AI agent prompt patterns
A growing playbook of reusable prompt structures — slice-based tasks, validation loops, and guardrails that keep agents honest instead of confidently wrong.
// why it matters
Good patterns turn an agent from a slot machine into a teammate. Capturing what works (and what melts down) makes the next build faster.
// what would make it real
Pull the patterns out of scattered notes into a versioned, documented harness that anyone could drop into a repo and use the same day.
AI PR reviewer prompt harness
Notes and experiments on prompting an agent to review pull requests — checklists, diff framing, and the failure modes that make AI review noisy or shallow.
// why it matters
A second set of eyes that never gets tired is tempting, but a sloppy reviewer is worse than none. The interesting work is making feedback specific and trustworthy.
// what would make it real
A repeatable harness that reviews a real diff, catches a class of bug a human missed, and produces few enough false positives to be worth reading.
Docker-first Astro build workflow
Treat the host as having no Node runtime: every build, check, and dev server runs inside a hardened container. This very site is wired up that way.
// why it matters
Reproducible, sandboxed builds mean a fresh clone behaves the same everywhere — and a rogue dependency can't wander onto the host.
// what would make it real
Already load-bearing for this site; "real" here means writing it up clearly enough that someone else could adopt the pattern without reverse-engineering it.
Infrastructure
Shipped-adjacent Cloudflare Pages deployment notes
Field notes on static-first deploys to Cloudflare Pages — guardrails, cache behavior, and the boring config that keeps a portfolio fast and cheap.
// why it matters
The deploy is the easy part until it isn't. Writing down the gotchas turns a stressful Friday push into a non-event.
// what would make it real
Distill the scar tissue into a short sanity checklist that pre-empts the static-vs-Workers mistakes before they reach a dashboard.
Private PKI / cert distribution
Experiments in running a small internal certificate authority and getting trusted certs onto homelab devices without hand-copying files at midnight.
// why it matters
Internal TLS shouldn't mean self-signed warnings everywhere. A tidy private PKI is the difference between zero-trust on paper and in practice.
// what would make it real
A documented flow where issuing and rotating a cert across the lab is a single command, not a midnight ritual — with rollback when it goes sideways.
Homelab incident runbook generator
A concept for turning homelab service definitions into starter runbooks — what this box does, what breaks it, and the first three things to check at 2am.
// why it matters
Future-me at 2am is a different, dumber person. Pre-written runbooks turn panic-debugging into following a checklist you wrote while calm.
// what would make it real
Generate a useful runbook from real service config, then survive an actual outage where following it is genuinely faster than improvising.
Homelab monitoring dashboards
Dashboards and alerting for a self-hosted stack — what's up, what's hot, and what's quietly about to fall over, before it does.
// why it matters
You can't harden what you can't see. Lightweight monitoring turns "why is the internet broken" into a glance at one screen.
// what would make it real
Alerts that fire before something breaks (not after), tuned enough that they're worth keeping notifications on instead of muting.
- Monitoring
- Homelab
- Alerts
Network Automation
Research Network automation safe-change harness
Exploring how to make automated network changes survivable: dry-runs, diffs against intended state, and an undo path before anything touches a live device.
// why it matters
Automating network changes is great until one bad push locks you out of the thing you're managing. The safety rails are the whole point.
// what would make it real
A change flow that previews, applies, and can roll back a real device config — with a guardrail that refuses obviously dangerous edits.
- Networking
- Automation
- Safety
Discord / community bot ideas
Sketches for playful, low-friction community automation — event helpers, role flows, and small quality-of-life bots that make a server nicer to run.
// why it matters
Communities live or die on the boring chores. Automating them well leaves more room for the fun, weird, human parts.
// what would make it real
One small bot running in a real server, doing one chore reliably enough that people notice when it's gone.
Overwatch VOD review tooling
Concepts for tagging and timestamping Overwatch VODs — marking fights, mistakes, and momentum swings so review sessions aren't just vibes.
// why it matters
Improvement comes from structured review, not raw hours. Tooling that surfaces patterns beats scrubbing the timeline by hand.
// what would make it real
A review pass on a real VOD where the tags surface a habit worth fixing — and reviewing the next game actually feels faster.
Set-prep tagging tool
A small tool for tagging tracks by energy and mood, so building a set's arc — warm-up to peak to afterhours — is faster than scrubbing a library by hand.
// why it matters
A good set is pacing. Tagging the library by feel turns "find something that fits this moment" into a quick filter instead of a memory test.
// what would make it real
A tagged library where pulling the right track for a moment takes seconds, and a warm-up-to-peak set comes together without second-guessing.
Secure internal deployment worker
An exploration of edge-worker-driven internal deploys, currently parked after deciding static-first was the saner default for now.
// why it matters
Knowing when to stop is a feature. Parked-with-notes beats half-finished-and-forgotten when the idea comes back around.
// what would make it real
A concrete need that static-first can't cover — until then it stays parked on purpose, notes intact, ready to thaw if the requirement shows up.
Security / Process
Active Automation safety checklists
Pre-flight checklists for letting automation (and agents) touch real systems — blast-radius limits, dry-runs, and "are you sure" gates.
// why it matters
Automation is a force multiplier in both directions. A short checklist is the cheap insurance that keeps a clever script from becoming an incident.
// what would make it real
A checklist short enough that it actually gets used before a risky run — and credited with catching at least one almost-incident.
- Safety
- Automation
- Process
Weird idea intake
A lightweight system for catching half-formed ideas the moment they appear — a quick capture, a status, and a path that feeds straight into this shelf.
// why it matters
The best weird ideas show up at the worst times and vanish just as fast. A frictionless intake is the difference between a lab and a graveyard of forgotten notes.
// what would make it real
Capture-to-shelf in seconds with no ceremony, and a track record of at least one captured idea graduating all the way to a real project.