Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Get Started Free →

Causal Claim Extraction · yogsoth-ai | DecimalAI

yogsoth-ai/causal-claim-extraction

SkillScore61Maintained

Extract all causal claims (X causes Y, X leads to Y, X enables Y) from an artifact, producing a structured list of cause-effect pairs.

researchby yogsoth-ai

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Research pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/yogsoth-ai-causal-claim-extraction/SKILL.md

decimalai skills pull yogsoth-ai-causal-claim-extraction --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/yogsoth-ai-causal-claim-extraction/SKILL.md https://app.decimal.ai/s/yogsoth-ai-causal-claim-extraction/SKILL.md

Raw file: /s/yogsoth-ai-causal-claim-extraction/SKILL.md ↗ · machine-readable summary at /s/yogsoth-ai-causal-claim-extraction.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

23 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 7d ago

Yes — +35% lift-2% tokens0% turns61% of whole cases pass with skill (14/23)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-07	✗→✓	▲ Improved	4%	0%
case-10	✗→✓	▲ Improved	-30%	0%
case-11	✗→✓	▲ Improved	41%	0%
case-12	✗→✓	▲ Improved	6%	0%
case-14	✗→✓	▲ Improved	17%	0%

Showing 5 of 23 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

yogsoth-ai / de-anthropocentric-research-engine

skills/causal-claim-extraction/SKILL.md

View on GitHub ↗

License Apache-2.0Imported 8/3/2026Version v1

Skill contents

SKILL.md

Causal Claim Extraction

Extracts all explicit and implicit causal claims from an artifact.

Execution

Subagent — spawned via subagent-spawning/spawn-agent.

Why Subagent

Causal claim extraction requires careful linguistic analysis of the entire artifact. Isolated context prevents premature evaluation of claims.

Input

artifact: The artifact to analyze
artifact_type: Type of artifact (gap, hypothesis, claim, etc.)

Output

causal_claims: List of {cause, effect, strength, evidence, location}
causal_graph: Directed graph of cause-effect relationships
claim_count: Total number of causal claims found

Budget

One unit = one extraction pass per artifact.

Available SOPs

Optional, no fixed order; the final leaf is always a sop.

| SOP | When to use | | --- | --- | | spawn-agent | Spawn a customized CC subagent with full MCP tool access. Used by SOPs that declare execution: subagent. |

Want to see if yogsoth-ai/causal-claim-extraction would activate on your phrasing? Try it before you install — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/22/2026, 03:50 UTC · 23 cases

