swarm-ai-safety
- 16 skills
- 0 followers
- 1 week ago last updated
- ▌ Paper Writing · swarm-ai-safety bundleScaffold research papers from SWARM run data with auto-populated tables
- ▌ Parameter Sweep · swarm-ai-safetyRun parameter grid sweeps across SWARM scenarios and generate summary statistics
- ▌ Statistical Analysis · swarm-ai-safety bundlePerform rigorous statistical analysis on SWARM experiment data with multiple-comparison corrections
- ▌ Session Close · swarm-ai-safetyEnd-of-session ritual — summarize what changed, what was learned, update memory, commit, and push. Ensures no work is lost and the next session has full context.
- ▌
- ▌ Experiment Loop · swarm-ai-safetyRun the full experiment lifecycle — design, sanity check, full sweep, synthesize, and claim review. Given a hypothesis, proposes sweep matrix, runs the experiment, generates artifacts, and updates the research log.
- ▌ Regression Check · swarm-ai-safetyRe-run the baseline suite and compare against the last known-good run. Flags metric drift, test failures, or behavioral changes.
- ▌ Run Scenario · swarm-ai-safetyExecute a SWARM simulation scenario and export standardized artifacts
- ▌ Run Query · swarm-ai-safetyQuery the run index and vault for experiment history. Search by tag, date, type, or claim. Returns run metadata and pointers without loading raw data.
- ▌ Synthesize · swarm-ai-safety bundleSynthesize a SWARM run into vault notes. Reads run.yaml and summary.json, generates an experiment note, checks existing claims for new evidence, updates the master index. Triggers on "/synthesize", "/synthesize [run_id]", "synthesize this run", "generate notes for run".
- ▌ Vault Init · swarm-ai-safety bundleInitialize or extend a SWARM Research OS vault. Creates directory structure, installs schema-enforced templates, generates master index, seeds starter notes, and optionally backfills run.yaml metadata for existing runs. Triggers on "/vault-init", "initialize vault", "set up research os vault", "create vault structure".
- ▌ Sanity Check · swarm-ai-safetyQuick validation run before committing to a full sweep. Runs a scenario with minimal seeds/epochs, checks for crashes and metric sanity, reports go/no-go.
- ▌ Claim · swarm-ai-safety bundleCreate or update a claim card in the vault. In create mode, guides through evidence collection, confidence assessment, and boundary conditions. In update mode, adds new evidence or changes status. Triggers on "/claim", "/claim create [proposition]", "/claim update [claim-id]", "/claim status", "add a claim", "update claim evidence".
- ▌ Verify · swarm-ai-safety bundleRun all vault integrity checks — schema validation, evidence provenance, wiki-links, index consistency, and claim health. Triggers on "/verify", "/verify --claims", "/verify --quick", "check vault health", "verify the vault".
- ▌ Kb Query · swarm-ai-safetyQuery the SWARM knowledge graph structurally — find pages, list backlinks, follow link paths, and surface related/semantic neighbors across docs, scenarios, slash commands, agents, roles, and code references. Prefer this over grep when you need *connected* answers ("what links to X", "how does X relate to Y", "what's similar to X").
- ▌ Swarm Safety · swarm-ai-safety bundleSWARM: System-Wide Assessment of Risk in Multi-agent systems. Simulate multi-agent dynamics, test governance, study emergent risks.