soofi-xyz-team-kit

Operating guide for the Oracle public-data ingestion agent: install and drive the elephant-xyz/skills (onboard-county + stage skills) against the oracle-node pipeline to discover, ingest, validate, refresh, and publish county property/permit plus Sunbiz/BBB data and coverage into Neon + Filebase/IPFS for MCP/Donphan/Miranda consumption. Covers the oracle-node checkout layout, sibling repos, AWS_PROFILE/AWS_REGION setup, the stage-skill map, and the coverage/IPNS handoff. Triggers on: oracle, onboard county, ingest county data, refresh Lee County, appraisal/permit/Sunbiz/BBB ingestion, oracle-node, county-discovery, coverage.

Operating guide for exploring Elephant data through the @elephant-xyz/mcp server in Cursor: Oracle open-data properties (appraisal, permits, Sunbiz, BBB), geo filters, lexicon schemas, and verified transform examples. Use when asking natural-language questions about Lee County properties, contractors, businesses, address mismatches, or schema fields — via MCP tools only. Not for Neon SQL (use-elephant-query-db) or county ingestion (use-oracle). Triggers on: elephant mcp, explore oracle data, donphan, contractors, BBB, Sunbiz, fort myers, address mismatch, nail salon, commercial property, property schema.

Extend the hypno Google Chat Asana initiative portfolio bot and WOW personal CLI workflows. Covers prompt-in-Markdown, gchat webhook flow, initiative project CRUD, acceptance-criteria parsing, portfolio analysis, my-tasks/story creation/store consolidation via manage-initiatives.mjs. Triggers on: hypno, asana initiatives, initiative portfolio, acceptance criteria, golden overlap, war plan initiatives, manage-initiatives, google chat asana bot, my tasks, WOW story, create-story, store-sync.

Use when changing or operating Lucario's M2D failed-document replay, run-status, manual approval, or restart-from-beginning behavior.

Use when changing or operating Lucario's M2D portfolio staging-config update flow, S3 config publication, GitHub PR creation, or approved sample-media verification.

Operating guide for the WOW Story Quality Agent (soofi-xyz/wow-website): when someone @-mentions the bot (Kirlia) in a comment on an Asana task of custom type Story, a Vercel runtime rewrites the story via LLM into the WOW format (clear title, Context / Description / Acceptance Criteria bullets / Demo Script), updates the task, and creates or refreshes a review subtask assigned to the mention author. Covers enabling/disabling projects via bot membership (discovery sweep), the org seed script and PATs, the webhook + 15-min sweeper trigger paths, the idempotent story_agent_tasks ledger, diagnosing missed or failed stories, and changing the WOW format prompt. Triggers on: story quality agent, story agent, WOW story format, rewrite Asana stories, mention the story bot, enable the story agent on a project, story agent missed a task, story_agent_tasks, seed-story-agent.

Use when implementing or changing the Persist service from its PRD: Neptune-backed graph persistence, Persist Blobs, GraphSON ingest, lexicon validation, deterministic IDs, Neptune CSV bulk load, Gremlin query APIs, and the lexicon-generated polymorphic GraphQL read surface with per-field resolution to Neptune, DynamoDB, or Interprose. Read reference/PRD.md first; combine with conkeldurr, machamp, and apply-engineering-guidelines.

Apply the repository Golden Path engineering standards. Use when building new services, refactoring existing code, setting up infrastructure, configuring CI/CD, choosing libraries, writing tests, adding observability, alerting, monitoring, or reviewing architecture decisions. Covers tech stack (TypeScript for all services, Python only for PySpark/Glue), AWS/CDK infrastructure, testing strategy (Vitest/Pytest), observability (structured logs, X-Ray, metrics), mandatory PagerDuty alerting on critical failures, self-resolving DLQ channel alarms, and AI policy. Triggers on: new service, scaffold, refactor, architecture review, tech stack question, infrastructure setup, CDK, testing setup, logging, observability, metrics, pagerduty, alerting, on-call, critical failure, dlq, dead-letter queue, cloudwatch alarm, channel notification, alarm fatigue, CI/CD pipeline. Do NOT trigger for general coding questions unrelated to the repository standards.

