swarm-ai-safetyOfficial·15 Agent Skills Included

swarm

Simulate multi-agent risks and test governance interventions

Simulates multi-agent ecosystems to measure emergent risks like deception, collusion, and adverse selection using probabilistic safety metrics. Runs governance experiments with taxes, audits, circuit breakers, and staking to compare safety interventions across reproducible seeded scenarios. Includes skills for parameter sweeps, statistical analysis, plotting, and paper scaffolding to turn raw simulation runs into rigorous research findings.

39|4

↗

npx skills add swarm-ai-safety/swarm --all -g -y

Available:

Directs the agent on how to run scenarios, sweeps, and red-team tests, select the right research persona for each task, and follow reproducibility and coordination protocols when working in this repository.

All Skills in This Repository (15)

Pure Emerald Level Indicators

📦 In Repo

swarm-ai-safety

swarm-safety

Simulate multi-agent AI systems to study distributional safety and emergent risks.

statistical-analysis

Perform pairwise t-tests, Cohen's d, and Bonferroni corrections on SWARM experiment CSV data.

plotting

Generate publication-quality plots from SWARM simulation sweep and time-series data.

parameter-sweep

Execute parameter grid sweeps across SWARM simulations to measure emergent risk metrics.

run-scenario

Execute SWARM simulation scenarios and export history.json and CSV metrics.

paper-writing

Scaffold markdown research papers from SWARM run data with methods and results tables.

regression-check

Re-run pytest suites and compare performance metrics against historical baselines.

verify

Run vault integrity checks for schema, evidence, wiki-links, and claims.

session-close

Summarize session changes, update memory logs, and commit code with Git.

run-query

Query run-index.yaml and vault claims for experiment metadata by tag, date, type, or claim.

vault-init

Initialize a SWARM Research OS vault with directory structure and templates.

sanity-check

Validate multi-agent simulation scenario configs with short runs and metric checks.

Official

Intermediate

Frequently Asked Questions

FAQPage Schema

How to install SWARM?▼

Run `npx skills add swarm-ai-safety/swarm --all -g -y` in your terminal to install all skills in this suite globally.

What is SWARM used for?▼

SWARM is a research framework for measuring emergent failures in multi-agent systems, such as deception, collusion, and adverse selection, even when individual agents appear safe.

How do I run a SWARM simulation?▼

Use the CLI with `swarm run scenarios/baseline.yaml --seed 42` or the Python API by registering agents with the Orchestrator and calling run().

Can SWARM test governance mechanisms?▼

Yes. You can stress-test interventions like transaction taxes, circuit breakers, audits, staking, and collusion detection against adaptive and deceptive agents.

Do I need coding experience to use SWARM?▼

Basic Python helps for custom scenarios, but pre-built scenario YAML files, CLI commands, and a no-setup Colab notebook let you run experiments quickly.