37 AI agent kernels. One bus.

Watches telemetry and emits typed signals to the bus.

lib/agents/detectorSignalEmitter

Weaves 1..N signals into a typed hypothesis with a confidence score.

lib/agents/reasonerHypothesisWeaver

Sandboxes the proposed action end-to-end and returns a verdict before any approval packet is built.

lib/agents/simulatorSandboxSpec

Drafts a typed engineering spec (goals, non-goals, risks, verification, rollback) from a problem statement.

lib/agents/specWriter

Reads a diff and proposes the must-have, should-have, and nice-to-have tests with rationale per row.

lib/agents/testCoverageProposer

Orders a multi-step refactor with per-step verdict (safe / needs-test / needs-review / blocked) and rollback.

lib/agents/refactorSequencer

Builds a multi-stage migration runbook with backwards-compat window, gate checks, and rollback per stage.

lib/agents/migrationCoordinator

Classifies a failed Actions run and proposes a typed PR patch (retry, pin, quarantine, raise runner, etc.).

lib/agents/githubPipelineRepairer

Drafts release notes from commits between two tags. Conventional-commit aware, recommends SemVer bump.

lib/agents/githubReleaseNotesDrafter

Weaves heterogeneous events into a phase-tagged incident timeline with TTD / TTM / TTR durations.

lib/agents/incidentTimelineWeaver

Renders a ready-to-merge postmortem from a timeline + action items. Refuses to draft while incident is ongoing.

lib/agents/postmortemDrafter

Drafts multi-channel social posts with closed-union risk tier. Customer mention auto-tiers high; incident → critical.

lib/agents/marketingContentDrafter

Sequences approved drafts under per-channel caps, blackout windows, dedup. Refuses critical+dual-approval on a single track.

lib/agents/socialPostScheduler

Audits tables / columns / indexes / FK constraints. Surfaces missing-FK, missing-index, naming drift, duplicate indexes, orphan columns.

lib/agents/databaseSchemaReviewer

Reads slow-query rows + schema, proposes add_index (with DDL), rewrite_query, partition_table, downsize, or no_action.

lib/agents/slowQueryProposer

Finds idle EBS / EIP / Lambda / Snapshot / RDS / ELB / empty S3. Production-tag refuses auto-delete; surfaces $/mo savings.

lib/agents/idleCloudResourceDetector

Analyzes alert fire / ack / incident rates, proposes raise_threshold / widen_window / require_paired_metric / mute / split_routing.

lib/agents/alertNoiseReducer

Builds a 4-phase onboarding plan (pre-start / day-one / week-one / month-one). Refuses interns with elevated access.

lib/agents/hrOnboardingPlanner

Classifies an inbound lead — segment (enterprise / mid-market / startup / personal / education), ICP fit score, suggested plan + next action.

lib/agents/salesLeadEnricher

Chains kernels into typed workflows with per-step gate handoff. Closed-union step states; safety-tier rejection blocks downstream dispatch.

lib/agents/agentWorkflowOrchestrator

The agent that picks agents. Ranks the kernel catalog against an operator problem with closed-union pick / ambiguous / no_kernel verdict.

lib/agents/metaReasonerKernel

N kernels propose, council reconciles, dissent surfaced explicitly. Safety-tier reject is a hard veto.

lib/agents/multiAgentDebate

Natural-language operator intent → typed ActionSpec (kind + domain + urgency + entities + destructive flag).

lib/agents/intentParser

MAD-based z-score on time series. Spike / dip / drift / missing_data with bounded false-positive rate.

lib/agents/anomalyDetector

Scores blast radius of a proposed change against service topology. Critical-tier reach → council_3_of_5 gate.

lib/agents/changeRiskAssessor

Scans a file / diff for AWS / Stripe / GitHub / OpenAI / Anthropic / PEM / Slack secrets. Redacts in output; never echoes the raw value.

lib/agents/secretsHygieneScanner

Maps audit events → SOC 2 / ISO 27001 / GDPR / HIPAA controls. Per-framework readiness % + blocking-gap list.

lib/agents/complianceControlMapper

Applies the tenant charter to every proposal — refuses anything outside the operator-signed scope.

lib/agents/policyGateEvaluator

Classifies the blast radius of every proposal into a closed-union severity tier.

lib/agents/boundaryGateCatalog

Weighted-vote consensus — ⅔ default — across the planning + safety agents.

lib/agents/council

Assembles the approval packet the operator sees — the only gate that ever lets autonomy act.

lib/agents/approverPacketAssembler

Post-execution check — confirms the action achieved the expected outcome.

lib/agents/verifierPostExecChecker

Writes the durable sha-256 rationale row that makes the action replayable and provable.

lib/agents/auditorRationaleWriter

Proposes method improvements from the audit trail — the loop that turns evidence into a better proposal.

lib/agents/improverProposalSynthesizer

Wilson-lower-bound calibration of kernel accuracy from approval / reject / rollback outcomes. Recommends trust_more / pause / trust_less.

lib/agents/confidenceCalibrator

Aggregates per-agent activity into the operator-visible cockpit timeline.

lib/agents/agentActivityAggregator

Compresses raw observations into episodic / semantic / procedural records. Recurrent observations promote to semantic. Prunes by relevance + age.

lib/agents/agentMemoryConsolidator