First phase of candidate test-task evaluation. Read the assignment user story (or assignment repository), candidate pull request, runtime URL, demo video, credentials, and the assignment-sent datetime, then derive the story's one-sentence intent — the business outcome that would make the work obviously useful — before any scoring. Fetch the candidate's latest commit timestamp from GitHub and compute elapsed delivery time. Build the evidence model (actors, workflow, data objects, scale signals, runtime proof, access boundaries, demo evidence), the pass/fail gates (PR to the designated assignment repo, runtime, credentials, demo), and the weighted 100-point scoring model. Use at the start of a hiring-candidate evaluation, before functional or implementation review. Not for designing new agents or products from scratch.

Evidence-and-functional-outcome phase of candidate test-task evaluation, run as a dedicated subagent. First enforce the gates from evaluate-candidate-intent — a missing PR to the designated assignment repo, demo artifact, or credentials are explicit Failed gates, and any runtime that is not a candidate-deployed hosted runtime (a locally run app such as localhost, a dev server, or local Docker), or that otherwise cannot be reached and exercised, is an absolute hard fail (score 0/100, verdict Fail, for any reason including environment or anti-automation blocks). Then drive the live deployed runtime with the Playwright browser to prove the story's intent through working runtime behavior, data evidence, output evidence, and demo artifacts, evaluate functional outcome before implementation, check assignment-specific access boundaries (for X Engagement, hosted investors-mcp read tools only — flag direct database, vector-store, or blob access as a violation), and return Pass/Partial/Fail/Blocked for each material ac

Guides creation of batch data processing workflows on AWS. Covers input data analysis, choosing Step Functions Distributed Map vs AWS Glue PySpark, testing pipelines, cost controls, throttling, idempotency, metrics, and mandatory PagerDuty alerting on critical failures. Triggers on: batch job, batch workflow, data pipeline, data processing job, ETL pipeline, bulk processing, distributed map, glue job, batch failure alerting, pagerduty, dead-letter queue.

Implementation phase of candidate test-task evaluation, run as a dedicated subagent, after the functional outcome is understood. Evaluate technical implementation quality — architecture, code structure, and the technical depth behind acceptance-criteria coverage — comparing against any reference repositories the story points at (for example the investors-mcp reference fork). Then score kit-usage conformance: determine which soofi-xyz kit agent(s)/skill(s) should have built a task like this by consulting arceus and README.md, consult each relevant builder agent as a read-only reviewer for a 0–100 coding score, and aggregate. Return separate implementation-quality and kit-usage scores with file-level evidence. Use after evaluate-candidate-product so implementation is judged third, never before the outcome.

Design, rebuild, and operate a generic Persist-backed short-URL service. Covers token issuance, GraphSON v3 persistence, graph-query resolution, public redirects, GraphFactProduced click facts on EventBridge, SQS DLQ replay, custom-domain discovery, and forbidden legacy patterns such as DynamoDB token tables, Snowflake dependencies, webhooks, metadata blobs, and business logic in the resolver.

Operate the Eevee editorial agent from Cursor: retrieve from the live Eevee RAG via the agent-eevee retrieve CLI, then draft pitches/propositions/website copy in Eevee's voice using Eevee's prompts. Triggers on: eevee, editorial draft, pitch, proposition, capability deck, white paper, rfp response, website copy, founder content.

Use every time the user creates a PR. Also use when the user asks to babysit a PR through merge and production release, watch PR comments and CI/CD until mergeable, monitor the release/prod deployment after merge, or open and babysit a follow-up fix PR for a failed production release.

User guide for consuming the Vercel Neon Postgres database named elephant-query-db: Vercel environment setup, schema imports from @elephant-xyz/query-db, and exact TypeScript/Drizzle query code for appraisal parcels/properties, Accela permits, BBB contractor reputation/quality scores, Sunbiz companies, and addresses. Use when building Vercel apps, route handlers, dashboards, reports, or agents that query Elephant oracle data. Not for changing loaders or schema migrations.

Assemble deterministic communication runtimes from reusable audience, template, and activity capabilities. Covers runtime workflow composition, internal data contracts, candidate generation, scoring, allocation, outputs, and runtime validation. Use when building or refactoring communication runtimes, workflow assembly agents, or end-to-end channel services.

