{"id":"41cdcd73-0705-40aa-bce8-3cfd13782f6b","entityType":"agent","slug":"clawhub-medstatstar-ct-literature","name":"Clinical Trial Literature Search / 临床试验文献检索专家","canonicalUrl":"https://www.xpersona.co/agent/clawhub-medstatstar-ct-literature","canonicalPath":"/agent/clawhub-medstatstar-ct-literature","generatedAt":"2026-10-11T03:56:21.171Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"editorial-content","verified":true,"confidence":"high","updatedAt":"2026-10-11T00:37:58.812Z","emptyReason":null},"description":"Search medical-domain scholarly literature with full data-source coverage (OpenAlex + Europe PMC/MeSH + bioRxiv/medRxiv preprints + arXiv methodology breadth), normalize, merge, and de-duplicate the results, produce a qualitative CSM safety-literature subset, and assist in providing open-access (OA) full-text PDF downloads. Usable without a key. / 全数据源覆盖检索医学领域学术文献（OpenAlex + Europe PMC/MeSH + bioRxiv/medRxiv 预印本 + arXiv 方法学广度），归一化合并去重，可产出 CSM 安全性定性子集，协助提供OA文献PDF下载。无key亦可使用。 Skill: Clinical Trial Literature Search / 临床试验文献检索专家 Owner: medstatstar Summary: Search medical-domain scholarly literature with full data-source coverage (OpenAlex + Europe PMC/MeSH + bioRxiv/medRxiv preprints + arXiv methodology breadth), normalize, merge, and de-duplicate the results, produce a qualitative CSM safety-literature subset, and assist in providing open-access (OA) full-text PDF downloads. Usable withou","descriptionLabel":"Technical summary","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 1.2K downloads reported by the source. Last updated 10/11/2026.","installCommand":"clawhub skill install s176fv8983h1rte6dmxwp9wt4n89j8p5:ct-literature","sourceUrl":"https://clawhub.ai/medstatstar/ct-literature","homepage":"https://clawhub.ai/medstatstar/skills/ct-literature","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/medstatstar/ct-literature","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/medstatstar/skills/ct-literature","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":62,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"Search medical-domain scholarly literature with full data-source coverage (OpenAlex + Europe PMC/MeSH + bioRxiv/medRxiv preprints + arXiv methodology breadth), "},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-11T00:37:58.812Z","emptyReason":null},"protocols":[{"protocol":"OPENCLEW","label":"OpenClaw","status":"self-declared","notes":"Declared in the public agent profile."}],"capabilities":[],"verifiedCount":0,"selfDeclaredCount":1,"capabilityMatrix":{"rows":[{"key":"OPENCLEW","type":"protocol","support":"unknown","confidenceSource":"profile","notes":"Listed on profile"}],"flattenedTokens":"protocol:OPENCLEW|unknown|profile"}},"adoption":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-11T00:37:58.812Z","emptyReason":null},"stars":null,"forks":null,"downloads":1217,"packageName":null,"latestVersion":"1.0.1","tractionLabel":"1.2K downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-11T00:37:58.796Z","emptyReason":null},"lastUpdatedAt":"2026-10-11T00:37:58.812Z","lastCrawledAt":"2026-10-11T00:37:58.796Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-12T00:37:58.796Z","lastVerifiedAt":null,"highlights":[{"version":"1.0.1","createdAt":"2026-10-02T03:01:40.254Z","changelog":"v1.2.0: Enable snowball+report-first by default; add few-results suggestion card; i18n fix; F19/F13/shared_sync compliance","fileCount":78,"zipByteSize":855698},{"version":"1.0.0","createdAt":"2026-09-07T07:53:26.336Z","changelog":"v1.0.0 full-feature stable release: OA PDF batch download chain; §16 pre-publish compliance cleanup; bilingual README refresh","fileCount":66,"zipByteSize":419448},{"version":"0.9.5","createdAt":"2026-08-26T06:25:18.402Z","changelog":"Release v0.9.5: unify version across SKILL.md frontmatter / dual READMEs / CHANGELOG (single source of truth); compress SKILL.md to <=200 lines; trim redundant Natural language dialogue section; publish GitHub + ClawHub (SkillHub already live)","fileCount":60,"zipByteSize":312576},{"version":"0.9.0","createdAt":"2026-08-22T10:04:44.493Z","changelog":"Add bug report feature: 11-key sanitized envelope via ct-bugreport endpoint with two-stage user confirmation and local fallback","fileCount":60,"zipByteSize":304394},{"version":"0.6.15","createdAt":"2026-08-14T09:54:43.033Z","changelog":"Major retrieval-performance optimization (pooled HTTP, verify dedup, 3x wider pool, two-phase delivery; benchmarked ~3.4x faster) + search-keyword auto-translation (Chinese→English, offline dictionaries ~900 entries, synonym OR, user-extendable) + progress event stream (--progress json). Clean package: tests/ excluded.","fileCount":54,"zipByteSize":251572},{"version":"0.6.14","createdAt":"2026-08-13T05:22:22.016Z","changelog":"Security-audit fixes (SkillSpector): unify API-key setup to conversational dual-path; drop R-only dead code; drug_name_resolver unique-candidate auto-mode; CLI anti-hallucination warnings; README activation boundary. Clean package: tests/ excluded.","fileCount":53,"zipByteSize":238397},{"version":"0.6.13","createdAt":"2026-08-13T05:18:45.002Z","changelog":"Security-audit fixes (SkillSpector): unify API-key setup to conversational dual-path; drop R-only dead code; drug_name_resolver unique-candidate auto-mode; CLI anti-hallucination warnings; README activation boundary. Clean package: tests/ excluded.","fileCount":53,"zipByteSize":238297},{"version":"0.6.12","createdAt":"2026-08-12T14:53:08.099Z","changelog":"引文验证新增标题/作者一致性深度校验（v0.6.11）：标识解析成功后拉 Crossref/EuropePMC/OpenAlex 元数据比对标题+第一作者，新增 mismatch 状态与 --no-consistency 开关；含 v0.6.10 系统逻辑审计与 README 重构。","fileCount":61,"zipByteSize":268413}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s176fv8983h1rte6dmxwp9wt4n89j8p5:ct-literature","setupComplexity":"low","setupSteps":["Setup complexity is classified as HIGH. You must provision dedicated cloud infrastructure or an isolated VM. Do not run this directly on your local workstation.","Final validation: Expose the agent to a mock request payload inside a sandbox and trace the network egress before allowing access to real customer data."],"contract":{"contractStatus":"missing","authModes":[],"requires":[],"forbidden":[],"supportsMcp":false,"supportsA2a":false,"supportsStreaming":false,"inputSchemaRef":null,"outputSchemaRef":null,"dataRegion":null,"contractUpdatedAt":null,"sourceUpdatedAt":null,"freshnessSeconds":null},"invocationGuide":{"preferredApi":{"snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-medstatstar-ct-literature/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-medstatstar-ct-literature/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-medstatstar-ct-literature/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-medstatstar-ct-literature/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-medstatstar-ct-literature/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-medstatstar-ct-literature/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-11T03:56:21.167Z"}},"retryPolicy":{"maxAttempts":3,"backoffMs":[500,1500,3500],"retryableConditions":["HTTP_429","HTTP_503","NETWORK_TIMEOUT"]}},"endpoints":{"dossierUrl":"https://www.xpersona.co/api/v1/agents/clawhub-medstatstar-ct-literature/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-medstatstar-ct-literature/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-medstatstar-ct-literature/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-medstatstar-ct-literature/trust"}},"reliability":{"evidence":{"source":"runtime-metrics","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No trust, reliability, or runtime telemetry is available."},"trust":{"status":"unavailable","handshakeStatus":"UNKNOWN","verificationFreshnessHours":null,"reputationScore":null,"p95LatencyMs":null,"successRate30d":null,"fallbackRate":null,"attempts30d":null,"trustUpdatedAt":null,"trustConfidence":"unknown","sourceUpdatedAt":null,"freshnessSeconds":null},"decisionGuardrails":{"doNotUseIf":["Contract metadata is missing or unavailable for deterministic execution."],"safeUseWhen":[],"riskFlags":["missing_or_unavailable_contract","trust_data_unavailable","schema_references_missing"],"operationalConfidence":"low"},"executionMetrics":{"observedLatencyMsP50":null,"observedLatencyMsP95":null,"estimatedCostUsd":null,"uptime30d":null,"rateLimitRpm":null,"rateLimitBurst":null,"lastVerifiedAt":null,"verificationSource":null},"runtimeMetrics":{"successRate":null,"avgLatencyMs":null,"avgCostUsd":null,"hallucinationRate":null,"retryRate":null,"disputeRate":null,"p50Latency":null,"p95Latency":null,"lastUpdated":null}},"benchmarks":{"evidence":{"source":"no-benchmark-data","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No benchmark suites or observed failure patterns are available."},"suites":[],"failurePatterns":[]},"artifacts":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"high","updatedAt":"2026-10-11T00:37:58.812Z","emptyReason":null},"readme":"Skill: Clinical Trial Literature Search / 临床试验文献检索专家\n\nOwner: medstatstar\n\nSummary: Search medical-domain scholarly literature with full data-source coverage (OpenAlex + Europe PMC/MeSH + bioRxiv/medRxiv preprints + arXiv methodology breadth), normalize, merge, and de-duplicate the results, produce a qualitative CSM safety-literature subset, and assist in providing open-access (OA) full-text PDF downloads. Usable without a key. / 全数据源覆盖检索医学领域学术文献（OpenAlex + Europe PMC/MeSH + bioRxiv/medRxiv 预印本 + arXiv 方法学广度），归一化合并去重，可产出 CSM 安全性定性子集，协助提供OA文献PDF下载。无key亦可使用。\n\nTags: latest:1.0.1\n\nVersion history:\n\nv1.0.1 | 2026-10-02T03:01:40.254Z | user\n\nv1.2.0: Enable snowball+report-first by default; add few-results suggestion card; i18n fix; F19/F13/shared_sync compliance\n\nv1.0.0 | 2026-09-07T07:53:26.336Z | user\n\nv1.0.0 full-feature stable release: OA PDF batch download chain; §16 pre-publish compliance cleanup; bilingual README refresh\n\nv0.9.5 | 2026-08-26T06:25:18.402Z | user\n\nRelease v0.9.5: unify version across SKILL.md frontmatter / dual READMEs / CHANGELOG (single source of truth); compress SKILL.md to <=200 lines; trim redundant Natural language dialogue section; publish GitHub + ClawHub (SkillHub already live)\n\nv0.9.0 | 2026-08-22T10:04:44.493Z | user\n\nAdd bug report feature: 11-key sanitized envelope via ct-bugreport endpoint with two-stage user confirmation and local fallback\n\nv0.6.15 | 2026-08-14T09:54:43.033Z | user\n\nMajor retrieval-performance optimization (pooled HTTP, verify dedup, 3x wider pool, two-phase delivery; benchmarked ~3.4x faster) + search-keyword auto-translation (Chinese→English, offline dictionaries ~900 entries, synonym OR, user-extendable) + progress event stream (--progress json). Clean package: tests/ excluded.