Ready to supercharge your AI agents?

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

Get Started Free →

Intersection Evaluation · yogsoth-ai | DecimalAI

yogsoth-ai/intersection-evaluation

NewSkillScoreMaintained

Evaluate exploration status of each cell in a method×problem matrix, annotating as explored, partial, or unexplored.

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-intersection-evaluation/SKILL.md

decimalai skills pull yogsoth-ai-intersection-evaluation --out .claude/skills/

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

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

Raw file: /s/yogsoth-ai-intersection-evaluation/SKILL.md ↗ · machine-readable summary at /s/yogsoth-ai-intersection-evaluation.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 18d ago

—+46% tokens0% turns17% of whole cases pass with skill (4/23)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-10	✓→✓	＝ Same ✓	3%	0%
case-11	✓→✓	＝ Same ✓	-28%	0%
case-12	✓→✓	＝ Same ✓	11%	0%
case-13	✓→✓	＝ Same ✓	5%	0%
case-21	✗→✗	＝ Same ✗	2%	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/intersection-evaluation/SKILL.md

View on GitHub ↗

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

Skill contents

SKILL.md

Intersection Evaluation

Evaluate the exploration status of each cell in a method×problem matrix and prioritize unexplored intersections.

Execution

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

Why Subagent

Evaluation requires careful assessment of each intersection's exploration depth, consulting literature and benchmarks. Benefits from focused context to maintain consistent evaluation criteria.

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. |

Illustrative Want to see what yogsoth-ai/intersection-evaluation 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

