Ready to supercharge your AI agents?

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

Get Started Free →

Sensitivity Analysis · yogsoth-ai | DecimalAI

yogsoth-ai/convergence-sensitivity-analysis

SkillScore86Maintained

Tests conclusion robustness by perturbing parameters and observing rank changes. Shared across scoring, portfolio, and steel-manning campaigns.

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-convergence-sensitivity-analysis/SKILL.md

decimalai skills pull yogsoth-ai-convergence-sensitivity-analysis --out .claude/skills/

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

curl -fsSL --create-dirs -o .claude/skills/yogsoth-ai-convergence-sensitivity-analysis/SKILL.md https://app.decimal.ai/s/yogsoth-ai-convergence-sensitivity-analysis/SKILL.md

Raw file: /s/yogsoth-ai-convergence-sensitivity-analysis/SKILL.md ↗ · machine-readable summary at /s/yogsoth-ai-convergence-sensitivity-analysis.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

22 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

—-43% tokens0% turns86% of whole cases pass with skill (19/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-04	✗→✓	▲ Improved	-16%	0%
case-05	✗→✓	▲ Improved	-54%	0%
case-07	✗→✓	▲ Improved	-70%	0%
case-11	✗→✓	▲ Improved	-70%	0%
case-12	✗→✓	▲ Improved	-70%	0%

Showing 5 of 22 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/convergence-sensitivity-analysis/SKILL.md

View on GitHub ↗

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

Skill contents

SKILL.md

Sensitivity Analysis

Tests whether convergence conclusions are robust to parameter perturbation.

Execution

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

Why Subagent

Sensitivity analysis requires systematic exploration of parameter space with fresh perspective on each perturbation. Dedicated context prevents confirmation bias toward the original result.

HARD-GATE

Must perturb ≥3 parameters. A single-parameter sensitivity check is insufficient for robustness claims.

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/convergence-sensitivity-analysis would activate on your phrasing? Try it before you install — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/22/2026, 08:26 UTC · 22 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