\n\nv0.6.14 | 2026-08-13T05:22:22.016Z | user\n\nSecurity-audit fixes (SkillSpector): unify API-key setup to conversational dual-path; drop R-only dead code; drug_name_resolver unique-candidate auto-mode; CLI anti-hallucination warnings; README activation boundary. Clean package: tests/ excluded.\n\nv0.6.13 | 2026-08-13T05:18:45.002Z | user\n\nSecurity-audit fixes (SkillSpector): unify API-key setup to conversational dual-path; drop R-only dead code; drug_name_resolver unique-candidate auto-mode; CLI anti-hallucination warnings; README activation boundary. Clean package: tests/ excluded.\n\nv0.6.12 | 2026-08-12T14:53:08.099Z | user\n\n引文验证新增标题/作者一致性深度校验（v0.6.11）：标识解析成功后拉 Crossref/EuropePMC/OpenAlex 元数据比对标题+第一作者，新增 mismatch 状态与 --no-consistency 开关；含 v0.6.10 系统逻辑审计与 README 重构。\n\nv0.6.11 | 2026-08-12T14:46:58.917Z | user\n\n引文验证新增标题/作者一致性深度校验（v0.6.11）：标识解析成功后拉 Crossref/EuropePMC/OpenAlex 元数据比对标题+第一作者，新增 mismatch 状态与 --no-consistency 开关；含 v0.6.10 系统逻辑审计与 README 重构。\n\nv0.6.0 | 2026-08-12T07:42:19.327Z | user\n\nv0.6.0: P0 citation verification + evidence provenance log; P1 PROSPERO reserved source (key-gated); outbound calls consolidated into adapters/; Open Access link column; SKILL.md <=200 lines\n\nv0.5.4 | 2026-08-11T03:15:41.740Z | user\n\nSelf-contained: inject ct-base shared assets, remove sibling fallback\n\nv0.5.3 | 2026-08-08T08:02:20.046Z | user\n\nAdd .env key XOR+base64 obfuscation (Ssd3 hardening); bump v0.5.3\n\nv0.5.2 | 2026-08-08T07:45:38.505Z | user\n\nRemove API-key-in-chat guidance (Ssd3); drop unused requests dependency; align v0.5.2\n\nv0.5.1 | 2026-08-08T06:49:34.599Z | user\n\nsecurity audit remediation: removed key-paste-to-assistant guidance, removed arbitrary R execution primitive (r_libs.run_r)\n\nv0.5.0 | 2026-08-08T05:37:00.981Z | user\n\ninit version\n\nArchive index:\n\nArchive v1.0.1: 78 files, 855698 bytes\n\nFiles: adapters/__init__.py (270b), adapters/bug_report.py (21369b), adapters/build_guidelines.py (12663b), adapters/coze_resolve.py (50906b), adapters/fetch_arxiv.py (10534b), adapters/fetch_coze_unified.py (33845b), adapters/fetch_europepmc.py (11958b), adapters/fetch_guidelines.py (21282b), adapters/fetch_openalex.py (18422b), adapters/fetch_preprints.py (17306b), adapters/fetch_prospero.py (9554b), adapters/fetch_semantic_scholar.py (8385b), adapters/guideline_corpus.py (5199b), adapters/http_utils.py (21902b), adapters/pdf_download.py (80345b), adapters/portal_fetch.py (7069b), adapters/preprint_fallback.py (13392b), adapters/verify_citations.py (28888b), AGENTS.md (2581b), assets/_preview_240.png (16615b), assets/ct-lifecycle-diagram.png (247402b), assets/ct-lifecycle-diagram.svg (10356b), assets/icon_4x.png (30028b), assets/icon_8x.png (60690b), assets/icon.svg (2417b), CHANGELOG.md (147699b), docs/prospero_access_guide.md (10922b), LICENSE (1089b), README_zh-CN.md (39147b), README.md (41786b), references/capabilities.md (18408b), references/citation_styles.md (2494b), references/dialogue.md (5558b), references/drug_name_map.json (23396b), references/errors.md (834b), references/guidelines/guidelines_index.json (102767b), references/language_policy.md (3810b), references/mesh_terms_mapping.json (7661b), references/multi-db-search.md (17603b), references/openalex_key.md (4544b), references/published_app.md (3963b), references/report_template.md (4246b), references/search_menu.md (18948b), references/sop.md (5607b), references/term_map.json (9096b), references/units.md (3538b), requirements.txt (534b), scripts/abstract_translator.py (10376b), scripts/ct_literature.py (86875b), scripts/doc_type_filter.py (17665b), scripts/drug_name_resolver.py (5168b), scripts/evidence_log.py (11243b), scripts/excel_style.py (24154b), scripts/export_html.py (43808b), scripts/export_xlsx.py (57355b), scripts/format_citations.py (13746b), scripts/hardware_id.py (4142b), scripts/i18n_messages.json (38477b), scripts/i18n.py (15685b), scripts/keyword_breadth.py (9756b), scripts/kw_lexicon.json (25723b), scripts/kw_localize.py (34167b), scripts/mesh_mapper.py (5941b), scripts/normalize.py (21645b), scripts/obsidian_exporter.py (10702b), scripts/rare_disease_sources.py (10463b), scripts/report.py (27972b), scripts/score_relevance.py (5578b), scripts/screen_prisma.py (10410b), scripts/source_guard.py (5915b), scripts/test_topic_online_fallback.py (5585b), scripts/topic_translator.py (12388b), scripts/update_xlsx_pdf_paths.py (5619b), scripts/verify_dois.py (4316b), scripts/zotero_exporter.py (4654b), skill-card.md (2226b), SKILL.md (13683b), _meta.json (132b)\n\nFile v1.0.1:SKILL.md\n\n---\r\nslug: ct-literature\r\nname: ct-literature\r\ndisplayName: Clinical Trial Literature Search / 临床试验文献检索专家\r\ncn_name: 临床试验文献检索专家\r\nversion: 1.2.0\r\ninvocable: true\r\nsummary: 全数据源覆盖检索医学领域学术文献（OpenAlex + Europe PMC/MeSH + bioRxiv/medRxiv 预印本 + arXiv 方法学广度），归一化合并去重，可产出 CSM 安全性定性子集，协助提供OA文献PDF下载。无key亦可使用。\r\nlicense: MIT\r\ndescription: \"Search medical-domain scholarly literature with full data-source coverage (OpenAlex + Europe PMC/MeSH + bioRxiv/medRxiv preprints + arXiv methodology breadth), normalize, merge, and de-duplicate the results, produce a qualitative CSM safety-literature subset, and assist in providing open-access (OA) full-text PDF downloads. Usable without a key. / 全数据源覆盖检索医学领域学术文献（OpenAlex + Europe PMC/MeSH + bioRxiv/medRxiv 预印本 + arXiv 方法学广度），归一化合并去重，可产出 CSM 安全性定性子集，协助提供OA文献PDF下载。无key亦可使用。\"\r\ntriggers:\r\n - \"systematic literature search\"\r\n - \"系统文献检索\"\r\n - \"文献证据基础\"\r\n - \"已发表安全性文献 / CSM\"\r\n - \"cross-database literature search\"\r\n - \"跨数据库 文献检索\"\r\n - \"Embase Cochrane Web of Science\"\r\n - \"多数据库 系统综述\"\r\n - \"ct-literature\"\r\nrequired_commands: [python]\r\nmetadata:\r\n openclaw: { emoji: \"📚\" }\r\n authors: [\"medstatstar\", \"phoe-zip\"]\r\n tags: [clinical-trial, literature, evidence, systematic-review, csm, openalex, pubmed, public-data]\r\n homepage: \"https://github.com/medstatstar/ct-literature\"\r\npermissions:\r\n scope: \"user-space-only\"\r\n network: \"optional\"\r\n network_note: \"Reads only public bibliographic sources: OpenAlex (api.openalex.org, no key), Europe PMC (ebi.ac.uk, MEDLINE/MeSH, no key; also indexes bioRxiv/medRxiv preprints via SRC:PPR), Semantic Scholar (api.semanticscholar.org, no key; rate-limited HTTP 429 -> gracefully skipped), arXiv (export.arxiv.org/api/query, no key). Europe PMC is ON by default (--no-with-europepmc to disable); bioRxiv/medRxiv are ON by default (--no-with-biorxiv / --no-with-medrxiv to disable); arXiv is opt-in via --with-arxiv. No WAF, no confidential input; ordinary input + public retrieval (A-tier). Opt-in, user-confirmed bug reports additionally reach https://ct-bugreport.coze.site/run with an 11-key sanitized envelope only (never raw data).\"\r\n filesystem: \"read-only to its own files; writes report files only to the current working directory\"\r\n data: \"no confidential data input; no external transmission of user data\"\r\n\r\n---\r\n\r\n## Published Application\r\n\r\nThe workbench is published as a Python HTTP service at `https://ct-literature.app.workbuddy.host/`\r\n(appId `wbapp_WuOPnJ4caJ32GTquNlR7vc`; bundle `workbench/publish-dist/`, workspace-independent,\r\nre-publish via `workbench/publish-kit/install_genie.py` — **overwrite, never `createNewApp`**;\r\nmis-bound paths are permanently retired and the returned `shareLink` must always be asserted).\r\n**Full registry table + re-publishing protocol → [references/published_app.md](references/published_app.md)** — load it before any deploy / re-publish operation.\r\n\r\n## Language\r\n\r\n- **English guide** → [README.md](https://github.com/medstatstar/ct-literature/blob/main/README.md) · **中文指南** → [README_zh-CN.md](https://github.com/medstatstar/ct-literature/blob/main/README_zh-CN.md)\r\n- Bilingual auto-switch: the answer language follows the user's question language (English question → English answer, Chinese question → Chinese answer).\r\n\r\n## Purpose\r\n\r\nRetrieve **published scholarly literature** (peer-reviewed papers, systematic reviews, conference abstracts, preprints) about a drug / disease / method, normalize heterogeneous records from multiple public bibliographic sources into one de-duplicated evidence base, and surface the evidence landscape plus a **CSM (cumulative safety monitoring)** qualitative subset. Supports trial-planning background, protocol / CSR introductions, and published-safety literature checks.\r\n\r\n## Positioning within the ct- library\r\n\r\nFour complementary A-tier public-intel skills: `ct-registry` (registered trials) · **`ct-literature` (published evidence)** · `ct-safety` (FAERS signals) · `ct-pipeline` (aggregate intel brief).\r\n**Boundaries:** `ct-literature --safety` is *qualitative* published-safety literature — never a substitute for `ct-safety`'s FAERS disproportionality. Unsure which skill? Route via `ct-advisor` (§15). Full four-skill matrix + boundary detail → [references/capabilities.md](references/capabilities.md).\r\n\r\n## Data Sources\r\n\r\nDefault pipeline: **OpenAlex (primary) + Europe PMC (default ON) + bioRxiv/medRxiv (default ON)**.\r\nOpt-in: arXiv (`--with-arxiv`), Cochrane CDSR via Europe PMC journal filter (`--cochrane`), Semantic Scholar (`--with-semantic-scholar`, key needed), PROSPERO (`--with-prospero` — **reserved source**, dormant no-op without token, never claimed functional). All public bibliographic APIs, no WAF; keyless OpenAlex is capped 100/day since 2026-02-13 — a free key lifts to 100k/day (`.env` auto-load, key never printed).\r\nPer-source access/status/role table + PROSPERO caveat → [references/capabilities.md](references/capabilities.md).