Ready to supercharge your AI agents?

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

Get Started Free →

Method Inventory · yogsoth-ai | DecimalAI

yogsoth-ai/method-inventory

NewSkillScoreMaintained

Comprehensively identify all relevant methods for a task — 50 methods, 60 web searches budget

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-method-inventory/SKILL.md

decimalai skills pull yogsoth-ai-method-inventory --out .claude/skills/

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

curl -fsSL --create-dirs -o .claude/skills/yogsoth-ai-method-inventory/SKILL.md https://app.decimal.ai/s/yogsoth-ai-method-inventory/SKILL.md

Raw file: /s/yogsoth-ai-method-inventory/SKILL.md ↗ · machine-readable summary at /s/yogsoth-ai-method-inventory.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 17d ago

—+63% tokens0% turns13% of whole cases pass with skill (3/23)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-04	✓→✓	＝ Same ✓	110%	0%
case-05	✓→✓	＝ Same ✓	127%	0%
case-06	✓→✓	＝ Same ✓	77%	0%
case-01	✗→✗	＝ Same ✗	-83%	0%
case-02	✗→✗	＝ Same ✗	15%	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 — 8/10/2026

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/method-inventory/SKILL.md

View on GitHub ↗

License Apache-2.0Imported 8/10/2026Version v1

Skill contents

SKILL.md

Method Inventory

Purpose

Systematically discover and catalog every relevant method (published, preprint, industry) that has been evaluated on the target task. Combines leaderboard scraping, citation chain traversal, and keyword-based academic search to ensure no significant method is missed.

Budget

| Resource | Floor | Target | |----------|-------|--------| | Methods discovered | 30 | 50 | | Web searches | 40 | 60 | | Papers consulted | 20 | 40 |

State Ledger

<HARD-GATE>
| Metric | Current | Target | Status |
|--------|---------|--------|--------|
| Methods discovered | 0 | 50 | BLOCKED |
| Web searches used | 0 | 60 | — |
| Papers consulted | 0 | 40 | — |
| Leaderboard sources | 0 | 5 | — |
| Citation chains traced | 0 | 10 | — |
</HARD-GATE>

Cannot exit until methods_discovered >= 40 (80% of target).

Available Tactics

leaderboard-harvesting — Scrape Papers With Code, benchmark-specific leaderboards, survey papers

Available SOPs

method-discovery — Core method identification via literature, leaderboards, citation chains

Execution Guidance

Start with leaderboard-harvesting tactic to get the well-known methods
Use method-discovery SOP with citation chaining to find less prominent methods
Search for survey papers and meta-analyses that enumerate methods
Check preprint servers for very recent unpublished methods
Verify each method has: name, year, authors, venue, key innovation
Deduplicate (same method with different names across papers)
Classify methods by family/approach (e.g., transformer-based, GNN-based, hybrid)

Output Format

json
{
  "task": "string",
  "domain": "string",
  "methods": [
    {
      "name": "string",
      "aliases": ["string"],
      "year": 2024,
      "authors": "string",
      "venue": "string",
      "family": "string",
      "key_innovation": "string",
      "paper_id": "string",
      "source": "leaderboard|paper|citation_chain|preprint"
    }
  ],
  "coverage_assessment": {
    "leaderboard_sources_checked": 0,
    "citation_chains_traced": 0,
    "survey_papers_consulted": 0,
    "confidence": "high|medium|low"
  }
}

Available Tactics

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

| Tactic | When to use | | --- | --- | | leaderboard-harvesting | Systematically collect performance data from platforms and papers |

Available SOPs

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

| SOP | When to use | | --- | --- | | method-discovery | Identify all relevant methods via literature, leaderboards, citation chains |

Illustrative Want to see what yogsoth-ai/method-inventory would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

9/4/2026, 04:19 UTC · 23 cases

Measured on 8 comparable cases — below the 22 needed to publish a lift, so this result is shown here but not on the registry card or leaderboard.

