convez.ai
Agente
Cheeryoh

signal-loop

Cheeryoh/signal-loop

Ver no GitHub
InicialPluginScore · de 0 a 100

HITL multi-agent orchestration turning customer signals into a ranked education-priorities brief + drafted asset. Claude Code Max plan.

Instalação

$npx convez-ai Cheeryoh/signal-loop

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 1 semana
Versão
—
Issues abertas
0
Contribuidores
1

README

A tradução para o português deste README ainda está na fila.

Signal Loop

A human-in-the-loop, multi-agent orchestration that turns the firehose of customer signals into a ranked education-priorities brief — and drafts the #1 asset before the meeting even starts.

This closes the loop the other two products open. Launch Content Studio produces launch assets; Drift Sentinel keeps courses current; Signal Loop decides what to build next, from evidence. It takes mixed signals — support tickets, usage telemetry, CS notes, reviews, sales notes — clusters them into friction themes, honestly triages which are education's job versus Product/Pricing/Engineering's, ranks the education-addressable ones, and writes a first draft of the top one.

Part of a three-product portfolio. See also Launch Content Studio (launch asset packages) and Drift Sentinel (product↔course drift).


Why it exists

"Gather customer signals and feedback to continuously improve educational programs" is easy to say and hard to operationalize. The trap is reacting to the loudest ticket instead of the highest-leverage gap — or worse, answering a product problem with a webinar. Signal Loop forces the discipline: cluster before reacting, triage before building, rank before drafting. A human signs off at each of those three decisions, so the prioritization stays accountable and the "not education's problem" calls are explicit and routed, not buried.

The pipeline

 SignalIngestor ────────────────────────────▸ (structure)
 ThemeClusterer ─────────────────────────────▸ [Gate 1 · Friction themes]
 InterventionMapper ─────────────────────────▸ [Gate 2 · Education vs deferred]
 StoryScout ─────────────────────────────────▸ (surface success stories)
 PriorityBriefWriter → InterventionDrafter ──▸ [Gate 3 · Priorities brief + #1 draft]
AgentRole
SignalIngestorAgentMixed signal blob → structured records (source, sentiment, product area)
ThemeClustererAgentSignals → 3–6 friction themes with frequency + severity (not one-per-signal)
InterventionMapperAgentEach theme → root-cause category + honest "education-addressable?" call + best-fit asset
StoryScoutAgentPositive signals → candidate success stories worth amplifying
PriorityBriefWriterAgentEducation-addressable themes → ranked brief (impact, effort, owner) for leadership
InterventionDrafterAgentWrites a ready-to-edit first draft of the #1 priority

The honesty of the triage step is the point. Mapping a billing complaint to category: pricing, educationAddressable: false, recommendedAsset: none — and routing it out — is treated as a correct, valuable answer, not a failure to find work. The reject path lets a human enforce exactly that (the included rejection test pushes a mis-categorized pricing theme back out of education's scope).

Run it

Requires Node 20+. No API key needed — agent calls run through your local claude CLI (Max plan) by default; zero API billing.

npm install
npm run dev          # http://localhost:3000

Paste a batch of customer signals (or click an example) and approve/reject each gate.

LLM backends

BackendWhenCost
cli (default)local claude CLI presentnone (Max plan)
apiANTHROPIC_API_KEY setAPI usage
fixtureUSE_FIXTURES=truenone — canned outputs for offline/demo/test

Test

npm test            # full pipeline, fixture mode, auto-approve
npm run test:reject # rejects the triage gate once to exercise the rewind path

Both assert all six artifacts are produced and the run completes.

Architecture

Shares the kernel (src/kernel/) verbatim with the other two portfolio products: promise-based HITL gates, dual-backend ClaudeClient, filesystem persistence under .data/, SSE live progress. Only agents.ts, schemas.ts, Conductor.ts, and public/config.js are product-specific. See HANDOFF.md.

src/
  kernel/        shared orchestration engine (identical across products)
  agents.ts      6 product agents
  schemas.ts     Zod artifact schemas
  Conductor.ts   the signal-loop pipeline
  index.ts       wires the conductor into the kernel server
public/          dashboard
fixtures/        canned agent outputs for offline tests
scripts/         testPipeline.ts

MIT licensed.

Ver o original no GitHub

Bom para

Relacionados