\r\n\r\n## Clinical guideline sources (`--with-guidelines`, opt-in · LOCAL corpus)\r\n\r\nVersion-pinned, **pointer-only** corpus (`references/guidelines/guidelines_index.json`) — build once (network, `adapters/build_guidelines.py`), read many (zero network). Full-text documents are NEVER shipped in the skill (author Coze KB / opt-in external cache). Corpus is honest by construction: failed fetch degrades to a `retrieved:false` pointer, never a fabricated citation.\r\nTier matrix, data-protection split, audit notes → [references/capabilities.md](references/capabilities.md).\r\n\r\n## Features\r\n\r\nCore: topic/drug/disease search · review-type & year filters · safety/CSM tagging · multi-source merge+dedupe · citation ranking · MeSH/concepts/funders · structured output (JSON/MD/Excel) · retry+backoff · safe link rendering · citations BibTeX/RIS · PRISMA funnel · doc-type exclusion (`--original-only` / `--only-type`, bidirectional) · relevance scoring · Obsidian/Zotero.\r\nAnti-hallucination: **P0 citation verification** (`--verify`, §17.1) · **P1 DOI audit** (`--verify-dois`) · **P0 evidence provenance log**.\r\nFull capability table (per-feature flags & semantics) → [references/capabilities.md](references/capabilities.md).\r\n\r\n### Retrieval structure model (2026-09-30, adapted from the 组小学 expert · P0/P1)\r\n\r\nPorted 组小学's (tencent-lit-search) auditable retrieval structure to ct-literature — all **additive, backward-compatible** (no new flag = identical old behavior):\r\n\r\n- **Structured PICO concepts `--concept TYPE=VALUE`** (repeatable): `TYPE∈{condition,population,intervention,comparator,outcome,endpoint,study_design,biomarker,drug,indication}`. Concepts are AND-combined into the query, with intervention / comparator / outcome on separate axes (no longer all crammed into `--topic`); each concept is MeSH-normalized and its `mapping_status` recorded (`exact_label` / `synonym` / `unmapped_literal`; fuzzy matches are rejected and literal terms are never fabricated).\r\n- **Retrieval depth budget `--depth quick|standard|deep`**: per-source cap + whether to trigger weak-result rescue (`quick` never rescues; `standard`/`deep` do). Not an evidence-quality rating.\r\n- **Evidence-dimension lane**: each merged record gets `evidence_dimension` (= `safety` if `is_safety`, else `general`); `--safety` plans a dedicated safety lane so safety signals are not diluted by efficacy records. SAFE PREVIEW prints the **lane plan view** (purpose / source / evidence_dimension / notes) for pre-run review.\r\n- **Four-state audit ledger + coverage verdict**: evidence-log adds `status∈{ok,empty,error,skipped}` and `coverage∈{healthy,partial,empty,critical_gap}`; explicitly-disabled sources are logged as `skipped` (not_run); gaps are surfaced honestly, never zeroed; OpenAlex failure → `critical_gap` with a warning against extrapolation.\r\n- **Weak-result rescue (P1-6)**: when `standard`/`deup` and unique merged works < 5 and concepts had narrowed the query, automatically re-run with a broader query (topic only); single-shot, non-fatal; logs `skipped` if nothing new is added.\r\n\r\n> What is NOT ported: 组小学's 7-scenario exhaustive set / first-class trial·drug·target objects / target-gene-variation entities — these fall outside ct-literature's clinical-literature scope and would overlap with ct-registry / ct-safety, so they are deliberately excluded.\r\n\r\n## Unified work schema\r\n\r\nNormalizes all sources into one record shape (`source / id / title / authors / year / … / doi / mesh / concepts / is_safety / is_preprint / sources`). Field list → [references/capabilities.md](references/capabilities.md).\r\n\r\n## Output\r\n\r\n**Standard deliverables = HTML + Excel only** — present `lit_report.html`; `lit_report.xlsx` keeps the complete filterable result. On-demand add-ons: citations (`.bib`/`.ris`/styled md), OA-PDF download (`--download-pdf`, legal sources only; **never open/list individual PDFs after download — point users at `out_dir/pdfs/`**), Obsidian notes, Zotero exports. PDF pipeline runs local-resolve ‖ Coze-decode ‖ download in parallel (2026-09-19).\r\nFull output catalogue, sheet layout, server-side supplement chain, concurrency notes → [references/capabilities.md](references/capabilities.md); command catalogue → [references/sop.md](references/sop.md).\r\n\r\n## Requirements\r\n\r\n- Python 3.10+ (Anaconda `C:\\Tools\\anaconda3\\python.exe` recommended).\r\n- `requests` optional (fetch scripts use stdlib `urllib`); `matplotlib` optional (future trend charts).\r\n- Network: read-only public bibliographic APIs.\r\n\r\n## ⚠️ Safety\r\n\r\n- Default **SAFE PREVIEW**: scripts only generate / display; network requests run only with explicit `--run`.\r\n- Reads **public publications ONLY**, zero confidential research / subject data input (A-tier — non-confidential input, per §11; API keys are local config, never research data).\r\n- `--safety` literature is **qualitative** — never feed it into FAERS disproportionality; it only corroborates `ct-safety` qualitatively.\r\n- Output is for reference / background only, not a regulatory submission.\r\n\r\n## Implementation\r\n\r\nPrimary entry: `python scripts/ct_literature.py --topic \"\" [--review-type …] [--year-from …] [--safety] [--cochrane] [--sources \"OpenAlex,EuropePMC\"] [--with-guidelines] --run --out-dir ./out` (omit `--run` = SAFE PREVIEW). Guidelines corpus build: `python adapters/build_guidelines.py --topic --run`.\r\nInvocation cookbook (worked examples, `--sources` semantics, OpenAlex key setup) → [references/capabilities.md](references/capabilities.md).\r\n\r\n## Errors\r\n\r\nSee [references/errors.md](references/errors.md) for the full error catalogue (network / 429 / 401 / empty results / DOI dedupe).\r\n\r\n## Pipeline\r\n\r\n- `ct-registry` → `ct-literature`: landscape hypothesis seeds the literature search topic.\r\n- `ct-literature` → `ct-pipeline` (intel evidence dimension), → `ct-protocol` / `ct-csr` (background), `--safety` → `ct-safety` (qualitative corroboration).\r\n\r\n## Cross-Database Search Mode\r\n\r\nA cross-database planning layer (Embase / Cochrane / Web of Science + preprint Tier P, adapted from `multi-database-literature-collector`, AIPOCH MIT) builds search strategy; live fetch still runs the six sources. See [references/multi-db-search.md](references/multi-db-search.md). The **Cochrane** leg is now directly automatable via `--cochrane` (a verified Europe PMC journal filter — identical to meta-analysis's in-skill dedup probe), so CDSR needs no manual browser step.\r\n\r\n## Natural-Language Dialogue — Triage-First (mandatory)\r\n\r\nEvery NL turn MUST start at Step 0 triage — never skipped, never jump straight to `--run`\r\n(§6.2 + [references/search_menu.md](references/search_menu.md)). Four buckets: **Simple** (topic +\r\n≥2 params → §4.2 preview) · **Complex** (menu §4.1 → §12 keyword gate → §4.2 → run) · **Vague**\r\n(grill-me ≤2 rounds → re-triage) · **Middle** (answer directly).\r\n**Red lines:** never skip triage; never decide for the user (Complex menu keeps the \"explain the differences\" entry ③); the §12 keyword-system confirmation gate is mandatory whenever search terms are parseable.\r\n**Cross-turn continuity (high-risk, mode A):** query spec lives only in the thread — after every run append the `## 当前检索设定：topic=… | type=… | year=… | safety=… | sources=… | max=… | verify=…` echo block (never omit fields); follow-ups override only changed fields and run through `merge_spec.py` by default; remotes are stateless, so continuity is local-only.\r\nFull triage tables, echo/merge_spec rules, execution flow → [references/dialogue.md](references/dialogue.md) — load it when executing an NL dialogue round.\r\n\r\n## Bug Reporting (§20.3, adapter: `adapters/bug_report.py`)\r\n\r\nTrigger = explicit user request (unlimited) or strong signal + one retry (≤1 proposal/session).\r\nHard two-stage confirmation: ① show full sanitized report → ② send only on explicit consent\r\n(endpoint `https://ct-bugreport.coze.site/run`, public credential). 11-key whitelist envelope,\r\n`description` is the only user-reviewed free text; never raw data. Client sends `report` only —\r\ngovernance belongs to `ct-update`.\r\nFull trigger/two-stage/sanitization detail → [references/dialogue.md](references/dialogue.md).\n\nFile v1.0.1:README.md\n\n# Clinical Trial Literature Search (ct-literature)\r\n\r\n[🇨🇳 中文](./README_zh-CN.md) | [🇺🇸 English (Current)](#)\r\n\r\n
\r\n\"ct-literature\r\n
\r\n\r\n> **No install needed to try it:** if you don't want to install this skill and just want to quickly use its basic features, visit the web app directly at **https://ct.medstatstar.com**.\r\n\r\n> **A `ct-` library skill (A-tier public-intel — non-confidential input) that retrieves published scholarly literature about a drug / disease / method, normalizes multiple public bibliographic sources into one de-duplicated evidence base, and surfaces the evidence landscape plus a CSM (cumulative safety monitoring) qualitative subset.**\r\n\r\n> No commands or manual needed. Just describe your literature question **in plain language inside a chat** — the skill fetches from **OpenAlex (primary) + Europe PMC (on by default) + bioRxiv/medRxiv (on by default)**, then writes a self-contained **HTML + Excel** report. (Semantic Scholar and arXiv are opt-in via flags, not part of the default pipeline.) A-tier (non-confidential input): fully local computation, only public retrieval. **Note: your topic query is sent to the public bibliographic APIs below — see the [outbound notice](#outbound--privacy).** The skill activates **only when you explicitly ask for a literature search**; it never retrieves on its own during unrelated conversations.\r\n\r\n> 💡 **Keyless by default, but a free key lifts the cap a lot:** OpenAlex has required an API key since 2026-02-13; without one you are in the keyless pool (100 credits/day, flagged *not suitable for production*). A free key lifts this to 100k/day. Apply in ~30s — see the [First-Time FAQ](#first-time-faq) and the key-notice the skill prints automatically when no key is detected.