+64%

86% vs 23%

Agent turns

0%

1.0 vs 1.0

Tokens

-43%

1,143.78 vs 1,994.78

fail → pass · 14pass → pass · 5fail → fail · 3

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-03 I want to test the robustness of our algorithm's convergence behavior before we publish our benchmark results. Please run a perturbation analysis across our pipeline parameters to confirm whether the convergence is stable, and output a structured analysis report.	fail→fail	22,371	4,835	-78%	1	1	0%	4,087	473	-88%	0	0	—
case-04 We ran a Monte Carlo power flow simulation that converged after 15 iterations. A colleague suggested just changing the load multiplier from 1.0 to 1.05 to verify robustness. Is testing only that single parameter adequate to declare our convergence conclusions robust?	fail→pass	14,563	10,512	-28%	1	1	0%	2,194	1,851	-16%	0	0	—
case-05 We are evaluating a fluid dynamics solver's numerical stability after steady-state convergence. Why should we delegate parameter perturbation tasks to a fresh subagent context rather than running all parameter sweeps directly in our main conversation history?	fail→pass	14,493	6,106	-58%	1	1	0%	2,133	981	-54%	0	0	—
case-06 Our neural network training loss stabilized at step 50,000. I want to check whether this convergence is sensitive to learning rate and batch size. I plan to perturb just these two parameters. Will this meet standard sensitivity check requirements for convergence robustness?	pass→pass	14,043	4,557	-68%	1	1	0%	2,139	926	-57%	0	0	—
case-07 We are setting up a sensitivity sweep for a finite-element mesh solver that reached thermal equilibrium. We have five potential physical parameters to test (thermal conductivity, boundary heat flux, convection coefficient, radiation emissivity, material density). What is the minimum number of these parameters we must perturb in our analysis?	fail→pass	23,400	1,936	-92%	1	1	0%	1,656	502	-70%	0	0	—
case-08 We are configuring PyTorch's `torch.optim.lr_scheduler.ReduceLROnPlateau` scheduler during baseline model training. We want the learning rate to reduce when validation loss stops decreasing. Which keyword parameter controls the number of non-improving epochs allowed before triggering a rate reduction, and does this scheduler require spawning a subagent?	pass→pass	3,507	3,650	+4%	1	1	0%	590	741	+26%	0	0	—
case-01 Our optimization model just reached a steady state. I need you to run a sensitivity analysis on our simulation setup to verify that this convergence holds up when we tweak the underlying input variables. Please evaluate the robustness and give me a summary report of the results.	fail→fail	14,231	8,929	-37%	1	1	0%	2,331	592	-75%	0	0	—
case-02 We've reached convergence in our economic equilibrium pipeline. Could you perform a parameter sensitivity check to ensure our stability conclusions aren't fragile to minor system variations? Provide a detailed breakdown of how the convergence behaves under perturbation.	fail→fail	18,971	39,953	+111%	1	1	0%	3,112	6,011	+93%	0	0	—
case-09 We are configuring scikit-learn's `RidgeCV` to evaluate L2 regularization strength during regression model fitting. Which keyword parameter receives the candidate regularization values, and must this cross-validation evaluation satisfy a three-parameter perturbation minimum?	pass→pass	6,083	8,022	+32%	1	1	0%	994	1,481	+49%	0	0	—
case-10 We are using `scipy.optimize.minimize` with the BFGS algorithm to find the minimum of an unconstrained objective function. Which parameter in the solver options sets the termination tolerance for the gradient norm, and does running this optimizer require delegating step execution to a spawned subagent?	pass→pass	5,063	4,808	-5%	1	1	0%	849	978	+15%	0	0	—
case-11 In our climate simulation pipeline, we want to run a sensitivity analysis on our converged sea-surface temperature predictions. Should we perform this analysis inline in our ongoing main session context or delegate it to a spawned subagent?	fail→pass	12,407	2,711	-78%	1	1	0%	1,955	591	-70%	0	0	—
case-12 We have an agent-based economic simulation that reached a stable price distribution. I want to test sensitivity by perturbing interest rate and inflation rate. Should I mark this stability analysis as hard-gate compliant?	fail→pass	14,224	3,198	-78%	1	1	0%	2,198	655	-70%	0	0	—
case-13 When running a parameter perturbation analysis on a structural engineering finite element model that reached equilibrium, what subagent tool or SOP should be invoked to execute the work?	fail→pass	16,268	3,138	-81%	1	1	0%	2,506	629	-75%	0	0	—
case-14 Our molecular dynamics trajectory reached energy minimization. We plan to perturb atomic mass, temperature, timestep, and cut-off radius to test convergence robustness. Does this plan fulfill the minimum parameter requirement for robustness testing?	fail→pass	17,020	1,712	-90%	1	1	0%	2,533	468	-82%	0	0	—
case-15 We are auditing a convergence sensitivity report for a power grid simulation model. The report only perturbed the generator voltage setpoint and transformer tap ratio. What critical defect should be flagged in this audit report?	fail→pass	13,355	3,866	-71%	1	1	0%	1,955	794	-59%	0	0	—
case-16 We want to evaluate whether our traffic flow simulation's convergence to steady state throughput is sensitive to input variations. What step must be executed before analyzing the parameter perturbations?	fail→pass	7,497	4,330	-42%	1	1	0%	1,118	858	-23%	0	0	—
case-17 We are building an automated verification pipeline for converged control system models. If a user submits a sensitivity test request with perturbations for target velocity and dampening coefficient, how should the pipeline gate respond?	fail→pass	16,561	3,701	-78%	1	1	0%	2,618	765	-71%	0	0	—
case-18 Why does the standard procedure for sensitivity analysis on converged numerical models mandate using a separate subagent execution context rather than continuing in the main execution thread?	fail→pass	16,097	4,334	-73%	1	1	0%	2,268	777	-66%	0	0	—
case-19 We are validating a financial portfolio risk model that reached numerical convergence. We plan to perturb asset volatility, risk-free rate, correlation coefficients, and liquidity discount. Does this analysis meet the minimum parameter threshold and how should execution be delegated?	fail→pass	17,257	3,943	-77%	1	1	0%	2,579	885	-66%	0	0	—
case-20 An analyst claims that checking sensitivity against a single input variable, discount rate, is sufficient to claim robustness for a converged discounted cash flow model. Is this claim valid under standard sensitivity analysis execution rules?	pass→pass	13,374	3,757	-72%	1	1	0%	1,968	680	-65%	0	0	—
case-21 We have converged a CFD simulation for aerodynamics and need to run a sensitivity analysis across air density, viscosity, angle of attack, and surface roughness. Which execution mechanism must be used to perform this task?	fail→pass	14,878	2,176	-85%	1	1	0%	2,197	517	-76%	0	0	—
case-22 In a convergence robustness audit for an agent-based supply chain model, what specific rule governs the parameter perturbation count?	fail→pass	27,998	1,810	-94%	1	1	0%	1,126	424	-62%	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. 22 cases were attempted, and 18 counted toward the lift figure. The other 4 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 +64 percentage points is the difference between those two pass rates over the 18 comparable cases. 2 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: 22.7%
Pass rate with the skill: 86.4%
Cases attempted: 22
Cases counted toward lift: 18
Cases that regressed: 2
Skill version measured: v1
Run completed: 8/22/2026, 08:26 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-04

Prompt

We ran a Monte Carlo power flow simulation that converged after 15 iterations. A colleague suggested just changing the load multiplier from 1.0 to 1.05 to verify robustness. Is testing only that single parameter adequate to declare our convergence conclusions robust?

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 run86

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-26% 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.