← 返回 Skills 市场
fubian-ai

Lifescience Target Intelligence

作者 FuBian-AI · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ⚠ suspicious
90
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install lifescience-target-intelligence-internal
功能描述
INTERNAL SKILL — invoked by lifescience-meta-router-internal only. Not for direct user invocation.
使用说明 (SKILL.md)

Routing Criteria

Competitive landscape analysis for biological targets in drug development. Trigger when: the query focuses on a biological target as the primary subject — receptors, kinases, enzymes, ion channels, immune checkpoints, oncogenic mutations acting as drug targets (e.g., EGFR, HER2, KRAS, KRAS G12C, PD-1, PD-L1, CDK4/6, GLP-1, BTK, PARP, TROP2, VEGF, IL-6R, TNF-α, PCSK9, SGLT2, JAK1/2, mTOR, FGFR, RET, MET, TIGIT, LAG-3) — and asks about: which companies or drugs are competing in this target space, pipeline overview across a target class, clinical progress for all drugs targeting X, patent landscape for a target, first-in-class vs best-in-class comparison, Red Ocean vs White Space assessment, combination therapy landscape, or technology modality trends. Also triggers for: "which companies are developing X inhibitors/antibodies", "how many drugs target X", "compare all X inhibitors in Phase 3", "what is the competitive landscape for X", "who are the leaders in X space", "latest therapeutic interventions for X", "what drugs are available for X mutation", "treatment landscape for X mutation", "combination strategies for X inhibition", "best-in-class emerging therapies for X", "target validation for X", "GO/NO-GO recommendation for X target". Zone 3 (Scientific Intelligence) — Tier P + Tier S co-equal. Tier P and Tier S presented in separate sections. NOT for: individual drug deep-dive (use lifescience-pharmaceuticals-exploration-internal), company-level pipeline overview (use lifescience-company-profiling-internal), or standalone patent FTO/expiration analysis (use lifescience-patent-intelligence-internal).

Target Intelligence

Zone 3 — Tier P + Tier S co-equal. Tier P and Tier S data presented in separate sections; never mixed in the same table row.

Role

Senior analyst specializing in target-level competitive intelligence. Focus areas:

  • Competitive landscape: what drugs/companies are competing in this target space?
  • Clinical progress: what trials are ongoing? what are the outcomes?
  • Patent position: who holds key patents? what are the technology trends?
  • Target validation: GO/NO-GO recommendation with quantitative scoring
  • Strategic insights: Red Ocean vs White Space identification

Data Collection

Search → Fetch pattern is mandatory.

Tier P (Primary — Patsnap MCP)

Step Tool Purpose
1 ls_target_fetch Confirm target identity, biology, pathway
2 ls_paper_searchls_paper_fetch Development history, review literature
2b hybrid_search(sources=["paper"]) High-impact literature supplement: use filters={cited_min:50} for seminal papers, or search_strategy=["semantic"] for conceptual cross-domain exploration
3 ls_drug_searchls_drug_fetch All drugs targeting this; use drug_type filter for modality breakdown
4 ls_drug_deal_searchls_drug_deal_fetch BD transactions in this target space
5 ls_clinical_trial_searchls_clinical_trial_fetch Clinical progress using DrugIDs from Step 3
6 ls_clinical_trial_result_searchls_clinical_trial_result_fetch Trial outcomes including failed trials
7 ls_patent_searchls_patent_fetch Patent landscape by core type and technology
8 ls_patent_vector_search Semantic patent fallback for novel technology areas
9 ls_news_vector_searchls_news_fetch Recent trial readouts, competitive moves
10 ls_antibody_antigen_search Antibodies against this target (biology-modality MCP); use for antibody/bispecific/ADC modality queries
11 ls_web_search Commercial pricing, reimbursement, ICER (for approved drugs or Phase 3 candidates)

Tier S Supplement (Curated Scientific — present in separate section)

For target biology and validation evidence not in Patsnap:

import requests

# UniProt — protein function, expression, disease associations
r = requests.get(f"https://rest.uniprot.org/uniprotkb/{uniprot_id}.json")

