Conciseness

The body is dense and instructional with no basic-concept explanations, but §0 '为什么必须有协议' spends lines on motivation with time-sensitive statistics ('GPT-3.5 虚构引用率 55%,即使 GPT-4 仍达 18%(Walters & Wilder, Scientific Reports 2023)') that do not guide execution and sit outside any deprecated/old-patterns section. That trimmable padding matches 'minor instances of over-explanation' (4) rather than the every-token-earns-its-place level 5.

Actionability

For an instruction-only skill the guidance is fully concrete: explicit thresholds ('至少 2~3 个独立来源', '1 个主问题 + 3~5 个研究子问题'), named tools (arXiv/PubMed/Crossref, 官方披露/财报/公告/监管), a copy-paste report template with naming convention 'YYYY-MM-DD-主题.md', a YAML frontmatter spec, and per-item citation statuses '✅已核验 / ⚠️未找到 / ❌疑似虚构'. No gaps found versus the 'mostly executable, minor gaps' level 4.

Workflow Clarity

The five stages are explicitly sequenced (阶段一…阶段五), stage 4 is a dedicated validation stage ('引用核验…输出逐条状态', '不一致时并列呈现分歧,不强行统一'), and §4 is a pre-delivery quality-gate checklist ('成稿前逐项自检') plus a hard rule '禁止输出未核验的…引用'. This matches clear sequence + explicit validation + error-handling + checklist; it is not a destructive or batch operation, so no cap applies.

Progressive Disclosure

The single file is well-sectioned (§0–§4, tables, code-block templates) with no buried or nested references, and no bundle files exist to mis-navigate. It does not reach 5 because at ~90 lines the inlined report template (§2) and knowledge-base spec (§3) could be split into one-level-deep reference files to keep SKILL.md a leaner overview, and the under-50-line simple-skill exception does not apply.

Specificity

Lists seven concrete pipeline actions — '问题拆解、多源检索、证据分级、交叉验证、引用核验、结构化成稿与知识库沉淀' — plus named domains '(学科、金融、知识探索)' and a purpose ('用于保证研究质量、降低幻觉风险'), comprehensively covering the research lifecycle. Not 4 because action coverage of the workflow is complete rather than having minor gaps.

Completeness

The 'what' is explicit (the seven-stage pipeline and its purpose) and the 'when' is present as equivalent explicit trigger guidance ('任何领域的研究任务(学科、金融、知识探索)都应当遵循本技能'), so the score-3 cap for a missing 'Use when' clause does not apply. Not 5 because the when-clause states scope rather than concrete user-sayable trigger phrases.

Trigger Term Quality

Includes the core term '研究任务' plus domain-level triggers '学科、金融、知识探索' and stage terms like '检索' and '引用核验', but misses common user phrasings such as '调研', '分析', '综述', or '了解某主题' (those appear only in the body's trigger note, not the description). This fits 'good keyword coverage; a few natural terms missing' (4) rather than 3, since coverage goes well beyond a single keyword like 'Works with PDF files'.

Distinctiveness Conflict Risk

By claiming '任何领域的研究任务' it positions itself as a universal default for all research, creating overlap risk with any domain-specific research, finance-analysis, or literature-review skill. The distinctive mechanisms ('证据分级', '引用核验', '知识库沉淀') keep it above the 'very broad' level 2, but the any-domain claim is more than minor overlap risk, so it does not reach 4.