kiro-with-harness

Create professional architecture, workflow, sequence, data-flow, and lifecycle/state diagrams as standalone HTML files with SVG graphics, a built-in dark/light theme toggle, and one-click export to PNG / JPEG / WebP / SVG. Accepts plain-language descriptions or pasted Mermaid code (flowchart, sequenceDiagram, stateDiagram) and lays the diagram out from scratch in archify style. Use when the user asks for system architecture diagrams, infrastructure diagrams, cloud architecture visualizations, security diagrams, network topology, technical workflows, approval flows, runbooks, CI/CD flows, process diagrams, API call sequences, request lifecycles, data pipelines, ETL/ELT maps, PII boundaries, data lineage, state machines, lifecycle diagrams, status transitions, or asks to convert/beautify a Mermaid diagram.

Data analysis methodology — how to frame a question, pick the right technique, avoid statistical traps, and connect results to decisions. NOT a tool tutorial (pandas/polars/duckdb live in python-data-analysis) — this is the judgment layer: problem framing, analysis-type decision tree, experiment/causal design, and domain playbooks (churn, cohort, funnel, anomaly). Trigger: "왜 늘었/줄었지", 이탈 분석, 코호트, 퍼널, A/B 테스트, 인과추론, 상관 vs 인과, 유의성, 세그먼트, 이상 탐지, "이 데이터로 뭘 봐야", 지표 설계, exploratory analysis.

AWS cost management and FinOps practice — Cost Explorer, Budgets, Cost Anomaly Detection, CUR/data exports, cost allocation tags, Cost Categories, Savings Plans vs Reserved Instances, Compute Optimizer / Cost Optimization Hub, rightsizing, unit economics, showback/chargeback. Grounded in the FinOps Foundation Framework (Inform / Optimize / Operate). Trigger: cost anomaly, Savings Plan, Reserved Instance, rightsizing, cost allocation tag, chargeback, showback, unit cost, budget alert, CUR, Cost Explorer, unblended/amortized cost, RI coverage, commitment, "왜 청구서가 늘었지".

Track and report AI model token usage, spending, and budgets from a local cost-tracking database.

Multi-source deep research using firecrawl and exa MCPs. Searches the web, synthesizes findings, and delivers cited reports with source attribution. Use when the user wants thorough research on any topic with evidence and citations.

Conduct market research, competitive analysis, investor due diligence, and industry intelligence with source attribution and decision-oriented summaries. Use when the user wants market sizing, competitor comparisons, fund research, technology scans, or research that informs business decisions.

기술 문서를 정확하고 명확하고 실행 가능하게 작성·윤문하는 스킬. 개발 가이드·API 문서·README·테크 블로그·기술 리포트를 대상으로, 번역투·hype·모호성을 제거하고 전제조건·코드 예제·용어 일관성·구조를 보강한다.

Amazon Bedrock — Converse API, model invocation (Claude/Llama/Nova/Mistral/Titan), Agents, Knowledge Bases, Guardrails, prompt caching, cross-region inference profiles, and cost tracking. Use when calling foundation models inside an AWS boundary (bedrock-runtime, Converse, retrieve_and_generate, guardrails, embeddings).

Capture and apply a consistent brand — colors, typography, logo usage, spacing, and voice — across generated documents, slides, and web artifacts. Use when output must match a company's visual identity, or when the user provides brand assets / a style guide to follow.

Anthropic Claude API patterns for Python and TypeScript. Covers Messages API, streaming, tool use, vision, extended thinking, batches, prompt caching, and Claude Agent SDK. Use when building applications with the Claude API or Anthropic SDKs.

Cost optimization patterns for LLM API usage — model routing by task complexity, budget tracking, retry logic, and prompt caching.

Generate native Word .docx documents — headings, styled paragraphs, tables, images, and sections — with python-docx, or convert Markdown to .docx via pandoc. Use when the user wants an editable Word document (reports, letters, contracts, specs) rather than a PDF or HTML page.

Edit long-form and web writing so it reads like a person wrote it — cut AI tells (filler openers, uniform rhythm, flagged vocabulary, em-dash overuse, listicle bloat) and raise real quality through specificity and varied sentence rhythm. Use when polishing blog posts, newsletters, docs, landing-page copy, or any prose that will be published on the web. English/general companion to humanize-korean.

Scaffold and harden Model Context Protocol (MCP) servers from a capability spec — choose stdio vs HTTP transport, define tools/resources/prompts, validate inputs, and register the server with Kiro. Use when the user wants to build, wrap an API as, or debug an MCP server.

Create, fill, and extract PDFs. Pick the right tool by task — reportlab for data-driven documents, WeasyPrint for HTML/CSS templates, pandoc+Typst for Markdown, and pypdf/pdfplumber for merging, splitting, form-filling, and extraction. Use when the user wants to generate a PDF report, invoice, certificate, or convert content to PDF.

Generate native PowerPoint .pptx decks programmatically — slides, titles, bullet bodies, tables, charts, and images — with python-pptx (or pptxgenjs in Node). Use when the user wants an editable .pptx file rather than an HTML deck. For browser-based animated decks use frontend-slides instead.