\r\n\r\n## Table of Contents\r\n\r\n- [Who This Is For](#who-this-is-for)\r\n- [How to Use It in a Chat](#how-to-use-it-in-a-chat)\r\n- [Data Sources](#data-sources)\r\n- [Why You Can Trust the Output — Anti-Hallucination](#why-you-can-trust-the-output--anti-hallucination)\r\n- [What Can It Do — Scenarios](#what-can-it-do--scenarios)\r\n- [First-Time FAQ](#first-time-faq)\r\n- [Security & Privacy](#security--privacy)\r\n- [Advanced Reference (Developers)](#advanced-reference-developers)\r\n\r\n---\r\n\r\n## Who This Is For\r\n\r\nct-literature is part of the `ct-` clinical-trial skill family, built for three groups:\r\n\r\n- **Clinical-trial practitioners at pharmaceutical companies** — sponsors, CROs, and medical / statistical / regulatory roles;\r\n- **Clinicians and nurses who take part in the hands-on conduct of trials**;\r\n- **Medical students who want to learn clinical-trial methodology in a structured way**.\r\n\r\n## How to Use It in a Chat\r\n\r\nct-literature is a **conversational skill**: you simply tell the assistant what you want to look up — no commands, no parameter names to remember. Once installed as a WorkBuddy skill, you invoke it in a chat via the Skill tool; there is no extra setup, but it activates only when you call it.\r\n\r\nBelow are 7 real conversational examples ordered from simple to advanced. Each shows **\"You say\"** and **\"The assistant replies\"**, plus how the report is produced. Replies are **key excerpts** of the real interaction (progress lines and details compressed); whenever your decision is needed, the skill lists the options **on screen for click-to-confirm**, or asks you to say \"default / go ahead\". The primary deliverables are a self-contained **`lit_report.html`** (offline, printable) and **`lit_report.xlsx`**. By default the workbook has **3 data sheets: Overview → Literature master → Evidence Log** (plus a README cover). The **Safety-related** sheet is opt-in — it appears only when you add `--safety` (the CSM qualitative subset), since the safety/CSM literature subset is a safety-oriented concern, not a default deliverable of every literature search.\r\n\r\n### Example 1 · A simple literature search (with preview confirmation)\r\n**You say:**\r\n> Search published literature on osimertinib in NSCLC, 2018–2025, focused on adverse events / post-marketing safety.\r\n\r\n**Assistant replies (excerpt):**\r\n> Recognized as a **simple request** (topic + year + safety angle all clear) → straight to a preview for your confirmation:\r\n\r\n| Setting | Value |\r\n|---|---|\r\n| Topic | osimertinib + NSCLC (synonym family AZD9291 / Tagrisso / 奥希替尼 merged automatically) |\r\n| Years | 2018–2025 · safety focus ✅ |\r\n| Sources | OpenAlex + Europe PMC + preprints (default three) · 50 works per source |\r\n| Verification | all (anti-hallucination, default ON) |\r\n\r\nConfirmed → keyword list reviewed → search runs (~30–60 s):\r\n> Done: **103** de-duplicated unique works (3 sources); **81 safety-related**. Citation verification 100/103 passed. Report written to `lit_report.html` and `lit_report.xlsx` (Literature master + Safety-related sheets).\r\n\r\n**📌 No key needed** — OpenAlex keyless pool covers a small search; the skill fetches and writes the report directly.\r\n\r\n### Example 2 · Evidence-synthesis layer only (systematic reviews / meta-analyses)\r\n**You say:**\r\n> Same topic, but keep only systematic reviews and meta-analyses — I want the synthesis, not single trials.\r\n\r\n**Assistant replies (excerpt):**\r\n> Restricting `--review-type systematic-review/meta-analysis` and re-searching. **3 systematic reviews hit** (after cross-source de-dup) — listed separately with full abstracts and citation verification. If a type has zero hits, the skill says \"no works of this type\" rather than padding with near-matches.\r\n\r\n**📌 Safety subset vs FAERS is a key distinction:** `--safety` (CSM qualitative subset = published case reports / PV articles) and `ct-safety` (structured FAERS signal detection, PRR / ROR / IC) are **clearly different data types** — literature complements but does not replace FAERS.\r\n\r\n### Example 3 · Broader coverage (Europe PMC + Semantic Scholar)\r\n**You say:**\r\n> Also pull MEDLINE via Europe PMC and citation-rank via Semantic Scholar.\r\n\r\n**Assistant replies (excerpt):**\r\n> Adding `--with-europepmc` (MeSH precision) and `--with-semantic-scholar` (citation ranking). S2 is skipped automatically if no key or on 429 — OpenAlex + Europe PMC still produce results.\r\n\r\n### Example 4 · Configure the OpenAlex key (self-config)\r\n**You say:**\r\n> How do I lift the 100/day limit for large searches?\r\n\r\n**Assistant replies (excerpt):**\r\n> Apply for a free key (~30s) at . Configure it yourself via (a) adding `OPENALEX_API_KEY=` to `ct-literature/.env`, (b) exporting an `OPENALEX_API_KEY` environment variable, or (c) passing `--openalex-key ` on the command line. **Note:** chat messages may be logged or retained by the platform — for maximum secrecy use the self-config routes (a)–(c). If you prefer, you can also ask the assistant to write the key into `ct-literature/.env` for you — it stores the key locally, never echoes it back, never logs it, and sends it only over HTTPS to the official OpenAlex API.\r\n\r\n**With a key, retrieval is faster AND your key never leaves your machine:** the key enters OpenAlex's keyed pool (~100k credits/day, rarely 429'd), and the OpenAlex / Semantic Scholar sources switch to **local direct-connect retrieval, bypassing the cloud** — your key is used in local memory only to call the official APIs and is never sent to any third-party service. Only keyless users go through the cloud channel (OpenAlex anonymous pool; Semantic Scholar auto-skipped).\r\n\r\n### Example 5 · Complex request → parameter confirmation (menu / card style)\r\n**You say:**\r\n> I want to look into diabetes treatment literature.\r\n\r\n**Assistant replies (excerpt):**\r\n> Triage → **Complex** (topic = diabetes is clear, but key parameters — **literature type / year / sources** — are still open). The skill does not decide for you; it lists what it detected and asks you to confirm two things on screen:\r\n\r\n> Detected: topic `diabetes treatment` ｜ To confirm: ① **Literature type** (all types [recommended] / systematic reviews + meta-analyses only / SR + RCT + case reports…) ② **Run now?** (✅ run / ✏️ adjust parameters / ❌ cancel)\r\n> Not sure about the differences? Say \"explain the differences between these choices in detail\" — it explains the clinical / statistical meaning first, then you decide.\r\n\r\nAfter parameters are confirmed there are **two more gates** before any network call: ① **keyword confirmation** — the expanded keyword list (e.g. `diabetes mellitus`, Chinese/English synonym families) is shown for your review; ② **preview confirmation** — the same settings table as Example 1; the search only runs after you approve.\r\n\r\n**📌 Note:** Simple requests (topic + ≥2 parameters clear) skip the confirmation menu and go straight to preview (Example 1); Vague requests use grill-me clarification (Example 6).\r\n\r\n### Example 6 · Vague request → bounded clarifying questions (grill-me style)\r\n**You say:**\r\n> Find me something about that new cancer drug.\r\n\r\n**Assistant replies (excerpt):**\r\n> Triage → **Vague** (\"that new drug\" cannot be pinned down) → asks back with **bounded grill-me (max 2 rounds)**, each round with a recommended default; if it is still unclear it falls back to defaults + a preview reminder:\r\n\r\n> **Round 1:** \"Can't recall the name — that's fine. Which **class** is it? A 3rd-gen EGFR-TKI (e.g. osimertinib), an immune-checkpoint inhibitor (e.g. pembrolizumab), or a target (PD-1 / EGFR)? Or paste the brand name / indication and I'll reverse-look it up.\"\r\n> You: osimertinib, the lung-cancer one.\r\n> **Round 2 (narrowing scope):** \"Locked to osimertinib. Should the cancer type be limited to **NSCLC**? (default: yes). Any year range or safety angle? (none → last 5 years, all types)\"\r\n\r\nAfter the two rounds the skill gives a **need profile + recommended settings** for confirmation: \"topic osimertinib + NSCLC ｜ last 5 years ｜ all types ｜ default three sources\" — approve to move to preview; if still unclear it runs with defaults and flags that in the preview.\r\n\r\n### Example 7 · On-demand batch download of legal open-access PDFs\r\n**You say:**\r\n> Download the PDFs of the 40 most recently published works in the report, and log how long it takes.\r\n\r\n**Assistant replies (excerpt):**\r\n> Takes the newest 40 works from the 103-work merged set (sorted by publication date, newest first) and starts the batch download — **before starting it tells you the expected pace**:\r\n\r\n> Starting batch download of 40 PDFs: **roughly 10–20 s per paper** (network / rate limits vary) — please be patient, no action needed.\r\n\r\nDownloading routes automatically: **OA / preprint direct links download locally first** (Frontiers, MDPI, PMC author manuscripts, …); only items that cannot be fetched locally but carry an OA/preprint signal go through the cloud decode chain; paywalled works with no legal copy are not forced. On completion it reports honestly:\r\n\r\n> Done: **36/40** succeeded, took **853 s (14.2 min)**. PDFs saved to `pdfs_latest40/`; **the Excel report now has the「PDF 本地路径」column updated** (36 absolute paths + 3 \"failed\" markers).\r\n> ⚠️ The 4 missing are all paywalled (NEJM / JCO / BMJ) — get them via your library or the authors; the skill never bypasses a paywall.\r\n\r\n**⏱ Pace:** roughly 10–20 s per paper (incl. rate-limit backoff + cloud decode); a 40-work batch runs ~10–15 min — that is normal wait, not a hang; single direct-connect papers (Frontiers-class) usually finish in 3–5 s.