Conciseness

The body is mostly efficient (tables, terse bullets, no explanation of concepts Claude already knows), but the F198 case is retold four times — in the epigraph ('两只大猫花 3 小时...'), as blockquote asides in Steps 2, 3, and 4, and again in a full '教学案例' section — and it carries time-sensitive specifics (spike date '2026-05-13') outside any deprecated/old-patterns section. This fits the 3 anchor (mostly efficient but could be tightened) rather than 4, where only minor trimming would be needed.

Actionability

Concrete, executable guidance dominates: a tool-priority ladder starting with '`strings <binary>` / 反编译', a copy-paste help-request format ('我卡在 X，试过 A/B/C，你能从 Y 角度看看吗？'), and the six-question Desperation Packet checklist. It is not 5 because portions of Step 5 and the rhetoric discussion stay abstract ('代码 > 文档 > 博客 > 推测', '拒绝投降') without an equally concrete procedure; it is above 3 since the core actions are directly executable rather than pseudocode-level.

Workflow Clarity

The five-step framework is explicitly sequenced (Step 1–5) with decision gates and a feedback loop: '任一 yes → 执行 Step 2-5', '卡 3 轮 → 必须喊人（硬触发器）', and the gate '六问全答 → 提交 operator 决策。有空白 → 回去补' — an explicit validate-and-retry checkpoint. No destructive/batch operations apply, so no cap; this matches the 5 anchor's checklists and error-recovery loop.

Progressive Disclosure

The skill has no bundle files and is a single well-sectioned document with clear headers, two lookup tables, and cross-skill references clearly labeled in a boundary table. It is not 5: at ~120 lines it exceeds the small self-contained exception, and the full F198 teaching case (plus its repeated asides) is inline content that would fit a one-level-deep reference file. It is clearly above 3 — structure is good and nothing is buried or nested.

Specificity

The description names the domain ('绝境反转方法论：...五步突破框架') and concrete deliverables ('Desperation Packet（六问证据评估）+ 新方向行动计划 / operator 升级请求'), which is several specific actions with minor gaps — it never enumerates what the five steps actually are, so it falls between the 3 anchor (1-2 concrete actions) and the 5 anchor (comprehensive coverage).

Completeness

Both 'what' and 'when' are answered explicitly and concretely: what ('五步突破框架' plus explicit Output artifacts) and when ('Use when: 任务看起来无解、多次尝试失败后想放弃...连续 3+ 轮在希望与绝望间摆动' with concrete trigger phrases). It exceeds the 5 anchor by also adding 'Not for:' exclusions; it is not the 4 anchor because 'when' is fully explicit, not merely present.

Trigger Term Quality

Comprehensive natural-language trigger coverage: the frontmatter lists 15 phrases users/models would actually emit ('没希望', '没救了', '无解', '不可能', '接受现实', '放弃', '现状最优', 'vision rescue') and the description itself quotes surrender rhetoric inline ('现状最优'/'没救了'/'接受现实'). This matches the 5 anchor's comprehensive synonym coverage; nothing natural appears missing.

Distinctiveness Conflict Risk

It carves a clear niche (desperation/oscillation signals on a vision goal) and explicitly disambiguates the nearest neighbors: 'Not for: 未出现绝境信号的常规 debugging（用 debugging）、常规探索/调研（用 deep-research）、operator 已签字降级的目标、trivial 任务'. Minimal conflict risk; not the 4 anchor since overlap is addressed head-on rather than merely minor.