Manage reusable communication activity loops across provider setup, send-file contracts, routing, execution handoff, delivery updates, response ingestion, and activity-state closure. Use when building provider adapters, communication activity agents, send workflows, delivery feedback processing, or channel operations for SMS or email.

Manage channel templates in Git-backed inventories, including template CRUD, metadata normalization, active/inactive state, family/variant structure, and one-time or recurring synchronization from operational stores into GitHub. Use when building or refactoring template-management agents, template repositories, template sync jobs, or channel template workflows for SMS or email.

User guide for calling a deployed Translate service — what a language and a runtime mapping are, the exact JSON shapes required to register them, the input shape Translate expects, and the request/response shape for validate, preview, and asynchronous executions. Use when explaining or making calls to /translate/* as a consumer. Not for changing the Translate codebase — use build-translate-service for that.

Build simple local TypeScript RAG CLIs for agents to query, with libSQL file databases, embedding-backed search, machine-readable commands, and AGENTS.md usage instructions. Use when prototyping local RAG, evaluating corpora, building query-only RAG tools, or teaching agents to use local retrieval from the terminal.

Build reusable AWS RAG systems with OpenSearch retrieval, Bedrock embeddings, SAM local Lambda emulation, Docker OpenSearch replay, historical ingestion, and webhook refresh. Use when working with cloud RAG, embeddings, OpenSearch retrieval, moving a local SQLite/libSQL RAG POC to AWS, knowledge bases, schema/header mapping, confidence thresholds, AWS production RAG, or local emulation.

Guides creation of AI agents using the rules-agent pattern. Covers Lambda-first agent runtime design, Chat SDK ingress via @soofi-xyz/chat-adapter-asana, Chat SDK state persistence via @soofi-xyz/chat-state-dynamodb, Vercel AI SDK ToolLoopAgent on Bedrock, Bedrock prompt caching, multi-intent request routing, LangSmith telemetry, AgentCore-backed AI conversation memory, tools, deployment, and testing. Triggers on: ai agent, build agent, lambda agent, asana bot, chat sdk, chat-adapter-asana, chat-state-dynamodb, webhook agent, tool loop agent, bedrock prompt cache, langsmith agent, agent memory, bedrock agent, agentcore memory.

Use when implementing or changing the Connect service from its PRD: partner integrations, flow specs, credentials and tokens, webhooks, static IP, SFTP, batch executions, and connector job APIs. Read reference/PRD.md first; combine with conkeldurr, machamp, and apply-engineering-guidelines.

Use when implementing or changing the Rules product from its PRD: tenant-local batch decisioning, rule contracts, Persist graph reads, callable-population outputs, audit reports, metrics, Glue preparation jobs, and Step Functions workflows. Read reference/PRD.md first; combine with conkeldurr, machamp, and apply-engineering-guidelines.

Use when implementing or changing the Build service from its PRD: TypeScript CDK source intake, CodeBuild synth and validation, CDK cloud assembly artifacts, artifact provenance, build manifests, and marketplace-ready bundle outputs. Read reference/PRD.md first; combine with conkeldurr, regigigas, machamp, and apply-engineering-guidelines.

Use when implementing or changing the Lexicon product from its PRD: governed graph vocabulary, ruleset data, metric definitions, source-system mapping artifacts, S3/SSM artifact publication, and the read-only schema browser. Read reference/PRD.md first; combine with conkeldurr, unown, porygon, and apply-engineering-guidelines.

Use when implementing or changing the Product service from its PRD: products, schemas, OpenAPI metadata, product flow templates, template-backed flows, invocations, waterfalls, reports, SMS, email, widgets, blobs, and operational telemetry. Read reference/PRD.md first; combine with conkeldurr, machamp, and apply-engineering-guidelines.

Use when implementing or changing the Marketplace service from its PRD: catalog, component bundles, releases, rollbacks, subscriptions, signed webhooks, review/status flows, and marketplace settings. Read reference/PRD.md first; combine with conkeldurr, regigigas, and apply-engineering-guidelines.

Use when implementing or changing the Translate service from its PRD: registered partner languages, versioned TypeScript mappings, validation, preview, asynchronous translation executions, mapping packs, and execution telemetry. Read reference/PRD.md first; combine with conkeldurr, machamp, and apply-engineering-guidelines.