\r\n\r\n---\r\n\r\n## Data Sources\r\n\r\n| Source | Key | Role |\r\n|---|---|---|\r\n| OpenAlex | recommended (free key; keyless = 100/day since 2026-02-13) | **Primary** — broad, citation-rich |\r\n| Europe PMC | no key | **On by default** (`--no-with-europepmc` to disable) — MEDLINE/MeSH biomedical precision |\r\n| Semantic Scholar | key recommended (manual form review) | **Opt-in only** `--with-semantic-scholar` — citation ranking; **not part of default sources**, skipped unless explicitly enabled with a key |\r\n| bioRxiv | no key (via Europe PMC PPR) | **On by default** (`--no-with-biorxiv` to disable) — biomedical preprints |\r\n| medRxiv | no key (via Europe PMC PPR) | **On by default** (`--no-with-medrxiv` to disable) — medical/clinical preprints |\r\n| arXiv | no key | Optional `--with-arxiv` — physics/CS/ML methodology breadth |\r\n| PROSPERO | token required (undocumented auth header) | Optional `--with-prospero` — systematic-review registry / protocol discovery; **reserved source**, degrades to a no-op skip until a working token + header is supplied |\r\n\r\n### How the sources fit together\r\n\r\nThe default trio — **OpenAlex (primary) + Europe PMC (on by default) + bioRxiv/medRxiv (on by default)** — already reaches almost the entire published landscape: through these three endpoints you get PubMed / PMC, the bioRxiv / medRxiv / arXiv preprints, and the Crossref, Semantic Scholar, CORE, and Unpaywall records (Semantic Scholar records are surfaced via OpenAlex's linkage even without querying S2 directly). The extra sources are opt-in, not because the trio is incomplete, but for two practical reasons:\r\n\r\n- **Preprint freshness** — **bioRxiv / medRxiv are now on by default**, pulling preprints straight from the source instead of waiting for them to propagate through Europe PMC's PPR feed.\r\n- **Resilience against rate limits** — if Europe PMC is throttled (HTTP 429), the standalone preprint endpoints let you keep widening coverage without depending on a single bottleneck. (Semantic Scholar is a separate opt-in source requiring `--with-semantic-scholar` + a key, and is not part of the default pipeline.)\r\n\r\n## Why You Can Trust the Output — Anti-Hallucination\r\n\r\nLLM-powered literature tools are notorious for **inventing papers that don't exist** — fabricated DOIs, wrong PMIDs, plausible-but-fake citations. ct-literature is built to make that impossible *by construction*, through four independent guardrails plus two operational safeguards:\r\n\r\n1. **Every citation is resolved against its live source (P0, default ON).** Before a work reaches your report, its identifier is checked against the real bibliographic API: DOI → `doi.org` (HTTP 2xx), PMID → Europe PMC `EXT_ID`, OpenAlex id → `api.openalex.org/works/`. Each work is tagged `citation_verified` plus a status of `verified` / `bot_blocked` / `unresolved` / `no_identifier` / `suspicious`. A **malformed DOI is flagged `suspicious`** — a likely hallucinated identifier is caught *before* it can appear in the report. Scope it with `--verify {all|top|background}`; the default `all` verifies every work.\r\n - **`bot_blocked`**: some publishers (NEJM, JAMA, Wiley, MDPI…) return **403** to programmatic access even though the DOI is real. The skill reports this distinctly — it is *not* a broken link, and the work stays `verified=True`.\r\n2. **Title / author consistency depth (v0.6.11).** Once an identifier resolves to a live resource, the skill fetches that resource's canonical metadata (title + first-author surname) from the authoritative, bot-friendly API — **Crossref** for DOIs (bot-friendly even when the publisher blocks `doi.org`), **Europe PMC** for PMIDs, **OpenAlex** for OpenAlex ids — and compares it to the work you hold. A resolved-but-**different** paper is flagged **`mismatch`** (not `verified`); a `bot_blocked` DOI whose Crossref metadata matches is **upgraded to `verified`**. A hallucinated-but-real DOI is thus caught *even when it resolves*. Metadata-fetch failure degrades gracefully to \"verified, consistency unchecked\" — it never invents a mismatch. Opt out with `--no-consistency`.\r\n3. **Full provenance is recorded, not summarized away.** Every merged work keeps its `sources` list (which API produced it), and `evidence_log.json` stores an immutable-style audit trail: query → source → hit count → retrieved_at → verification rate. You can always trace a claim back to the exact API call that produced it.\r\n4. **The report never pads gaps with fluent prose.** Every factual line in the report carries a source label or an explicit `⚠️ needs official verification` marker. The skill does **not** generate plausible-looking evidence to fill holes — if a source failed or a work is unverified, that is shown, not hidden.\r\n\r\nOperational safeguards reinforce this: **Safe Preview** keeps normalization / reporting on your machine (no remote code execution), and **source-aware skip** avoids redundant re-checks while still trusting each identifier *by provenance* (a paper OpenAlex returned already carries a real OpenAlex id, so it isn't re-queried there). All of this follows the anti-hallucination spec (§17.1).\r\n\r\n**Net:** the references this skill gives you are real, resolvable, and traceable — safe to put in a slide, a protocol, or a CSR appendix, provided you validate against the official source before any regulatory submission (see the [First-Time FAQ](#first-time-faq)).\r\n\r\n## What Can It Do — Scenarios\r\n\r\nThe skill covers published-evidence retrieval across the clinical-trial lifecycle. Each row gives the typical **situation** and a line you can **copy verbatim** under \"Try saying\".\r\n\r\n### ① Published-evidence search (OpenAlex, primary)\r\n| Situation | Try saying in chat |\r\n|:---|:---|\r\n| Evidence on a drug / disease / method | \"Find systematic reviews on osimertinib in NSCLC\" |\r\n| Recent literature with a year filter | \"Papers on CAR-T in lymphoma since 2020\" |\r\n| A topic with a safety angle | \"Post-marketing safety literature for drug X\" |\r\n\r\n### ② Broader / deeper coverage (optional sources)\r\n| Situation | Try saying in chat |\r\n|:---|:---|\r\n| MEDLINE / MeSH biomedical precision | \"Also search Europe PMC for this topic\" |\r\n| Citation-ranked relevance | \"Rank these by citation count via Semantic Scholar\" |\r\n\r\n### ③ Output formats & exports\r\n| Situation | Try saying in chat |\r\n|:---|:---|\r\n| Excel deliverable | \"Export the literature as an Excel file\" |\r\n| Self-contained HTML report only | \"Just the HTML report, skip Excel\" |\r\n| Import into **Zotero** (reference manager) | \"Export as Zotero RIS / CSV\" — get `zotero.ris` / `zotero.csv`, import into the Zotero desktop app or browser connector |\r\n| Browse as an **Obsidian** knowledge graph | \"Export to Obsidian\" — get one Markdown note per paper + a `Literature MOC.md` index; open the folder as a vault to see the paper network |\r\n| **Batch-download OA PDFs** (v1.0.0) | \"Download the newest 40 PDFs and log the time\" — files land in a local `pdfs*/` folder (DOI-named), the Excel report gets a 「PDF 本地路径」column, failures are marked honestly |\r\n\r\n### ④ Evidence verification & provenance (P0, default ON)\r\n| Situation | Try saying in chat |\r\n|:---|:---|\r\n| Verify every DOI/PMID is real (anti-hallucination) | \"Verify the citations are real before you report\" |\r\n| Also confirm title/author match the paper (v0.6.11) | \"Make sure the DOI actually points to this paper\" |\r\n| Trace where each hit came from | \"Show me the evidence provenance / source log\" |\r\n| Verify only the top-N works (fast large searches) | \"Just verify the top 15 citations\" |\r\n| Skip verification (faster, preview-only) | \"Don't verify citations this time\" |\r\n\r\n### ⑤ Key / setup\r\n| Situation | Try saying in chat |\r\n|:---|:---|\r\n| Lift the OpenAlex rate limit | \"How do I raise the rate limit?\" |\r\n| Check what's configured | \"What keys does the skill currently see?\" |\r\n\r\n> The sibling skills are described in their own READMEs; ordinary users only need to say what they want in plain language — the skill routes the right sources and writes the report.\r\n\r\n---\r\n\r\n## First-Time FAQ\r\n\r\n**Q: Do I need a key to run?** A: No. OpenAlex keyless pool = 100 credits/day (enough for small searches); a free key lifts it to 100k/day. Europe PMC and Semantic Scholar need no key.\r\n\r\n**Q: Where does my query go?** A: Your topic query and filters are sent to the public bibliographic APIs — OpenAlex, Europe PMC, and Semantic Scholar (when enabled). No confidential or sponsor data is ever sent.\r\n\r\n**Q: What's the difference from `ct-safety`?** A: `ct-literature` = published *qualitative* evidence (papers / reviews / case reports); `ct-safety` = structured FAERS disproportionality (PRR / ROR / IC). They are explicitly distinct data types — literature complements but does not replace FAERS.\r\n\r\n**Q: On a Chinese system, is the output in Chinese?** A: Yes. **Conversation answers and the reports** (HTML / Excel) follow your OS setting by default (Chinese on a Chinese-OS, English otherwise), and you can force-switch anytime with one sentence (e.g. \"switch to English\"). **Console progress lines** (fetch / download process hints) are Chinese auxiliary output — runtime noise, they do not affect the language of the answers or deliverables.\r\n\r\n**Q: Semantic Scholar keeps failing / being skipped?** A: The S2 key requires a manual form review (not auto-issued, waits after applying), so it is usually absent short-term. When no key is configured the source is **skipped entirely** (no network request) rather than attempting-and-degrading. Configure it later if you need citation ranking.\r\n\r\n**Q: How long does a search take? What are the rate limits?** A:\r\n- **Typical latency:** Enabled sources run **in parallel with each other** (one worker per source), but **each source pages serially** — requests inside a source are chained one after another, because parallel paging would raise rate-limit / ban risk (e.g. on the OpenAlex keyless pool). Europe PMC ~1s/page, OpenAlex ~2s/page, so the wall-clock is the *slowest* source, not the sum. A 3-source search pulling ~50 works typically finishes in **10–30 seconds** (plus ~1–4 min more when full citation verification is on — see the pre-run time estimate). Adding preprints (bioRXiv/medRXiv/arXiv) adds a few seconds more.\r\n- **Result cap:** Default `max_results` is **50 works per source**; there is no hard ceiling (raising it is allowed), but time and API usage scale linearly. **Measured 2026-08-14** (two sources, `osimertinib ILD`): `--max 100` → fetch+merge **~20 s**, full verification **~78 s** (≈0.64 s/work after cross-source dedup, which keeps **~62%** of fetched works with two sources). **Suggested ceiling for a ~5-minute run: 300 works per source with two sources, ~250 with three sources** (either way ≈370–380 unique merged works; dedup is automatic, so 3×250 does not mean 750 in the report). Beyond that total time grows ~linearly — for very large harvests, configure an OpenAlex key and use `--verify top 15` instead of pushing `max_results`.