# STRING — protein-protein interactions (physical + functional)
r = requests.get("https://string-db.org/api/json/network", params={
    "identifiers": "EGFR",
    "species": 9606
})
# Interpret: escore (experimental), dscore (database), tscore (text-mining)
# Physical interactions: escore > 0.4; functional: combined_score > 0.7

# BioGRID — curated protein interactions
r = requests.get("https://webservice.thebiogrid.org/interactions", params={
    "searchNames": "true",
    "geneList": "[gene_name]",
    "taxId": 9606,
    "format": "json",
    "accessKey": "[key]"
})

# ChEMBL — bioactivity data
r = requests.get("https://www.ebi.ac.uk/chembl/api/data/activity.json", params={
    "target_chembl_id": "[chembl_target_id]",
    "format": "json"
})

# OpenTargets — disease associations and tractability
r = requests.post("https://api.platform.opentargets.org/api/v4/graphql", json={
    "query": """
    query TargetTractability($ensemblId: String!) {
      target(ensemblId: $ensemblId) {
        tractability { label modality value }
        associatedDiseases { rows { disease { name } score } }
      }
    }
    """,
    "variables": {"ensemblId": "[ensembl_id]"}
})

# RCSB PDB — experimental structure availability
r = requests.get("https://data.rcsb.org/rest/v1/core/entry/[pdb_id]")
# PDBe — structure quality scores
r = requests.get(f"https://www.ebi.ac.uk/pdbe/api/validation/residuewise_outlier_summary/entry/{pdb_id}")

# DisGeNET — gene-disease associations with evidence scoring
r = requests.get("https://www.disgenet.org/api/gda/gene/[gene_id]", params={"format": "json"})

# GenCC — gene-disease validity (clinical evidence grading)
r = requests.get("https://search.thegencc.org/genes/[hgnc_id]")

Present Tier S data in a section labeled "Curated Scientific Data (Tier S)" with source attribution.


Analysis Framework

Target Validation Score (0–100)

When the user asks for a GO/NO-GO recommendation:

Dimension Weight Scoring Guidance
Disease Association 15% Genetic/GWAS evidence, expression data, animal models
Druggability 10% Target class, structural data, small molecule vs biologic tractability
Clinical Precedent 15% Approved drugs on target, clinical-stage assets, failure history
Competitive Landscape 10% Number of competitors, differentiation opportunity, FTO
Safety 15% On-target toxicities, normal tissue expression, knockout phenotype
Deal Activity 5% Recent deals validating target; deal values as market signal
Literature Evidence 5% Publication volume, KOL activity, conference trends
Pathway Context 10% Pathway position, redundancy risk, biomarker availability
Commercial Potential 15% Patient population, unmet need severity, pricing precedent

GO/NO-GO Thresholds:

  • 75–100: Strong GO — compelling target with strong validation
  • 50–74: Conditional GO — promising but gaps exist; specify conditions
  • 25–49: Caution — significant risks; needs more validation
  • 0–24: NO-GO — insufficient evidence or critical red flags

Competitive Landscape

For each drug in the competitive landscape:

  • Biological characteristics (indication, target, drug type, MoA)
  • Developer (company, region)
  • Clinical performance (key efficacy data: ORR, PFS, OS; safety data)
  • Failed/terminated trials: must state specific failure reasons

List by development stage: Approved → Phase 3 → Phase 2 → Phase 1.

Failed Trial Forensic Audit

For all terminated/failed trials, execute Four-Dimensional Audit:

  1. Target Engagement (TE): Was the mechanism properly engaged?
  2. Exposure Adequacy: Was drug concentration sufficient at target site?
  3. Patient Stratification: Was patient selection appropriate?
  4. Endpoint Design: Was the right endpoint measured?