Use when implementing or changing the Bootstrap CLI from its PRD: initial tenant bootstrap, Account bootstrap manifest intake, local first-Deployer install, Marketplace Puller install through Deployer, resume state, and status checks. Read reference/PRD.md first; combine with conkeldurr, regigigas, and apply-engineering-guidelines.

Use when implementing or changing the Marketplace Puller service from its PRD: tenant-side subscription intake, marketplace webhooks, desired-state reconciliation, dependency subscriptions, drift repair, and deployer handoff. Read reference/PRD.md first; combine with conkeldurr, regigigas, and apply-engineering-guidelines.

Use when implementing or changing the Deployer service from its PRD: tenant-local CloudFormation/CDK deployment execution, bundle contracts, regional stack orchestration, Docker image handling, callbacks, and status inspection. Read reference/PRD.md first; combine with conkeldurr, regigigas, and apply-engineering-guidelines.

Use when implementing or changing the Account service from its PRD: customer and organization identity, API keys, AWS sub-account provisioning, DNS configuration, bootstrap manifest, and maintenance access. Read reference/PRD.md first; combine with conkeldurr, regigigas, and apply-engineering-guidelines.

Lexicon-first metrics for contact-center / operational platforms: combine row-level (atomic) facts with vendor daily rollups, Parquet in object storage, CloudWatch publication, and clear lineage in cloudwatch-metrics.json and lexicon.json. Use when designing or implementing atomic vs rollup layers, reconciliation, inbound/outbound classification rules (e.g. LIVEVOX_INBOUND_RULES_JSON), metrics_mapper business keys, Athena reconciliation views, or per-agent daily reporting. Triggers on: atomic metrics, call_fact, vendor rollup, contact center metrics, agent daily metrics, metrics reconciliation, Parquet derived layer, blob vs atomic storage, operational platform metrics.

Define reusable audience-selection and eligibility-handoff patterns for communication services. Covers upstream filter boundaries, runtime intake contracts, external input schemas, and how eligible populations are packaged for downstream optimization. Use when building audience-selection agents, filter-to-runtime handoffs, communication segmentation, or `input_s3_uri`-style intake contracts.

Guides building the Tenant Domain Router — a standalone marketplace product that owns a single root domain (e.g. `provider.xyz`), mints one delegated subdomain per tenant environment (e.g. `customer.provider.xyz`), and exposes a uniform contract for any other product to attach a base path under that subdomain. Covers the marketplace-account parent hosted zone, NS-delegation to per-tenant child hosted zones, ACM certificate strategy, and the API Gateway / CloudFront base-path-mapping interface that other products consume. Triggers on: tenant subdomain, root domain delegation, route53 hosted zone delegation, per-tenant DNS, base path mapping, multi-tenant custom domain, ACM wildcard, marketplace domain product.

Integrate shared CI/CD GitHub Actions workflows into a project. Covers creating caller workflows, justfile with required recipes, and environment variable configuration. Triggers on: CI/CD, continuous integration, continuous deployment, GitHub Actions, pipeline, ci-cd, workflow, deploy pipeline, add CI, add CD, setup CI/CD.

Build HTML-to-PDF generation workflows on AWS Lambda using Playwright and Chromium. Covers typed request contracts, template registries, deterministic HTML rendering, Lambda Chromium packaging, PDF metadata and policy, and verification. Use when building or refactoring server-side PDF generation, print-first HTML templates, Lambda PDF renderers, or Chromium-based document generation. Triggers on: html to pdf, pdf generation, pdf renderer, chromium pdf, playwright pdf, lambda chromium, server-side pdf, print html, document rendering.

Design and implement lexicon-first metrics workflows across the repositories, including candidate metric discovery, snapshot semantics, dashboard delivery, and PR follow-up. Use when defining a new metric, mapping vendor metrics to canonical names, extending CloudWatch metric registration, integrating a metric into a pipeline, adding dashboard widgets, or preparing coordinated lexicon/pipeline/dashboard delivery. Triggers on: new metric, metrics workflow, cloudwatch metric, vendor metric mapping, metric pipeline, dashboard widget, lexicon metric, livevox metric, business metric.