\r\n- **Rate limits:**\r\n - **OpenAlex (keyless):** 100 credits/day (since 2026-02-13). A single multi-page search can use 5–20 credits. A free key lifts this to **100k/day**.\r\n - **Europe PMC:** No hard key limit, but please keep request frequency reasonable (no tight loops).\r\n - **Semantic Scholar (no key):** Prone to HTTP 429; the skill skips it entirely when no key is configured.\r\n- **Tip:** Start with the default sources (OpenAlex + Europe PMC) and a modest `max_results`; only enable extra sources if you need broader coverage.\r\n\r\n**Q: Why can't the fetch be faster?** A: Because the skill only uses the **official public access methods each site provides** (their public APIs / endpoints) and **never violates any site's terms or policies** — it fetches politely, source by source, page by page, so it cannot deliver the \"crawl a huge dataset in minutes\" effect of an aggressive scraper. Concretely: (1) **Different sources already run in parallel** (one worker per source) — adding more cross-source parallelism won't help. (2) **Each source must page serially** — the public bibliographic APIs (OpenAlex keyless pool, Europe PMC polite pool) throttle or ban clients that fire many parallel requests; serial paging is what keeps you under the ban radar. (3) If a run feels slow, the usual bottleneck is **full citation verification** (default ON; one or more HTTP lookups per work) — switch to `--verify top` (default top-15) to cut ~1–4 minutes. (4) Keep `max_results` moderate — time and API usage scale linearly with it. Bulk PDF fetching is the other multi-second-per-work operation (each request follows a redirect chain).\r\n\r\n**Q: Can I search in Chinese?** A: Partially — the skill auto-translates Chinese topics to English through **bundled offline dictionaries** (~900 entries: medical terms + drug INN names + brand names like 泰瑞沙→Tagrisso/osimertinib + MeSH synonyms; no network call) before querying the APIs, and the report banner shows the original and the translation as `中文 → English`. Equivalent names are combined with boolean OR to widen recall (e.g. `osimertinib OR Tagrisso`, `lung cancer OR pulmonary neoplasm`). Terms the dictionaries do not cover pass through as-is (recall may suffer) and a notice lists the unmapped ones — you can extend the dictionaries yourself by adding entries to `references/user_terms.json` (same `{中文: \"English\"}` format, values may be a list of synonyms; the file is git-ignored so your additions are never published). For best recall, use English terms — especially for rare conditions or novel compounds.\r\n\r\n**Q: Why don't you support Chinese domestic databases (e.g. CNKI / 知网)?** A: Deliberately not supported, for three reasons. (1) **Marginal value** — this skill targets the publicly retrievable international evidence base (OpenAlex / Europe PMC / ...); the incremental coverage of Chinese-only databases is small, and much of their content overlaps or is already indexed internationally. (2) **No compliant channel exists** — CNKI and similar Chinese databases **do not offer public APIs to individuals** (only to contracted, paying institutions), and they aggressively block — and have sued — web crawlers; automated retrieval would have neither a legal interface nor a defensible risk posture, violating this skill's rule of \"official public access only, never breach a site's terms\". (3) **ROI** — paying that compliance/legal risk for marginal coverage is not worth it. If you need a specific Chinese paper, search CNKI yourself and export the citation (RIS / BibTeX) for your records.\r\n\r\n**Q: Can the skill download full-text PDFs?** A: Yes — but **for open-access (OA) works only**, in two ways. (1) The Excel and HTML reports always include an **\"Open Access\"** column with a direct link to a free OA copy when one exists (typically 60–80% of recent works); **non-OA, paywalled papers show \"—\" and are not supported**. (2) Since v1.0.0 the skill can also **download the OA PDFs to disk for you**. The compliance boundary first:\r\n- **OA-only assistance; non-OA is not supported:** the skill pulls only from legitimate OA channels (publisher OA links, Europe PMC, PMC author manuscripts, open preprints) — **it performs no illegal operation of any kind**: no cracking, no bypassing, nothing touches a paywall. Paywalled works (NEJM / JCO / BMJ …) are honestly marked **\"failed\"** — get them via your institution's subscription, interlibrary loan, or the corresponding author.\r\n- **For personal use only; no commercial use:** this feature exists to make it convenient for an individual to get OA copies. Do not use the downloaded content for commercial purposes.\r\n- **OA direct-fetch fails → it looks for a preprint substitute:** when the canonical OA link cannot be pulled, it searches for an earlier preprint or author manuscript (bioRxiv / medRxiv / PMC) as a substitute; only when no usable copy exists is the work marked failed.\r\n- **Use the free resources wisely and pace your requests (avoid an IP block):** OA providers commonly block programmatic high-frequency downloads — to stay clear of legal trouble, the skill **strictly enforces a ≥5 s gap between consecutive download requests** and never pushes against a server's load threshold; please also control your own batch frequency and **do not over-use the free resources**. The skill has built-in cross-domain concurrency limits and per-domain throttling, and **a single batch above 50 works is refused outright** — split it (top-N / single source).\r\n\r\nExperience details: before a batch starts it states the pace (roughly **10–20 s per paper**; a 40-work batch runs ~10–15 min); afterwards PDFs land in a `pdfs*/` folder (DOI-named), it reports elapsed time (\"N/M succeeded, took X s\"), and **the Excel report gets a 「PDF 本地路径」column** (absolute path on success, \"failed\" otherwise). Keyless auto channels give a real-world success rate of roughly **70–90%** (OA coverage and publisher blocking vary). How to ask: say \"download ALL OA PDFs\", \"download the newest 40 PDFs and log the time\", or give a DOI/PMID list.\r\n\r\n**Q: What if I found an error in the result — how do I report it?**\r\nA: This skill follows the bug-report workflow. If you suspect the result is wrong (or the engine errored), just say **\"report a bug\" / \"上报问题\" / \"提交错误报告\"**. The skill also **proactively asks** whether to report when it detects a likely defect (e.g. the engine errors or retries still fail) — at most **once per session**, and you can always decline. Either way, the assistant will:\r\n1. **Propose a sanitized report** (11-field whitelist: skill / skill_version / test / error_type / error_code / engine_status / description / locale / query_origin / session_hash / attempts — **no raw input values or personal data**, except the `description` field where you decide what to disclose, e.g. the algorithm/function used and the error message);\r\n2. **Show the full report text for your review** — you can add a problem description or correct anything before confirming;\r\n3. **Send after your explicit confirmation** — to the unified endpoint `https://ct-bugreport.coze.site/run` (if this session called coze) or saved locally + emailed to the author (if purely local, data never leaves your machine);\r\n4. **Receive an acknowledgment** — including whether a previously submitted report from your source has already been fixed (with the fix note) or is still pending.\r\n\r\nYou stay in full control: the report is shown to you **before** anything is sent, and nothing is transmitted without your explicit \"send\" confirmation.\r\n\r\n---\r\n\r\n## Security & Privacy\r\n\r\n### Safe Preview (local computation)\r\n- **Runs locally:** The normalize / report / Excel rendering steps run entirely on your machine — no code is executed on any remote server beyond the bundled scripts.\r\n- **Traceable, not fabricated:** Every factual claim in the report carries a source label (`sources` list per work) or an `⚠️ needs official verification` marker; it never fills evidence gaps with fluent prose.\r\n- Outputs are for reference only; validate against official sources before regulatory submissions.\r\n\r\n### Outbound & Privacy\r\n- **Bibliographic search (public APIs only):** your topic + filters go to **OpenAlex** / **Europe PMC** / **Semantic Scholar** (only the sources you enable), plus **doi.org** and **Crossref** during citation verification. No confidential / sponsor data is ever sent.\r\n- **Metadata sent with every outbound request:** each request carries ① your business parameters (topic / filters) plus ② `query_origin` (a SHA-256 hash of your machine hostname — used for rate-limit attribution and billing; **this is a one-way hash, NOT your plaintext hostname**) and `locale` (your OS language, used to localize UI prompts). These fields are computed locally on your machine and sent as opaque identifiers — no PII, no hostname, no IP leaves your device.\r\n- **Bug reports (opt-in, user-confirmed):** `adapters/bug_report.py` sends an **11-key sanitized envelope** (skill / version / error_type / description / … — never raw data or subject records) to `https://ct-bugreport.coze.site/run` **only after you explicitly confirm** a two-stage prompt; without cloud access it falls back to a local file.\r\n- **Keys stay on your machine:** keys are read from your local `ct-literature/.env` and never ship with the package (only `.env.example` ships). Apply for your own OpenAlex key at and configure it yourself via the [First-Time FAQ](#first-time-faq) (`.env` / env var / `--openalex-key`); never commit `.env` to a repo. (Optional, not recommended: you may paste a key and ask the assistant to write it locally — stored only on your machine, never echoed or logged. Self-configuring is preferred.)\r\n\r\n---\r\n\r\n## Advanced Reference (Developers)\r\n\r\nCLI helpers, runtime requirements, the architecture tree, and the unified work-mode schema have moved here so everyday users don't need them. See [`SKILL.md`](SKILL.md) and [`CHANGELOG.md`](CHANGELOG.md) for the agent-facing spec and version history.