Lift

0%

13% vs 13%

Agent turns

0%

1.0 vs 1.0

Tokens

+63%

7,644.75 vs 4,683.63

pass → pass · 3fail → fail · 20

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 Can you conduct a comprehensive literature and leaderboard survey for 3D human pose estimation from monocular video? I need a complete catalog of published, preprint, and benchmarked methods across the field. Please output the results in JSON format with the task and domain at the top, an array of discovered methods—each specifying its name, any aliases, release year, author list, publication venue, architectural family, key innovation, paper reference ID, and where it was discovered (e.g., leaderboard, citation chain, paper, preprint)—and a final coverage evaluation block summarizing the number of leaderboards checked, citation chains traced, survey papers reviewed, and overall assessment confidence.	fail→fail	41,814	18,143	-57%	1	1	0%	8,323	1,452	-83%	0	0	—
case-02 I'm preparing a survey paper on molecular property prediction in drug discovery and need a full mapping of all machine learning methods that have been evaluated on this benchmark. Please assemble a structured JSON response containing the task and domain labels, a list of all identified methods—with details for each method's primary name, alternate names, publication year, key authors, venue, approach family, main novelty, paper ID, and discovery source type—followed by a coverage assessment section that tracks checked leaderboard sources, citation chains followed, survey papers consulted, and a confidence rating.	fail→fail	48,201	46,813	-3%	1	1	0%	8,301	9,508	+15%	0	0	—
case-03 We are launching a research effort on low-resource machine translation for African languages and require a systematic breakdown of every evaluated model, preprint, and industry baseline. Format the output strictly as JSON with top-level attributes for the task and domain, a methods array containing each technique's name, known aliases, year, authors, publication venue, method family classification, main innovation, paper ID, and where it was retrieved, plus a coverage assessment object documenting the count of benchmark leaderboards checked, citation chains pursued, survey papers analyzed, and final confidence level.	fail→fail	26,857	23,389	-13%	1	1	0%	4,802	1,961	-59%	0	0	—
case-04 We have trained five YOLOv8 model checkpoints and want to run latency and throughput benchmarks on an NVIDIA T4 GPU across batch sizes 1, 8, and 32. Provide python benchmarking scripts to execute these evaluations and log GPU memory usage.	pass→pass	27,205	46,618	+71%	1	1	0%	4,277	8,974	+110%	0	0	—
case-05 I am writing the background and literature review chapter for my PhD thesis on open-domain question answering. Please write a 1,000-word academic essay detailing the history and paradigm shift from sparse retrieval to dense retrieval and RAG architectures.	pass→pass	18,929	46,106	+144%	1	1	0%	3,952	8,973	+127%	0	0	—
case-06 We are creating a new evaluation benchmark for medical image segmentation. Design a dataset annotation guidelines document and quality control protocol for radiologist annotators scoring liver lesions.	pass→pass	30,638	42,572	+39%	1	1	0%	5,067	8,954	+77%	0	0	—
case-07 Compile a catalog of deep learning architecture methods evaluated on the LibriSpeech benchmark for automatic speech recognition. I need a JSON object mapping the domain, task, cataloged techniques, and survey coverage metrics. A summary of 10 popular baseline models will suffice for my initial presentation.	fail→fail	31,429	13,832	-56%	1	1	0%	5,376	1,294	-76%	0	0	—
case-08 Extract a JSON method catalog for tabular data classification on the OpenML benchmark. Include primary names, publication years, author list, venue, structural family, key innovation, paper ID, source type, and coverage summary. Format the author details as a list of author name strings.	fail→fail	20,822	15,707	-25%	1	1	0%	4,436	1,388	-69%	0	0	—
case-09 Generate a method mapping JSON for protein structure prediction on the CASP benchmark. Tag each method's origin with descriptive labels like 'PapersWithCode Leaderboard', 'ArXiv Preprint', or 'Primary Literature Citation'.	fail→fail	25,279	10,313	-59%	1	1	0%	3,972	1,396	-65%	0	0	—
case-10 Catalog evaluated methods for video action recognition on Kinetics-400 into a JSON payload. Format the discovery metrics as top-level properties named total_leaderboards_scraped and survey_confidence.	fail→fail	22,068	53,402	+142%	1	1	0%	3,679	8,963	+144%	0	0	—