Patent Landscape

Categorize patents by type and analyze technology direction evolution:

  • Identify dominant mechanisms/modality trends
  • Highlight emerging technology directions and early-mover patent holders
  • Do not simply list patent numbers — analyze trends

First-in-class / Best-in-class Analysis

Category Definition Analysis Points
First-in-class First drug to target this mechanism Timeline, current status
Best-in-class Superior efficacy/safety data Compare ORR, PFS, OS, safety
Fast-follower Me-too with differentiation Timing, differentiation strategy

Output

No mandatory template. Structure to best answer the specific question. Typical sections for a full target intelligence report:

Tier P section:

  1. Target rationale (biological function, disease association, pathway context)
  2. Development history (first approved drug, key milestones, major failures)
  3. Pipeline arena map (players by stage)
  4. Competitive positioning (leaders, challengers, followers)
  5. Clinical forensic analysis (failed trial audit)
  6. Patent landscape and technology trends
  7. Commercial assessment (pricing, reimbursement)
  8. Risk assessment matrix
  9. Strategic recommendations (R&D / BD / Investor)

Tier S section (if used):

  • Protein biology (UniProt)
  • Protein interactions — physical (BioGRID) and functional (STRING)
  • Experimental structure availability and quality (RCSB PDB, PDBe)
  • Bioactivity data (ChEMBL)
  • Disease associations and tractability (OpenTargets)
  • Gene-disease validity evidence (DisGeNET, GenCC)

For simple factual queries (e.g., "what drugs target EGFR"), return a concise direct answer.

Visual Output

Use templates from middleware/references/artifact-templates.md. Apply the three-layer model.

Layer A (top artifact — when ≥3 drugs retrieved):

  • Metric row: total drugs / approved / Phase 3 / most recent deal value
  • Card grid grouped by stage: 已批准Phase 3Phase 2Phase 1/早期
  • Card line 1: [Generic name] · [Company]; line 2: [Key differentiator — ORR/OS/modality/milestone]
  • Card color: by modality (middleware Principle 7 color table)
  • Chip row: modality distribution
  • BD highlights: 2–3 recent major transactions below grid
  • If market size data available: add bar chart (A2) for revenue/market trend

Layer B (markdown after artifact):

  • Competitive landscape table: drug / company / stage / key efficacy / key differentiator
  • Failed trial forensic table: trial / failure reason / dimension (TE / Exposure / Stratification / Endpoint)
  • GO/NO-GO scoring table (9 dimensions) — when validation query
  • Strategic recommendations: R&D / BD / Investor paragraphs

Layer C (inline in Layer B prose):

  • Stage progress strip (C1) when describing individual drug's current stage
  • Score gauge bar (C2) when citing a GO/NO-GO dimension score
  • Region badge row (C4) when describing approval/reimbursement status
  • Delta indicator (C5) when citing efficacy data (ORR %, LDL-C reduction %)

Patent landscape query → Layer A: SVG timeline (A3) with filing density by year and technology swim lanes. Layer B: patent trend analysis in markdown.

GO/NO-GO only → skip Layer A card grid; use Layer A metric row (total score) + Layer B scoring table only.


Related Analysis

Topic Skill
Specific drug ADMET, PK/PD, safety lifescience-pharmaceuticals-exploration-internal
Disease treatment landscape, SoC lifescience-disease-investigation-internal
Company pipeline, patents, deals lifescience-company-profiling-internal
Patent FTO, expiration, litigation lifescience-patent-intelligence-internal
Regulatory pathway, approval odds lifescience-regulatory-analysis-internal
Market size, revenue, pricing lifescience-commercial-analysis-internal
Biomarker, companion diagnostics lifescience-biomarker-analysis-internal

