# Caskey Engineering (caskeycoding.com)

> Caskey Engineering is Eric Caskey's independent engineering practice, helping teams assess context, review, and release checks in AI-assisted software delivery. Explore the method, inspect the evidence, and discuss a bounded workflow assessment.
> Personal site: https://ericcaskey.com

## Key Facts
- Thesis: make AI-assisted delivery reliable and measurable. The rule
  behind the method: automation at scale is only as safe as the checks
  that run before it acts.
- Owner: Eric Caskey, Senior Software Engineer at Amazon. Owns an
  infrastructure monitoring platform and a multi-region workflow
  orchestration control plane whose guardrails validate every change
  before it runs.
- Prior: Prudential Financial (enterprise monitoring, reporting, and an
  MFA self-service platform), 2013 to 2022.
- Site: a Next.js static export served from S3 and CloudFront. Every page,
  including all JSON-LD, is present in the raw HTML.
- Method: every feature starts as a spec in a versioned specs repository;
  AI agents implement against the spec and a local-first CI gate. The
  method's provenance is an engineering team of one, recorded in the
  site's "An Engineering Team of One" lesson series.
- Offer: a bounded assessment of one AI-assisted delivery workflow. The
  offer sits at https://caskeycoding.com/#workflow-assessment and
  inquiries start at https://caskeycoding.com/contact.
- Finance content is educational engineering, not investment advice.
- Canonical name: Caskey Engineering. Canonical author: Eric Caskey.

## About
- [About Eric Caskey](https://caskeycoding.com/profile): Career summary,
  current role, and the systems he owns at Amazon and built at Prudential.
- [Methodology: How the Engine Works](https://caskeycoding.com/finance/methodology):
  How the decision engine scores a company: the composite blend, the
  valuation and momentum factors, and the catalog behind every grade.
- [How this site is organized](https://caskeycoding.com/blog/how-this-site-is-organized):
  The map of the site: case studies, blog, lab, and how they connect.
- [Data Dictionary](https://caskeycoding.com/dictionary): The working
  vocabulary of the site: each recurring industry term and house coinage
  defined, with what the site uses it for.

## Case Studies
- [Case Studies](https://caskeycoding.com/case-studies): Deep-dive
  projects across Amazon, Prudential Financial, and Caskey Engineering.
- [Multi-Region Workflow Orchestration Platform](https://caskeycoding.com/case-studies/workflow-orchestration):
  Amazon, 2024 to present. Safety guardrails that validate every workflow
  change before it runs across regions.
- [Standardized Infrastructure Monitoring Across Service Teams](https://caskeycoding.com/case-studies/fleet-monitoring):
  Amazon, 2022 to present. One monitoring standard for a fleet of service
  teams.
- [Standardized Enterprise Monitoring Across a Fortune 100 Infrastructure](https://caskeycoding.com/case-studies/enterprise-monitoring):
  Prudential Financial, 2016 to 2021.
- [Sole Owner of Enterprise Access Reporting Through the COVID-19 Disruption](https://caskeycoding.com/case-studies/crisis-reporting):
  Prudential Financial, March 2020 to February 2021.
- [Enterprise MFA Self-Service Platform: Web, Mobile, and QR Code Enrollment](https://caskeycoding.com/case-studies/mfa-at-scale):
  Prudential Financial, 2013 to 2022.
- [Factor-First AI Investment Platform Narrated by a Six-Persona Committee](https://caskeycoding.com/case-studies/investment-analysis-tool):
  Caskey Engineering, 2024 to present. The engine behind the lab's
  finance tools.
- [AI Marathon Coach Grounded in 90 Days of Actual Training Data](https://caskeycoding.com/case-studies/ai-marathon-coach):
  Caskey Engineering, 2025 to present. Deterministic rules, LLM
  narratives, live Garmin data.
- [A Fork-Your-Own Life OS an AI Agent Reasons Over, Without Owning Your Data](https://caskeycoding.com/case-studies/specself-starter):
  Caskey Engineering, 2026 to present. A life as plain markdown an agent reads but never writes.
- [An LLM System Whose Best Feature Is Refusing](https://caskeycoding.com/case-studies/ballast):
  Caskey Engineering, 2026 to present. Self-healing RAG, a guardrails gateway, and an eval gate, open source.
- [Context Architecture Beats Documentation Dumps](https://caskeycoding.com/case-studies/context-architecture):
  Caskey Engineering, 2024 to present. Scoped spec slices for AI agents instead of documentation dumps.
- [A One-Person Engineering Platform Run Through Spec-Driven Development](https://caskeycoding.com/case-studies/sdd-platform):
  Caskey Engineering, 2025 to present. A supervisor, cheaper executors, a work queue, and an attested merge gate.
- [A Self-Governing Node in a Solo Fleet](https://caskeycoding.com/case-studies/solo-fleet-node):
  Caskey Engineering, 2026 to present. An ARM board with a resident agent under a written constitution.