\r\n\r\n### Runtime & requirements\r\n| Item | Requirement |\r\n|---|---|\r\n| Runtime | Python 3.11+ (CPython). The pipeline uses **only the Python standard library** (`urllib`) for HTTP — **no third-party dependency is required**. |\r\n| Keys (optional) | OpenAlex free key (recommended for scale); Semantic Scholar key optional (lifts ~1 req/s limit). Both via `.env` / env var / `--openalex-key`. |\r\n| Sibling skills | `ct-registry` (trial registries), `ct-safety` (FAERS), `ct-pipeline` (intel brief) — ct-literature seeds topics and is seeded by them; all install from GitHub. |\r\n\r\n### Optional tool · English→Chinese abstract term-annotation (`abstract_translator.py`)\r\nA small standalone CLI that annotates English text with Chinese glosses for matched medical terms — **term-level substitution, not full-text translation** (unmatched words stay in English). It is **not** part of the retrieval pipeline; run it on demand on a text or file:\r\n\r\n```bash\r\n# annotate a text snippet\r\npython scripts/abstract_translator.py --text \"Osimertinib is a third-generation EGFR-TKI used in NSCLC.\"\r\n# annotate a file (e.g. an abstract), output ASCII or JSON\r\npython scripts/abstract_translator.py --file abstract.txt --format ascii\r\npython scripts/abstract_translator.py --file abstract.txt --format json --output out.json\r\n```\r\n\r\nOutput shows the original and the annotated version (e.g. `randomized controlled trial` → 【随机对照试验】, `NSCLC` → 【非小细胞肺癌】, `overall survival` → 【总生存期】). The dictionary is a bundled offline EN→ZH medical-terms list (~130 entries, study types / trial-design / statistics) plus English entries from the shared `term_map.json`; no network call. For fluent full-sentence translation, use a general translation service instead.\r\n\r\n### Architecture\r\n```\r\nct-literature/\r\n├── SKILL.md # agent-facing spec (English body)\r\n├── CHANGELOG.md # version history\r\n├── adapters/ # one fetcher + verifier per public API\r\n│ ├── fetch_openalex.py # primary source\r\n│ ├── fetch_europepmc.py # MEDLINE/MeSH (on by default)\r\n│ ├── fetch_semantic_scholar.py # optional citation rank (skippable)\r\n│ ├── fetch_preprints.py # bioRxiv / medRxiv\r\n│ ├── fetch_arxiv.py # arXiv\r\n│ ├── fetch_prospero.py # PROSPERO (reserved, dormant until token set)\r\n│ ├── http_utils.py # shared retry / headers / key load\r\n│ └── verify_citations.py # P0 citation verification + title/author consistency\r\n├── scripts/\r\n│ ├── ct_literature.py # orchestration: fetch → normalize → verify → report/export\r\n│ ├── normalize.py # multi-source merge + dedupe\r\n│ ├── score_relevance.py # relevance scoring\r\n│ ├── screen_prisma.py # deterministic PRISMA title/abstract screen\r\n│ ├── export_xlsx.py # Excel deliverable (excel_style)\r\n│ ├── export_html.py # self-contained HTML report\r\n│ ├── format_citations.py # APA/Nature/Vancouver/IEEE/GB7714 + BibTeX/RIS\r\n│ ├── evidence_log.py # provenance audit trail (evidence_log.json/.md)\r\n│ ├── obsidian_exporter.py # Obsidian notes + MOC\r\n│ ├── zotero_exporter.py # Zotero RIS/CSV\r\n│ ├── i18n.py # bilingual single source of truth\r\n│ └── excel_style.py, … # shared style\r\n├── references/ # SOP, key setup, search menu, multi-db method\r\n└── assets/icon.svg # A-tier logo\r\n```\r\n\r\n### CLI examples (developers)\r\n```bash\r\n# Primary (OpenAlex, no key)\r\npython scripts/ct_literature.py --topic \"osimertinib\" \\\r\n --review-type systematic-review --year-from 2018 --safety --run --out-dir ./out\r\n\r\n# Add Europe PMC (MeSH) + Semantic Scholar (citation rank)\r\npython scripts/ct_literature.py --topic \"osimertinib\" \\\r\n --with-europepmc --with-semantic-scholar --run --out-dir ./out\r\n\r\n# Recommended (zero extra flags): put the key in the skill .env, then just run\r\ncp .env.example .env # edit .env -> OPENALEX_API_KEY=your_key\r\npython scripts/ct_literature.py --topic \"osimertinib\" --safety --run --out-dir ./out\r\n\r\n# P0 · citation verification (default ON, mode=background) + evidence log are automatic under --run.\r\n# Scope it with --verify {all|top|background}; source-aware skip avoids redundant same-source\r\n# re-resolution (a paper from OpenAlex/Europe PMC is trusted by provenance).\r\n# Verification is an anti-hallucination gate (§17.1 P0): it cannot be fully disabled —\r\n# \"none\" is not a valid mode (v0.9.6 removed the CLI bypass).\r\npython scripts/ct_literature.py --topic \"osimertinib\" --run --out-dir ./out\r\n# Best speed/coverage balance for large result sets: verify only the top-ranked works (default top 15; --verify-top-n adjusts N — verification is never disabled)\r\npython scripts/ct_literature.py --topic \"osimertinib\" --run --verify top --out-dir ./out\r\n# Non-blocking default: background verification streams in without delaying the report\r\npython scripts/ct_literature.py --topic \"osimertinib\" --run --verify background --out-dir ./out\r\n# v0.6.11 · skip the title/author consistency layer (verification still resolves identifiers)\r\npython scripts/ct_literature.py --topic \"osimertinib\" --run --no-consistency --out-dir ./out\r\n# v0.7.0 · stream progress as NDJSON events on stdout (agent-facing: --progress json\r\n# redirects sub-module prints to stderr, so stdout stays parseable; events:\r\n# run_start / source_done / source_failed / fetch_done / verify_progress / verify_done /\r\n# evidence_log / export_done / export_failed / run_done, one JSON object per line, flushed)\r\npython scripts/ct_literature.py --topic \"osimertinib\" --run --progress json --out-dir ./out\r\n# v0.7.0 · two-phase delivery: unverified report in seconds, verification backfills later\r\npython scripts/ct_literature.py --topic \"osimertinib\" --run --verify background --out-dir ./out\r\n\r\n# P1 · PROSPERO systematic-review registry (opt-in, reserved source — dormant until a token is set)\r\npython scripts/ct_literature.py --topic \"osimertinib\" \\\r\n --with-prospero --prospero-token \"$PROSPERO_API_TOKEN\" --run --out-dir ./out\r\n```\r\n\r\n### Unified work mode (output schema)\r\n```\r\n{\r\n source, id, title, authors, year, publication_date, publication, journal_iso,\r\n type, study_type, cited_by_count, url, open_access_url,\r\n pmid, pmcid, doi,\r\n abstract_snippet, # full text, not truncated\r\n mesh, concepts, keywords, funders,\r\n language, is_retracted, is_safety,\r\n volume, issue, page,\r\n affiliations, # Europe PMC only\r\n sources, # contributing source list\r\n # --- attached by P0 verification (verify_citations.py) ---\r\n citation_verified, # bool\r\n citation_verify_status, # verified | bot_blocked | mismatch |\r\n # unresolved | no_identifier | suspicious | unverified_sampled\r\n citation_verify_note, # human-readable detail\r\n citation_consistency, # bool | None (v0.6.11)\r\n citation_title_ratio # float | None (normalized title similarity)\r\n}\r\n```\r\n\r\n---\r\n\r\n**Version**: v1.0.1 | **License**: MIT | **Authors**: medstatstar, phoe-zip\r\n\r\nFor feature requests, bug reports, or other feedback, feel free to contact the author directly at medstatstar@gmail.com (Wintone Zhang / 张文彤).\r\n\r\n---\r\n\r\n## Confidentiality Notice\r\n\r\n> The CT series consists of 20+ specialized domain skills, organized into **two tiers — A, B** — by \"whether the input contains confidential information\" (network / egress / publish are independent orthogonal attributes), providing full coverage of the entire new-drug clinical trial (Clinical Trial) lifecycle.\r\n>\r\n> - **Tier A (non-confidential input)**: run fully locally using only ordinary data; Tier A may need external public retrieval but involves no confidential information. These skills are published openly on GitHub.\r\n> - **Tier B (confidential input)**: accept strictly confidential clinical-trial data / protocols / CRFs from pharma sponsors (e.g., ct-analysis, ct-sdtm, ct-protocol, ct-eligibility); Tier B is processed locally and never leaves the boundary (egress=none), or additionally requires policy approval (egress=approval-req, e.g. ct-eligibility). Tier B packages contain zero confidential data but are NOT publicly published (stays fully local) — confidential input never ships with the package or leaves the machine. For custom / on-prem deployment, contact the author.\r\n>\r\n> 📧 Contact: medstatstar@gmail.com (Wintone Zhang / 张文彤)\r\n\r\n> 🌐 Other languages: [中文 README](README_zh-CN.md)\n\nFile v1.0.1:_meta.json\n\n{\n \"ownerId\": \"kn7amqq1jv28skb63wavr6shah89jsm5\",\n \"slug\": \"ct-literature\",\n \"version\": \"1.0.1\",\n \"publishedAt\": 1790910100254\n}\n\nFile v1.0.1:references/capabilities.md\n\n# Capabilities & Reference Detail (ct-literature)\r\n\r\n> Extracted verbatim from `SKILL.md` (2026-09-27, line-budget refactor — SKILL.md ≤ 200 lines per ct-base §16.1 / F02). No content removed; SKILL.md keeps one-line pointers here. Load this file when you need the full source / feature / output / invocation detail.\r\n\r\n## Positioning within the ct- library\r\n\r\nThe four A-tier public-intel skills (non-confidential input, `network=public-retrieval`) are complementary:\r\n\r\n| Skill | Answers | Object retrieved | Source family |\r\n|---|---|---|---|\r\n| `ct-registry` | What trials are registered / ongoing / completed? | Trial-registry metadata | Registry libraries |\r\n| `ct-literature` | What evidence has been *published*? | Publications | Literature libraries |\r\n| `ct-safety` | Is a drug–event over-reported (signal)? | FAERS cases | Adverse-event databases |\r\n| `ct-pipeline` | Aggregate the above into a strategic intel brief | Consumes the three JSONs | Public-intel layer |\r\n\r\n**Boundaries:** `ct-registry` never fetches paper full-text/abstracts; `ct-literature` never fetches registry structured metadata. `ct-literature --safety` surfaces *published* safety literature — **qualitative**, must NOT replace `ct-safety`'s FAERS disproportionality. Not sure which skill? Route via `ct-advisor`; full competitive-intel picture → `ct-pipeline` directly (§15).