Generate Excel .xlsx workbooks — sheets, styled tables, live formulas, and charts — with openpyxl (formulas/editing), XlsxWriter (fast/large/streamed), or pandas.to_excel (quick DataFrame dumps). Use when the user wants a spreadsheet report, financial model, or data export as a real .xlsx file.

Loop engineering for Kiro's autonomous features — how to drive `/goal` (iterative iterate-until-done loops, default 5 / `--max N`, self-verifying completion), Subagent Review Loops (reviewer rejects work back to the implementer until it passes), and DAG delegation (parallel sub-agents, verify-then-converge). Centers on writing goals as evidence-backed completion contracts (each success criterion verifiable by tool output, not narrative), designing RED→GREEN verification gates, and avoiding loop failure modes (declaring done without proof, no-progress thrashing, goal drift, cross-iteration context bloat). Use when setting up an autonomous `/goal` run, a review loop, or a multi-stage delegation pipeline. Triggers: /goal, loop engineering, autonomous loop, completion criteria, verify then converge, subagent review loop, max iterations, self-verification, agent keeps looping, when is it done.

Decision guide for AWS ETL / data integration / CDC — choose the right service and pattern for the user's situation across AWS DMS, AWS Glue, Kinesis Data Streams, Amazon Data Firehose (formerly Kinesis Data Firehose), Amazon Managed Service for Apache Flink (formerly Kinesis Data Analytics), Amazon MSK + MSK Connect (Debezium), EventBridge Pipes, Lambda, Step Functions, and zero-ETL integrations. Routes by batch vs streaming, CDC vs bulk, and — critically — by transform type (passthrough/format-only, stateless per-record filter/enrich, stateful windowed aggregation, heavy batch). Use when designing a pipeline to move/transform data between databases, streams, S3, OpenSearch, Redshift, or a lakehouse. Triggers: AWS ETL, CDC, change data capture, DMS, Debezium, MSK Connect, Glue job, Glue streaming, Kinesis, Firehose, Flink, EventBridge Pipes, zero-ETL, ingest pipeline, stream processing, batch transform, replicate database.

Always resolve, pin, and record the current latest stable/supported version of every versioned infrastructure component before provisioning or upgrading — never trust training-data defaults. Authoritative version-check commands for Amazon EKS (Kubernetes versions + add-ons + version skew), Amazon MSK (Apache Kafka versions), Terraform CLI & providers, Docker base images (pin by digest), Helm charts, and AWS CLI v2. Use whenever choosing or upgrading an EKS cluster version, MSK Kafka version, Terraform provider, container base image, or Helm chart. Triggers: latest version, EKS version, eksctl, describe-addon-versions, MSK Kafka version, list-kafka-versions, terraform provider version, .terraform.lock.hcl, pin image digest, helm chart version, version skew, upgrade EKS, end of support.

Offload high-volume log / append-only / time-series data out of an RDBMS into Amazon S3 (cheap retention + analytics via Iceberg/Athena) or Amazon OpenSearch (full-text search + observability dashboards + alerting). Covers the S3-vs-OpenSearch decision, RDBMS-side time partitioning and fast purge (DROP PARTITION, not DELETE), offload pipelines (AWS DMS full-load+CDC, Kinesis Data Firehose, Debezium/MSK, scheduled export), OpenSearch ingestion (Firehose, OpenSearch Ingestion/Data Prepper, Fluent Bit), index rollover & retention (ISM), UltraWarm/Cold tiers, and hot-warm-cold tiered architecture. Use when an RDBMS log/event/audit table is bloating, slow, or expensive and you want to move it to S3 or OpenSearch. Triggers: log table bloat, audit log, event log, time-series in RDBMS, archive old rows, DMS to S3, Firehose, OpenSearch ingestion, ISM rollover, UltraWarm, CDC, Debezium, partition pruning, cold storage, retention policy, move logs out of Postgres/MySQL.

S3-based lakehouse on Apache Iceberg — Amazon S3 Tables (managed Iceberg table buckets with auto compaction/snapshot maintenance), self-managed Iceberg on plain S3 (Glue/REST catalogs), querying with Athena (engine v3, MERGE/UPDATE/DELETE), and processing with Spark on EMR/Glue/EMR Serverless. Covers catalog choice, Iceberg V3 (deletion vectors, row lineage) and its Athena incompatibility, Spark engine/language speed (Scala vs PySpark vs Kotlin), native accelerators (Comet/Gluten/Photon), and S3 Tables cost pitfalls. Use when building or querying a data lake on S3, choosing between S3 Tables and self-managed Iceberg, picking a query engine, or tuning Spark performance. Triggers: S3 Tables, table bucket, s3tables, Apache Iceberg, Iceberg REST catalog, Glue Data Catalog, Athena Iceberg, MERGE INTO, time travel, deletion vectors, Iceberg V3, EMR Spark, Glue ETL, PySpark slow, Spark accelerator, Comet, Gluten, lakehouse, partition projection.