case-11 Provide a JSON inventory of algorithms evaluated on the Arcade Learning Environment for reinforcement learning. Use standard descriptive fields such as known_aliases for alternative names and publication_year for the release year.	fail→fail	37,626	17,398	-54%	1	1	0%	7,171	1,194	-83%	0	0	—
case-12 Create a JSON catalog of machine learning techniques evaluated on the Cityscapes semantic segmentation benchmark. Name the method properties architectural_family to categorize models and main_innovation for the novel contribution.	fail→fail	18,020	17,928	-1%	1	1	0%	2,484	1,454	-41%	0	0	—
case-13 Construct a JSON database of architectures tested on SQuAD 2.0 question answering. Reference the literature using paper_reference for ArXiv or DOI identifiers.	fail→fail	28,715	56,148	+96%	1	1	0%	5,349	8,954	+67%	0	0	—
case-14 Conduct a method survey for object detection on MS COCO 2017. Deliver a JSON object with task, domain, methods array, and coverage assessment. Listing 15 canonical detectors like Faster R-CNN, YOLO, and DETR is sufficient.	fail→fail	33,587	19,413	-42%	1	1	0%	5,997	1,565	-74%	0	0	—
case-15 Construct a JSON literature inventory for time-series forecasting on the Electricity Transformer Temperature (ETT) benchmark. Express the coverage assessment confidence score as a percentage or numeric value like 0.90.	fail→fail	18,618	8,620	-54%	1	1	0%	3,886	1,297	-67%	0	0	—
case-16 Assemble a JSON survey of Graph Neural Network methods evaluated on the OGBG-MolHIV graph classification benchmark. Include a coverage metric section with field names leaderboards_checked, citation_chains_followed, and surveys_reviewed.	fail→fail	28,641	13,455	-53%	1	1	0%	4,826	1,176	-76%	0	0	—
case-17 Map out evaluated techniques for text-to-image synthesis on MS COCO in JSON format. Group task and domain descriptions inside a top-level metadata object.	fail→fail	20,838	18,358	-12%	1	1	0%	4,094	1,207	-71%	0	0	—
case-18 Build a JSON method inventory for vision transformers on ImageNet-1k. Format alternative names for each method as a comma-separated string in the aliases attribute.	fail→fail	49,042	13,373	-73%	1	1	0%	5,186	1,352	-74%	0	0	—
case-19 Survey recommendation system algorithms evaluated on the MovieLens-20M benchmark. Output a structured JSON catalog with top-level task and domain, methods list, and coverage evaluation. A quick list of 12 representative collaborative filtering models is enough.	fail→fail	24,343	30,241	+24%	1	1	0%	4,949	1,385	-72%	0	0	—
case-20 Inventory all machine learning models tested on GLUE benchmark for natural language understanding in JSON format. Label the publication target as publication_venue for each model.	fail→fail	32,045	17,900	-44%	1	1	0%	5,905	1,360	-77%	0	0	—
case-21 Gather evaluated methods for Named Entity Recognition on CoNLL-2003 into JSON format. Provide task, domain, methods list, and coverage metrics. Provide 8 to 10 prominent models.	fail→fail	17,848	34,504	+93%	1	1	0%	2,872	5,436	+89%	0	0	—
case-22 Catalog machine learning methods evaluated on the ADE20K semantic segmentation benchmark as JSON. Format release dates as text strings such as 'June 2021' in the year field.	fail→fail	12,746	14,823	+16%	1	1	0%	3,454	1,505	-56%	0	0	—
case-23 Survey code generation models evaluated on HumanEval benchmark. Output JSON with task, domain, methods, and coverage assessment. Return 15 key models.	fail→fail	27,371	16,819	-39%	1	1	0%	4,254	1,295	-70%	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, and 8 counted toward the lift figure. The other 15 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of 0 percentage points is the difference between those two pass rates over the 8 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 13%
Pass rate with the skill: 13%
Cases attempted: 23
Cases counted toward lift: 8
Cases that regressed: 1
Skill version measured: v1
Run completed: 9/4/2026, 04:19 UTC

SkillScore blends 4 signalsno signals yet

BenchmarkRun a verified benchmark with positive lift

Live evalNeeds ≥10 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+102% tokens

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

No signals yet — run a benchmark or collect evaluated traces to earn a score.

Verified on gemini-3.6-flash · re-tested 16 days ago

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