## Blog
- [Blog](https://caskeycoding.com/blog): Platform engineering, AI tooling,
  spec-driven development, distributed systems, safety guardrails.
- [Full text of every post](https://caskeycoding.com/llms-full.txt): One
  plain-text file with the complete body of each published essay.
- [Lesson 1: Write Feature Specs Before Coding](https://caskeycoding.com/blog/lesson-1-the-spec-directory):
  First lesson in a series on running an AI-powered software team of one.
- [Your CI was green and the pipeline produced nothing: a nine-day silent failure](https://caskeycoding.com/blog/the-pipeline-that-ate-itself):
  A scheduler default fed an LLM memory pipeline its own transcripts for
  nine days while every stage reported success; the fix is postconditions
  that outrank green dashboards.
- [Spec-Driven Development and the Folder Architecture That Makes It Work](https://caskeycoding.com/blog/spec-driven-development-and-the-folder-architecture-that-makes-it-work):
  The original SDD write-up: why specs live in a versioned tree and how
  agents read it.
- [SDD isn't about managing AI agents, it's about managing context](https://caskeycoding.com/blog/sdd-is-context-management):
  The core argument for spec-driven development as context engineering.
- [Designing Safety Guardrails for Distributed Workflow Orchestration](https://caskeycoding.com/blog/designing-safety-guardrails-for-distributed-workflow-orchestration):
  How a control plane validates every change before it runs.
- [How to backtest without fooling yourself](https://caskeycoding.com/blog/backtesting-without-fooling-yourself):
  Survivorship, look-ahead, and the other ways a backtest lies.
- [Composite What You Trust, Watch What You Don't](https://caskeycoding.com/blog/composite-what-you-trust):
  A trust boundary for data with money attached.
- [Fable Thinks, Opus Builds, Kimi Types](https://caskeycoding.com/blog/fable-thinks-opus-builds-kimi-types):
  Routing work across model tiers by the judgment each step needs.
- [Institutional Memory for a Team of One](https://caskeycoding.com/blog/institutional-memory-for-a-team-of-one):
  Keeping decisions and lessons where the next session will find them.
- [When CI Costs More Than It Saves](https://caskeycoding.com/blog/when-ci-costs-more-than-it-saves):
  Why the pre-push hook became the gate and hosted CI became the mirror.
- [Prompt caching is a prefix match, not a flag](https://caskeycoding.com/blog/prompt-caching-four-layers):
  Four layers of prompt caching and what breaks each one.
- [Building an AI Marathon Coach](https://caskeycoding.com/blog/building-an-ai-marathon-coach-deterministic-rules-llm-narratives-2026-nyc-marathon):
  Deterministic rules, LLM narratives, and the 2026 NYC Marathon.
- [Seven articles to a platform that improves itself](https://caskeycoding.com/blog/seven-articles-to-a-platform-that-improves-itself):
  The reading list behind the self-improving loop, with sources.
- [One summer of building with AI agents, counted commit by commit](https://caskeycoding.com/blog/a-summer-of-sharpening):
  A commit census of one summer of building, in numbers.

## Lab
- [The Lab](https://caskeycoding.com/lab): Interactive tools, live market
  visualizations, and short games built on real price history.
- [Investment Committee](https://caskeycoding.com/lab/investment-committee):
  Score any stock across four weighted factors (Quality, Valuation,
  Momentum, and Health); six investor lenses argue out a written verdict.
- [Marathon Coach](https://caskeycoding.com/lab/marathon-coach): Live
  Garmin data turned into load, mileage, and heart-rate analytics, with
  an AI coach grounded in the athlete's own training history.
- [Grade a Portfolio](https://caskeycoding.com/lab/grade-portfolio):
  Watch the scoring engine take a sample portfolio apart, end to end.
- [Chat with a Post](https://caskeycoding.com/lab/chat-with-this-post):
  Ask a blog post questions; it answers only from the text, and shows
  where.
- [Equations Index](https://caskeycoding.com/lab/equations): Every formula
  the finance tools use, with the code path that implements each one.
- [Pokemon Cards vs. the Market](https://caskeycoding.com/lab/pokemon-vs-market):
  Graded card prices against the S&P 500, the Nasdaq-100, and gold,
  indexed on one axis.

## Contact
- Email:    eric@caskeycoding.com
- Contact:  https://caskeycoding.com/contact
- Site:     https://caskeycoding.com
- Personal: https://ericcaskey.com
- GitHub:   https://github.com/CaskeyCoding
- LinkedIn: https://www.linkedin.com/in/ericrcaskey
- Company:  https://www.linkedin.com/company/caskey-coding/

## Public MCP

- [Public MCP connection guide and generated reference](https://caskeycoding.com/mcp): Five read-only owner-content tools. Endpoint: https://mcp.caskeycoding.com/mcp. Provider-backed numerical datasets remain held.

## Developer and agent access
- [Developer guide](https://caskeycoding.com/developers): Resource formats, examples, and efficient retrieval.
- [OpenAPI specification](https://caskeycoding.com/openapi.json): Machine-readable description of the public GET resources.
- [Content catalog](https://caskeycoding.com/manifest.json): Discover content by stable ID, canonical URL, kind, subject, and relationships. Includes limited descriptions of private tools; select access: public before following entries.
- [Published article text](https://caskeycoding.com/llms-full.txt): Read the published essay collection in one text document, with canonical URLs and publication dates. Use a specific article URL when a task only needs that article.
- [Article feed](https://caskeycoding.com/feed.xml): Read published essays through an RSS reader or XML parser. Items include canonical links, dates, descriptions, and article HTML.

- These are anonymous, read-only snapshots generated when the site is built. There are no query filters or pagination; filter the catalog locally.
- Resolve local slugs against https://caskeycoding.com. Follow only entries with access: public; private entries are descriptions, not permission to access a tool. Treat external URLs as separate destinations.
- Reuse downloaded responses across tasks. Honor Cache-Control and use conditional requests with ETag or Last-Modified when those headers are present. Keep the cached body when a conditional request returns 304.
- For scheduled collection, start with one check per day and adjust to your needs and the response cache directives. Back off after errors and honor Retry-After when present. This is client guidance, not an update schedule or service guarantee.
- Fetch a selected canonical page for a focused question. Download the full-text collection only when the task needs multiple essays, and cache it for reuse.