Lift

+35%

61% vs 26%

Agent turns

0%

1.0 vs 1.0

Tokens

-3%

1,450.83 vs 1,488.35

fail → pass · 8pass → pass · 6fail → fail · 9

Per-case results (23)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 Please review this post-mortem technical report for our infrastructure failure. Identify every causal link—both direct and implied—and output them with their cause, effect, relative strength, quote evidence, and section reference. Also include the overall count of claims and a directed network graph illustrating how causes lead to effects.	fail→fail	4,443	4,551	+2%	1	1	0%	669	928	+39%	0	0	—
case-02 We have a whitepaper detailing economic policy hypotheses regarding inflation and interest rates. Parse the artifact to extract all explicit or subtle cause-and-effect claims. Provide a structured output with the claim count, a directed cause-effect graph, and itemized records containing cause, effect, strength score, textual evidence, and location.	fail→fail	16,551	30,558	+85%	1	1	0%	3,372	5,684	+69%	0	0	—
case-03 We have a climate modeling paper 'Global Surface Temperature Trends 1980-2025'. Validate the statistical validity of the claimed correlation between aerosol emissions and ocean cooling. Run a regression diagnostic or assess if the statistical methodology supports true causality rather than mere correlation.	pass→pass	20,694	26,348	+27%	1	1	0%	3,155	4,414	+40%	0	0	—
case-04 Here is the Q3 Strategic Strategy Briefing for Apex Corp. Estimate the ROI and potential financial impact if Apex Corp had chosen project B instead of project A last quarter. Perform a counterfactual scenario analysis on revenue.	fail→fail	8,292	11,213	+35%	1	1	0%	1,489	1,981	+33%	0	0	—
case-05 Analyze the customer survey document '2024 Enterprise Feedback Summary'. Classify the overall sentiment (positive, neutral, negative) and highlight top customer feature requests.	fail→fail	2,757	2,322	-16%	1	1	0%	437	531	+22%	0	0	—
case-06 Analyze the product roadmap document 'SaaS Gateway v2 Gap Analysis'. We want to pull out every cause and effect statement. Will you process this directly in your primary context window?	fail→fail	1,982	5,275	+166%	1	1	0%	329	1,057	+221%	0	0	—
case-07 Extract causal assertions from the clinical trial manuscript 'Phase 3 Clinical Trial of Compound X'. Why should this analysis be handed off to a separate worker instance?	fail→pass	10,631	8,334	-22%	1	1	0%	1,556	1,620	+4%	0	0	—
case-08 Process the medical research paper 'Beta-Blockers in Post-MI Management'. Extract all causal mechanisms and format each item. Should textual evidence and paragraph position be separated or combined into a general notes string?	pass→pass	11,401	14,967	+31%	1	1	0%	1,772	2,752	+55%	0	0	—
case-09 Parse the incident post-mortem 'Database Outage 2024-05-12'. We need to visualize how root causes cascaded into secondary failures. How should the relationships between causes and effects be formatted in the final output bundle?	pass→pass	9,638	10,497	+9%	1	1	0%	1,599	2,200	+38%	0	0	—
case-10 We are submitting a research artifact 'Dark Matter Null Detection Report' which presents several scientific hypotheses. What metadata classification key and value should accompany the artifact content during extraction setup?	fail→pass	7,099	3,363	-53%	1	1	0%	1,117	786	-30%	0	0	—
case-11 Extract all causal statements from 'Cybersecurity Incident Response Summary 2024'. In addition to the itemized list, how should the total number of identified causal links be reported?	fail→pass	9,794	12,249	+25%	1	1	0%	1,552	2,188	+41%	0	0	—
case-12 Extract the cause-effect relationships from 'Supply Chain Bottleneck Report Q2'. What exact key name in the output structure holds the list of extracted items?	fail→pass	6,134	4,723	-23%	1	1	0%	965	1,021	+6%	0	0	—
case-13 When analyzing the economic policy paper 'Monetary Expansion and Housing Prices', should the degree of certainty or impact of each cause-effect relation be recorded, and under what field name?	pass→pass	11,877	3,289	-72%	1	1	0%	2,004	780	-61%	0	0	—
case-14 We have a 50-page architecture review document 'Enterprise Cloud Migration Plan'. How many extraction passes per artifact are budgeted for this analysis?	fail→pass	3,384	1,798	-47%	1	1	0%	410	479	+17%	0	0	—
case-15 In the bug report 'Kernel Memory Leak Analysis', some causes are explicitly stated ('X caused Y') while others are implied ('following X, Y degraded'). Should implicit causal claims be included alongside explicit ones?	fail→fail	11,484	4,425	-61%	1	1	0%	1,659	879	-47%	0	0	—
case-16 Parse the system failure post-mortem 'Payment Service Outage'. What key in the JSON output schema holds the graph representation of causal links?	pass→pass	6,812	1,692	-75%	1	1	0%	1,095	477	-56%	0	0	—
case-17 We need to extract causal claims from 'Autonomous Vehicle Sensor Failure Log'. Which tool or SOP is designated to spawn the worker agent?	fail→pass	11,687	2,012	-83%	1	1	0%	1,651	566	-66%	0	0	—
case-18 In the legal brief 'Patent Infringement Defense Brief', how should the specific position of a causal claim within the text be represented in the structured output?	pass→pass	13,766	5,507	-60%	1	1	0%	2,267	1,082	-52%	0	0	—
case-19 Extract causal assertions from 'Epidemiological Study on Water Contamination'. What five specific fields must be populated for every record in the extracted list?	fail→pass	10,440	4,794	-54%	1	1	0%	1,568	1,024	-35%	0	0	—
case-20 We have a position paper titled 'Argument for Decentralized Energy Grids'. When structuring the input parameters for extraction, what value should be assigned to artifact_type?	fail→fail	4,342	3,633	-16%	1	1	0%	597	777	+30%	0	0	—
case-21 When generating the graph of cause and effect relationships for 'Chemical Reaction Kinetics Paper', should the graph edges be undirected or directed?	fail→fail	11,250	2,469	-78%	1	1	0%	1,742	581	-67%	0	0	—
case-22 When the extraction subagent is spawned for 'Fintech Security Audit Report', what tool access rights does the subagent possess?	fail→fail	9,087	4,903	-46%	1	1	0%	1,297	539	-58%	0	0	—
case-23 Why does the extraction protocol enforce spawning a subagent rather than evaluating the claims in the main conversational context for 'Global Logistics Delay Analysis'?	fail→pass	13,229	5,803	-56%	1	1	0%	1,930	1,023	-47%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +35 percentage points is the difference between those two pass rates over the 23 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 26.1%
Pass rate with the skill: 60.9%
Cases attempted: 23
Cases counted toward lift: 23
Cases that regressed: 0
Skill version measured: v1
Run completed: 8/22/2026, 03:50 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-07

Prompt

Extract causal assertions from the clinical trial manuscript 'Phase 3 Clinical Trial of Compound X'. Why should this analysis be handed off to a separate worker instance?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run61

Live evalNeeds ≥20 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns-2% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 7 days ago

Other measured skills in the registry, with their headline benchmark lift.