\r\n\r\n## Data Sources (detail)\r\n\r\n| Source | Access | Status | Role |\r\n|---|---|---|---|\r\n| OpenAlex | Public REST; free key recommended (100k/day via `.env` auto-load) — keyless capped 100/day since 2026-02-13 | Required (primary) | Broad coverage + citation counts |\r\n| Europe PMC | Public REST (MEDLINE / PubMed Central), no key, MeSH-indexed | **Default ON** (`--no-with-europepmc`) | Biomedical precision + MeSH |\r\n| Cochrane (CDSR) | Via Europe PMC journal filter | Opt-in `--cochrane` | Cochrane Database of Systematic Reviews only |\r\n| Semantic Scholar | Public Graph API, no key, rate-limited (429) | **Opt-in only** `--with-semantic-scholar` (not part of default sources; requires key to be useful) | Citation-aware ranking; skipped when no key |\r\n| bioRxiv / medRxiv | Via Europe PMC `SRC:PPR` + publisher filter | **Default ON** (`--no-with-biorxiv` / `--no-with-medrxiv`) | Preprints (Tier P) |\r\n| arXiv | Public Atom API, no key | Optional `--with-arxiv` | Methodology breadth |\r\n| PROSPERO | Public REST (CRD York); **auth header undocumented** | Optional `--with-prospero` (key-gated, **reserved source**) | Duplication-avoidance / protocol discovery |\r\n\r\n> All are public bibliographic APIs — no WAF. The **default data sources are OpenAlex (primary) + Europe PMC (on by default) + bioRxiv/medRxiv (on by default)**; everything else (arXiv, Cochrane, PROSPERO, Semantic Scholar) are opt-in. OpenAlex keyless = 100 credits/day since 2026-02-13; a free key lifts to 100k/day (`--openalex-key` / env `OPENALEX_API_KEY` / skill `.env` auto-load; key never printed). Semantic Scholar is an explicit opt-in source (`--with-semantic-scholar` + key) and is **not considered in the default pipeline**.\r\n>\r\n> **PROSPERO is a reserved source (2026-08-12):** its public REST auth header is undocumented; unauthenticated probes return `{\"status\":\"error\",...}`. `--with-prospero` is a dormant interface: without a token it degrades to a graceful no-op skip (returns `None`, no file written) and is **not** claimed functional. Supply `--prospero-token` (+ `--prospero-header`) to exercise it; parser is schema-tolerant (JSON + XML) but must be re-validated against a real 200 before declared done. No token application planned.\r\n\r\n## Clinical guideline sources (`--with-guidelines`, opt-in · LOCAL corpus)\r\n\r\nGuidelines are **version-pinned** reference standards — at analysis time we read a **pre-built LOCAL corpus**, never \"fetch latest\" per run.\r\n\r\n> **Corpus boundary (audit note 2026-09-07):** every entry is a manually curated, version-pinned pointer to an organisation-issued document, but the builder's 12+ sources also surface reviews / consensus / adherence analyses and AI-in-medicine commentary. Such items are kept as background pointers only and are **not** authoritative clinical guidance — always resolve the linked source before relying on it; a `retrieved:false` pointer is an honest placeholder, never a fabricated citation.\r\n\r\n> **🔒 Data-protection split.** The skill tree ships **pointer-only** (`references/guidelines/guidelines_index.json`: org/title/URL/version — publish-safe). **Full-text documents are NEVER written into the skill** — they live in the author's self-controlled Coze KB (or an EXTERNAL local cache `~/.workbuddy/ct-guideline-docs`, opt-in via `--download`, off by default); ct-advisor consults that KB for native guideline Q&A.\r\n\r\n- **Author / build-time** (network): `python adapters/build_guidelines.py --topic --run` aggregates 12+ sources → `guidelines_index.json` (96 curated entries, schema v1). SAFE PREVIEW: omit `--run` (dry-run, no network/write).\r\n- **Analysis-time** (zero network): `--with-guidelines` on the main pipeline → `adapters/guideline_corpus.load()` reads the local index → `guidelines.json` + `guidelines` block in `.merged.json`.\r\n\r\n| Tier | Sources | How it got into the corpus |\r\n|---|---|---|\r\n| `api` | OpenAlex, Europe PMC, GIN, WHO IRIS | fetched by the builder (OA-PDF download attempted) |\r\n| `api` (key-gated) | NICE¹, MAGICapp, TRIP² | fetched if a key configured; else skipped |\r\n| `portal`→`api` | CPIC | genuine fetch via free keyless PostgREST API (`api.cpicpgx.org/v1`) |\r\n| `portal` pointer | NCCN, ADA, AHA, SIGN, CMA | best-effort public-portal scrape → graceful fallback to honest pointer (`retrieved:false`), never fabricated |\r\n\r\n**Build-time portal fetch (`adapters/portal_fetch.py`):** every fetcher is wrapped so it never raises — failed fetch degrades to the honest pointer, so the corpus is always honest (build once, read many). Each record carries `access` (`api`/`portal`) + `retrieved`; `guideline_corpus.load()` filters by topic/org and returns `corpus_missing` (with the builder command) if the index is absent. ¹ NICE REST auth undocumented (like PROSPERO) — skip until a token works. ² TRIP requires a commercial key.\r\n\r\n## Features (full catalogue)\r\n\r\n| Capability | Source | Typical scenario |\r\n|---|---|---|\r\n| Topic / drug / disease search | All | Build the published-evidence base |\r\n| Review-type filter | All | `systematic-review` / `meta-analysis` / `rct` / `case-report` |\r\n| **Cochrane retrieval (focus)** | Europe PMC | `--cochrane` → restrict to the Cochrane Database of Systematic Reviews (verified journal filter, shared with meta-analysis) |\r\n| Year-range filter | All | Focus on recent evidence |\r\n| Safety / CSM tagging | All | Every work is flagged `is_safety` (amber-highlighted in the Works sheet) when its title/abstract mentions AE / PV / toxicity — useful for quick scanning. **The standalone Safety-Related sheet is opt-in** (`--safety`); a plain search does NOT emit it by default. |\r\n| Multi-source merge + dedupe | normalize | One unified list, DOI/title de-duped, provenance kept |\r\n| Citation ranking | OpenAlex / S2 | Most influential works |\r\n| MeSH terms | Europe PMC | Biomedical concept indexing |\r\n| Concepts / Keywords / Funders | OpenAlex | Topic classification + COI signals |\r\n| PubMed/PMC ID, OA full-text URL, complete abstract | All | Direct links + full evidence preservation |\r\n| Structured output | — | JSON + Markdown + Excel workbook (`excel_style`). Default sheets: README → Overview → Works → Evidence Log. `Safety-Related` sheet appears **only with `--safety`**. |\r\n| Chained invocation | — | → `ct-pipeline` / `ct-protocol` / `ct-csr` |\r\n| Resilient fetch (retry + backoff) | All | Honors `Retry-After`; OpenAlex Bearer via key |\r\n| Safe link rendering | All | `_normalize_link()` sanitises every hyperlink |\r\n| Citations + BibTeX/RIS | All | `--citation-style` (apa/nature/vancouver/ieee/gb7714) + `--export-bib` |\r\n| PRISMA screening funnel | All | `--prisma` deterministic rule screen → SVG funnel in HTML |\r\n| **Doc-type exclusion** | All | 双向类型甄别（`doc_type_filter`，移植自 meta-analysis 实证规则）：① 只要原创研究 → `--original-only`（排除 review/guideline/protocol，unknown 放行）；② **反向模式** → `--only-type review` / `--only-type guideline,protocol` 等（只保留指定类型，unknown 也排除）。判别依据：标题/摘要信号 + pubType 元数据；PRISMA 漏斗记录 `non-original-type:*` / `type-not-required:*`。也可单独用 `scripts/doc_type_filter.py`（`--exclude-non-original` / `--only TYPES`，支持 `--use-pdfs` PDF 级兜底）。甄别异常一律放行，绝不阻断检索流程 |\r\n| Relevance scoring | All | `--rank relevance` → `relevance_score` (title .6 + abstract .4) |\r\n| Obsidian / Zotero integration | All | `--obsidian` notes + MOC; `--zotero` CSV/RIS |\r\n| **P0 · Citation verification** | All | Anti-hallucination (§17.1). `--verify {all\\|top\\|none}`; source-aware skip; DOI cross-checked via doi.org; title/author consistency vs Crossref/Europe PMC/OpenAlex; flags `verified/bot_blocked/mismatch/unresolved/...` |\r\n| **P1 · DOI audit** | All | `--verify-dois` (opt-in, needs `--run`): writes additive `doi_verified` / `doi_mismatch` onto every work and emits `doi_mismatch.md` — a human-review list isolating the real-but-wrong / hallucinated-DOI case (Crossref title+author fuzzy match). Standalone: `scripts/verify_dois.py --in .merged.json --run` |\r\n| **P0 · Evidence provenance log** | All | `evidence_log.json/.md` + workbook sheet + HTML block: query→source→hits→retrieved_at→verification rate |\r\n| **P1 · PROSPERO registry** | Review register | `--with-prospero` (opt-in, key-gated, **reserved**) — dormant no-op skip without token; never claimed functional |\r\n| **G · Guideline corpus** | Guideline orgs | `--with-guidelines` → local pointer corpus (see above) |\r\n| **P0 · 结构化 PICO 概念** | topic/干预/对照/结局 | `--concept TYPE=VALUE`（可重复，TYPE∈{condition,population,intervention,comparator,outcome,endpoint,study_design,biomarker,drug,indication}）：概念 AND 拼入检索式、独立成轴；MeSH 精确归一化 + `mapping_status` 审计（exact_label/synonym/unmapped_literal，模糊匹配拒绝、未命中保留原词） |\r\n| **P0 · 检索深度预算** | resource control | `--depth quick\\|standard\\|deep`：决定每源上限与弱化结果救援开关（**非证据质量评级**）；不传则沿用 `--max` 旧语义 |\r\n| **P0 · 证据维度分轨** | safety vs efficacy | 合并后每篇打 `evidence_dimension`（=`safety` 若 `is_safety`，否则 `general`）；`--safety` 触发独立 safety lane，安全性信号不被疗效文献稀释 |\r\n| **P1 · 四态审计 ledger** | provenance | `evidence_log` 新增 `status∈{ok,empty,error,skipped}` + `coverage∈{healthy,partial,empty,critical_gap}`；被显式关闭的源记 `skipped`（not_run）、缺口诚实呈现、**不归零** |\r\n| **P1 · lane 规划视图** | dry-run | SAFE PREVIEW 直接打印 lane 规划（purpose / 源 / evidence_dimension / 备注），跑前可审查 |\r\n| **P1 · 弱化结果救援** | recall rescue | `standard`/`deep` 且合并唯一文献 < 5 且概念曾收窄检索式时，放宽至主题级回补一轮（单次、非致命；无新增记 skipped） |\r\n\r\n## Unified work schema\r\n\r\n```\r\n{ source, id, title, authors, year, publication_date, publication, journal_iso,\r\n type, study_type, cited_by_count, url, open_access_url,\r\n pmid, pmcid, doi, abstract_snippet, mesh, concepts, keywords, funders,\r\n language, is_retracted, is_safety, is_preprint, volume, issue, page,\r\n affiliations, sources }\r\n```\r\n\r\n## Output (full detail)\r\n\r\n**Standard deliverables = HTML + Excel only.** Present `lit_report.html` to the user as the result, and remind them of the optional add-ons below (the report itself shows a \"还能做什么 / Options\" strip with these hints).\r\n\r\n- Per-source payloads: `openalex.json` / `europepmc.json` / `semantic_scholar.json` / `biorxiv.json` / `medrxiv.json` / `arxiv.json` (enabled only)\r\n- `lit_report.xlsx` — Excel delivery (`excel_style`; `--no-xlsx` to skip): Overview → Literature master → Safety-related, KPI cards, charts, `is_safety` amber highlighting. **The complete result — user can keep filtering / pivoting on top of it.**\r\n- `lit_report.html` — self-contained HTML report (inline CSS, offline; `--no-html` to skip); inline-SVG PRISMA funnel when `--prisma`. **Default deliverable to open.**\r\n- **On-demand citation formats** (default OFF — ask the user or use `--export-bib`): `references.bib` / `references.ris` / `references_