9/3/2026, 00:57 UTC · 23 cases

Measured on 18 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

-33%

33% vs 67%

Agent turns

0%

1.0 vs 1.0

Tokens

+46%

6,837.22 vs 4,695.89

pass → pass · 4fail → fail · 19

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-21 Compare 4 open-weights code LLMs against 7 programming language translation tasks. Assess research depth for every cell in this matrix and produce a prioritized list of open research intersections.	fail→fail	52,721	50,923	-3%	1	1	0%	8,225	8,385	+2%	0	0	—
case-04 We have five self-supervised visual representation algorithms and eight downstream medical imaging segmentation tasks. We need to identify which algorithm-task pairs remain untested in published research and rank them by research priority. Give me your initial evaluation directly in this turn if possible.	fail→fail	21,366	31,621	+48%	1	1	0%	2,940	4,790	+63%	0	0	—
case-01 I have a grid comparing four machine learning optimization algorithms against six standard reinforcement learning benchmarks. Can you assess the current research landscape for each algorithm-benchmark pair, figure out which combinations haven't been thoroughly tested yet, and give me a ranked list of the most promising unexplored pairings to try next?	fail→fail	20,045	104,904	+423%	1	1	0%	3,339	1,256	-62%	0	0	—
case-02 We're planning our next research cycle and need to map our candidate graph neural network architectures against our set of molecular property prediction tasks. Please evaluate the coverage depth across all method-task combinations based on existing literature, highlight where the major empirical gaps lie, and order the unstudied intersections by priority.	fail→fail	35,579	52,914	+49%	1	1	0%	5,882	6,950	+18%	0	0	—
case-03 I'm trying to organize a systematic literature survey on transformer compression techniques applied to different speech processing tasks. Could you analyze the matrix of techniques versus tasks to check how well each cell has been explored, identify the clear research gaps, and produce a prioritized list of combinations that warrant immediate experimental investigation?	fail→fail	51,822	49,516	-4%	1	1	0%	8,138	8,408	+3%	0	0	—
case-05 Our lab wants to map out 6 dense retrieval methods against 5 specialized legal and clinical document domain adaptation tasks. We need to analyze each intersection cell's coverage status and prioritize the unstudied gaps. Please carry out this grid evaluation now.	fail→fail	53,425	40,779	-24%	1	1	0%	8,238	8,398	+2%	0	0	—
case-06 We are building a research roadmap cross-referencing 4 variational quantum eigensolver variants against 7 molecular Hamiltonian simulation problems. Evaluate the research coverage for every intersection and rank the highest-value unexplored combinations.	fail→fail	92,955	64,454	-31%	1	1	0%	7,851	8,390	+7%	0	0	—
case-07 I need to map 5 reinforcement learning locomotion controllers across 6 rough-terrain simulation environments to find research gaps. Assess the exploration depth of each controller-terrain cell and output a prioritized list of open combinations.	fail→fail	23,592	58,192	+147%	1	1	0%	3,834	8,591	+124%	0	0	—
case-08 We are reviewing 4 diffusion-based audio architectures against 6 specific music synthesis subtasks such as polyphonic transcription and timbre transfer. Evaluate the literature matrix to spot unexplored intersections and rank them.	fail→fail	23,422	23,494	+0%	1	1	0%	3,773	906	-76%	0	0	—
case-09 We have 7 state-space time series models and 5 high-frequency industrial sensor datasets. Assess which model-dataset combinations have sufficient empirical validation in literature and highlight the highest priority unstudied pairings.	fail→fail	39,820	57,910	+45%	1	1	0%	3,667	8,390	+129%	0	0	—
case-10 I need a Python script using PyTorch to train a ResNet-18 model on CIFAR-10 and log accuracy metrics. Please provide the complete runnable script with data loading and training loop.	pass→pass	20,500	24,916	+22%	1	1	0%	4,652	4,798	+3%	0	0	—
case-11 I already have a CSV file containing numerical benchmark scores for a 5x5 matrix of models and tasks. Please write a Python snippet using Matplotlib to render this score matrix as a formatted heatmap image.	pass→pass	26,695	7,683	-71%	1	1	0%	2,420	1,737	-28%	0	0	—
case-12 Can you draft a Slurm batch script to launch a parallel grid search over learning rates [0.001, 0.0001] and batch sizes [32, 64] across 4 GPUs for our training pipeline?	pass→pass	13,810	30,011	+117%	1	1	0%	2,746	3,038	+11%	0	0	—
case-13 Can you provide a concise summary of the key findings and contributions from the 2023 survey paper on Large Language Model alignment techniques?	pass→pass	20,816	16,987	-18%	1	1	0%	2,145	2,258	+5%	0	0	—
case-14 We have 3 vision-language pretraining architectures and 5 multi-modal reasoning benchmarks. Assess the literature exploration status for each pair and rank the unstudied gaps to guide our next experiment.	fail→fail	26,698	73,445	+175%	1	1	0%	3,704	8,388	+126%	0	0	—
case-15 Our cybersecurity group is mapping 5 federated aggregation algorithms against 4 membership inference attack scenarios. Evaluate existing literature to determine cell-by-cell coverage and produce a ranked list of unexplored intersections.	fail→fail	39,551	48,120	+22%	1	1	0%	6,313	8,388	+33%	0	0	—
case-16 We are comparing 4 sequential recommendation algorithms against 6 e-commerce vertical datasets. Please review the empirical research coverage for each combination and highlight the top unexplored research opportunities.	fail→fail	25,799	42,980	+67%	1	1	0%	3,411	473	-86%	0	0	—
case-17 We want to study 5 neural network pruning methods across 4 embedded hardware accelerators. Evaluate the literature matrix to find which method-device pairs lack thorough empirical benchmarking.	fail→fail	14,635	43,356	+196%	1	1	0%	1,435	7,010	+389%	0	0	—
case-18 Map 6 offline RL algorithms against 8 MuJoCo continuous control benchmarks. We need an assessment of research coverage for each combination and a ranked list of the most critical gaps.	fail→fail	44,650	19,274	-57%	1	1	0%	8,225	1,864	-77%	0	0	—
case-19 We have 4 spatial graph convolution variants and 6 protein-protein interaction prediction tasks. Evaluate literature coverage across this grid and prioritize which unexplored cells to target in our lab's upcoming papers.	fail→fail	21,061	50,008	+137%	1	1	0%	3,315	8,388	+153%	0	0	—
case-20 Evaluate 5 automated NLG evaluation metrics across 6 specialized domain text corpora such as medical discharge summaries and legal contracts. Determine exploration status for each cell and order the research gaps.	fail→fail	19,389	52,702	+172%	1	1	0%	3,417	8,384	+145%	0	0	—
case-22 We have 5 adversarial defense mechanisms and 4 vision transformer variants. Assess which combinations have been benchmarked in existing papers and order the unexplored combinations by priority.	fail→fail	29,901	6,520	-78%	1	1	0%	4,010	594	-85%	0	0	—
case-23 Map 6 feature attribution explainability methods against 5 oncology image classification tasks. Evaluate literature coverage depth for each matrix cell and prioritize the unexplored pairs.	fail→fail	195,951	78,349	-60%	1	1	0%	5,604	8,379	+50%	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 18 counted toward the lift figure. The other 5 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 -33 percentage points is the difference between those two pass rates over the 18 comparable cases. 7 cases got worse with the skill loaded, and they are 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: 66.7%
Pass rate with the skill: 33.3%
Cases attempted: 23
Cases counted toward lift: 18
Cases that regressed: 7
Skill version measured: v1
Run completed: 9/3/2026, 00:57 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-1% 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 17 days ago

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