomi annual + Shunwei) - musk : en Honest Boundary + Sources (Isaacson + X + Rogan/Lex) - lixiang : zh Honest Boundary + Sources (微博 6800+ + MEGA letter) - hexiaopeng : zh Honest Boundary + Sources (财报会 + 知乎 + 王凤英) - libin : zh Honest Boundary + Sources (NIO Day + NIO House) Honest Boundary content: research cutoff, register split between stage and informal, missing private-decision detail, peer-comparison rules. Source attribution: primary (judge's own production) vs secondary (media triangulation) classification reused from existing research file headers — no new fact claims. All 15 perspectives now pass check_structure.py.
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 join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
Summary
三路并行审计（编排层 / 评委 perspective 层 / 工具脚本 + CI）后的清理批次：7 个原子 commit，Phase A 修 7 类 bug，Phase B 把 15 位 perspective 的 SKILL.md H2 段统一到一份新的可执行 spec。无新 perspective、无 panel 扩张。
Phase A — Bug fixes
8e30881—${MAC_MBA_ROOT}路径符号修复 wuying-browser.md 三处硬编码~/mba/；SKILL.md / dimensions.md 把 "Default Dimensions (7)" 澄清为 "7 core + 2 advanced" 并标 1..9 filename 范围2593607—jobs-perspective加 Self-Conflict Rule（Steve Jobs Archive 是 post-mortem IP）；leijun-perspective把## Xiaomi Conflict Rule重命名为## Self-Conflict Rule并扩列 SU7 / YU7 / 玄戒 / 顺为08ae426— 3 个脚本不再静默吞掉失败：download_subtitles.sh加 DEBUG=1 透传 stderr；sync_mba_desc.sh加json.JSONDecodeError+ 防御性 dict 访问；wuying/smoke_test.pycleanup 失败时返回 exit 3 防止 leak 付费 session224f402—published/reports/lenovo/report.html删除指向不存在的_raw/synthesis.md的死链；grow_metric_brand_auditor.sh把 macOS-only/private/tmp改成${TMPDIR:-/tmp}；__pycache__/*.pyc加进 .gitignorePhase B — Perspective structure consistency
3b6b6f6— 新增metric-brand-auditor/references/perspective-structure-spec.md（5 段 H2 spec，接受 zh / en alias）+scripts/perspective-tools/check_structure.py（无依赖 regex 校验器）+ CI 加Validate perspective SKILL.md structurestep66cbb7d— fusheng / jobs / likejia / wu-jundong / zhang-yiming 补 Anti-Fabrication + Self-Conflict 段（之前规则只 inline 在角色扮演规则里）9349046— leijun / musk / lixiang / hexiaopeng / libin 补 Honest Boundary + Sources 段（之前融在 Anti-Fab 内）校验前后：
check_structure.py从5/15 pass→15/15 pass。Test plan
python3 scripts/validate_panels.py— 所有 panel + skeleton OKpython3 scripts/perspective-tools/check_structure.py— 15/15 perspectives passpython3 -m py_compile所有 Python（排除 skills/ + reports/）通过bash -n所有 shell 脚本通过bash site/build.sh→published=2 missing=0（lenovo + chengshi-auto）~/mba已替换为${MAC_MBA_ROOT:-~/mba}_raw/synthesis.md在 lenovo report.html 中已不再出现panel-validation.yml已加入新 stepPanel validationworkflow 跑绿/fusheng-perspective "评价一下 OpenClaw"对照新 Anti-Fab 段不应捏造（merge 后人工 dogfood）/mba lenovo --quick --refresh跑 default panel 一遍，5 份reviews/*.md都按新结构产出（merge 后人工 dogfood）不在本 PR 范围（保留供下一轮）
/mba --dry-runTODO 实现（docs/08-extending.md L168）Plan 文件存档：
/root/.claude/plans/mba-graceful-sloth.md。Generated by Claude Code