skill_zone: 3
tier_policy: "P + S co-equal (separate sections)"
version: "5.0.0"
parent_middleware: "lifescience-middleware-internal"
安全使用建议
This skill appears to be a legitimate target‑intelligence workflow, but it has important packaging gaps. Before installing or enabling it: 1) Confirm whether your agent runtime already has access to the Patsnap MCP tools (ls_*). If not, the skill will fail or may attempt network calls without declared credentials. 2) Ask the publisher to declare required credentials/config (Patsnap API credentials, BioGRID accessKey, any institutional proxies) so you can vet and scope them. 3) Clarify the 'INTERNAL' vs 'user-invocable' mismatch — if this is restricted IP or sensitive internal tooling, ensure only authorized agents/users can invoke it. 4) Consider restricting autonomous invocation for this skill until you validate credentials and data‑access policies, because it will perform outbound requests to many external services and may retrieve commercial or sensitive data. 5) If you must proceed, run it in a monitored environment first and verify no unexpected endpoints or credentials are accessed.
功能分析
Type: OpenClaw Skill Name: lifescience-target-intelligence-internal Version: 1.0.0 The skill bundle 'lifescience-target-intelligence-internal' is a well-structured tool designed for professional life sciences analysis. It provides detailed instructions for an AI agent to perform competitive intelligence on biological targets using a specialized MCP server and several reputable scientific APIs (e.g., UniProt, STRING, OpenTargets). The files, including SKILL.md and its references, focus on data collection, target validation scoring, and generating structured reports with Mermaid visualizations. No indicators of malicious intent, data exfiltration, or unauthorized command execution were identified.
能力评估
Purpose & Capability
The SKILL.md documents a coherent target‑intelligence workflow (calls to many ls_* Patsnap MCP tools and public bioinformatics APIs), which matches the advertised purpose. However, the registry metadata lists no required credentials/config even though the workflow clearly depends on an enterprise MCP (Patsnap) and may need API keys for some public services — this mismatch is a packaging/configuration issue that affects trustworthiness.
Instruction Scope
Runtime instructions direct the agent to run many external searches/fetches (ls_* MCP tools) and to call public REST APIs (UniProt, STRING, ChEMBL, OpenTargets, BioGRID, RCSB PDB, etc.). Those actions are within scope for target intelligence, but they necessarily perform outbound network requests and fetch possibly paywalled or sensitive commercial data. The doc includes placeholders for access keys (e.g., BioGRID), yet does not instruct where to obtain them or declare them in the skill manifest.
Install Mechanism
Instruction-only skill with no install spec and no code files — lowest install risk. Nothing in the manifest attempts to download or write code to disk.
Credentials
The skill expects access to a proprietary Patsnap MCP server and many ls_* tools (commercial connectors) and includes example calls that require API keys, but requires.env and primary credential are empty. That is disproportionate: a skill that relies on external commercial APIs should declare required credentials/config paths. Also the skill is labeled 'INTERNAL' but the registry flags it as user-invocable; this raises policy/authorization concerns.
Persistence & Privilege
No 'always: true' or other privileged persistence is requested. The skill does not request system-level changes or modify other skills' configs. Autonomous invocation is allowed (platform default) but not combined here with other privilege escalations.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install lifescience-target-intelligence-internal
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /lifescience-target-intelligence-internal 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.0
lifescience-target-intelligence-internal v1.0.0 — Initial Release - Provides internal-only, target-level competitive intelligence for drug development. - Triggers on queries about competitive landscape, clinical progress, patent position, validation, or strategic context for a biological target (e.g., EGFR, KRAS, PD-1). - Separately presents Tier P (primary competitive data) and Tier S (curated scientific evidence). - Mandates "search → fetch" data gathering pattern for rigorous, reproducible analysis. - Includes framework for GO/NO-GO recommendations with quantitative scoring and a structured failed trials forensic audit. - Not intended for drug deep-dives, company pipelines, or stand-alone patent FTO.
元数据
Slug lifescience-target-intelligence-internal
版本 1.0.0
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 1
常见问题

Lifescience Target Intelligence 是什么?

INTERNAL SKILL — invoked by lifescience-meta-router-internal only. Not for direct user invocation. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 90 次。

如何安装 Lifescience Target Intelligence?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install lifescience-target-intelligence-internal」即可一键安装,无需额外配置。

Lifescience Target Intelligence 是免费的吗?

是的,Lifescience Target Intelligence 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Lifescience Target Intelligence 支持哪些平台?

Lifescience Target Intelligence 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 Lifescience Target Intelligence?

由 FuBian-AI(@fubian-ai)开发并维护,当前版本 v1.0.0。

💬 留言讨论