AWS SDK and CLI usage patterns — boto3 / aioboto3 (Python), AWS SDK for JavaScript v3, and AWS CLI v2. Covers credential resolution (SSO/profiles/OIDC), retry & timeout config (adaptive mode), paginators & waiters, ClientError handling, batch/transaction calls, async clients, JS v3 command/middleware pattern, and CLI v2 --query/--profile/pager. Use when writing AWS SDK code (not architecture — see aws-cloud), debugging throttling, credential chain issues, pagination, or scripting the AWS CLI. Triggers: boto3, aioboto3, botocore Config, paginator, waiter, ClientError, @aws-sdk/client-*, AWS CLI, aws configure sso, --query JMESPath, retry mode, NoCredentialsError, ThrottlingException.

Vite 6/7+ build tool patterns: config, plugins, HMR, env vars, dev proxy, library mode, dependency pre-bundling, build optimization, security pitfalls. Trigger keywords: vite.config.ts, vite.config.js, defineConfig, loadEnv, import.meta.env, VITE_, vite-plugin-checker, vite-tsconfig-paths, vite-plugin-dts, server.proxy, optimizeDeps, manualChunks, build.lib, rolldown, rollup, esbuild, oxc, HMR, import.meta.hot, vite preview, vite build, monorepo, docker vite, type check vite.

Generate draw.io (diagrams.net) diagrams using a grid cell coordinate system, catching left/right and top/bottom imbalance, arrow/text overlap via an MCP tool loop (create→export→verify→edit). Triggers — "draw draw.io", "make diagram", "drawio diagram", "architecture diagram", "draw flowchart/sequence", "mxGraph XML", "fix diagram alignment/overlap".

Applied to DynamoDB table design, key/index strategy, query patterns, and boto3/aioboto3 connection management. Triggers: DynamoDB table design, PK/SK, GSI, LSI, Single Table Design, access pattern, partition key, sort key, overloading, Query vs Scan, TTL, DynamoDB Streams, boto3, aioboto3, TransactWrite, BatchWrite related tasks.

Git workflow patterns including branching strategies, commit conventions, merge vs rebase, conflict resolution, and collaborative development best practices for teams of all sizes.

Lightweight learning log that accumulates one-line lessons extracted from repeated corrections in a single file. When the same review findings, build failure patterns, or user corrections repeat, this prevents making the same mistake twice across sessions. Lighter than /learn (pattern→skill extraction), more explicit than continuous-learning-v2's instinct (automatic observation).

Production machine-learning engineering workflow for data contracts, reproducible training, model evaluation, deployment, monitoring, and rollback. Use when building, reviewing, or hardening ML systems beyond one-off notebooks.

MongoDB 7.0+ schema design, collection/index creation, query optimization, sharding, and async connection management. Triggers: createCollection, createIndex, aggregation, $lookup, $match, $group, TTL index, async client, embedded document, content schema, metadata, document design, chat/log storage, user settings, content/metadata operations.

MongoDB runtime patterns — aggregation pipelines, query/index usage, transactions, async (motor) connection pools, and diagnostics. Complements mongodb-guideline (which owns schema/naming/shard-key design).

MySQL 8.0+ schema design, table/index creation, query optimization, partitioning, connection management, development principles and anti-patterns, JDBC driver selection. Triggers: CREATE TABLE, ALTER TABLE, slow query analysis, index design, RANGE partition, MySQL migration, utf8mb4, InnoDB, transaction management, UPSERT, Covering Index, composite index, normalization, data type selection, INET_ATON, UUID_TO_BIN, DATETIME TIMESTAMP, stored procedure, COUNT(*), random PK, JSON column, JDBC, Connector/J, AWS Advanced JDBC Wrapper, Aurora failover related tasks.

PostgreSQL 16+ schema design, table/index creation, query optimization, partitioning, and psycopg3 connection management. Triggers: CREATE TABLE, GENERATED ALWAYS AS IDENTITY, EXPLAIN ANALYZE, GIN/BRIN/GiST indexes, RLS, PARTITION BY RANGE, pg_partman, LISTEN/NOTIFY, Advisory Lock, UPSERT ON CONFLICT, CTE, timestamptz operations.

Common RDBMS naming and data type conventions (single source for MySQL and PostgreSQL). Table/column snake_case, singular, active voice (with date/time column exceptions), prefix/postfix, abbreviation dictionary, column prefix/suffix system, index naming (_IDX/_UIDX/_FTX), data type selection. Triggers: table/column/index design, DDL authoring, schema review, naming conventions, snake_case, abbreviations, PK/FK naming, boolean columns, DECIMAL, settlement amount columns.

Head-to-head comparison of coding agents on custom tasks with pass rate, cost, time, and consistency metrics

Design and optimize AI agent action spaces, tool definitions, and observation formatting for higher completion rates.

Operate as an agentic engineer using eval-first execution, decomposition, and cost-aware model routing.

Engineering operating model for teams where AI agents generate a large share of implementation output.

Regression testing strategies for AI-assisted development. Sandbox-mode API testing without database dependencies, automated bug-check workflows, and patterns to catch AI blind spots where the same model writes and reviews code.

Clean Architecture patterns for Android and Kotlin Multiplatform projects — module structure, dependency rules, UseCases, Repositories, and data layer patterns.