Posts
-
Persistent Workspaces in Terminal Multiplexers
-
One Admission Gate for Three Tools, and a Reply Channel Back to the Model
-
Format Contamination: Why Your AI Agent Stopped Calling Tools
-
Two Months of Parallel Tool Calls Failing on a Gemini Gateway
-
Put Subagent Spawns Behind an Approval Gate
-
Give a Handoff One Planning Turn, Then Switch Back
-
The Eight-Line Bug at the Bottom of a $40 Detour
-
Put Agent Skill Creation Behind a Confirmation Gate
-
When PuTTY's Saved Session List Stops Scaling
-
The Second Copy: Why Reusing Your Agent's Internal UI Crashes It
-
The Console Flash and the Watchdog That Lied About It
-
Your Agent Started a Background Process. Where Did It Go?
-
Building Attention Pause for Herdr and OMP
-
Git-Backed Local Workboard
-
Running Hindsight as OMP's Memory Backend on Windows
-
When a New Terminal Tab Is the Wrong Tab
-
Ambient Attention Without Interruption
-
Two Layers Cannot Share a Mouse
-
Why taskset Could Not Fix a DPDK Affinity Failure
-
One Shortcut, Five Launch Modes: Right-Click Presets on Windows
-
Agent Zero: A Control Group for Your Coding Agent
-
Zen at Window 13: A Prompting Governor for Windows
-
A Cost Circuit Breaker for Long-Running OMP Sessions
-
Your Review File Exists. The Review Still Didn't Happen.
-
Search Should Widen, Not Explode
-
The File Path Was Fine. The Invisible Hyperlink Target Wasn't
-
Before the Agent Acts: Designing a Prompt-Intake Layer for Coding Agents
-
OMP's hidden foreground agents: /tan, Agent Hub, and the ownership trap
-
Your AI Agent Does Not Need a Hosted Reader to Turn Web Pages into Markdown
-
One Browser, Two Identities: Route Links to the Right Edge Profile
-
Don't Screenshot Yourself: Capturing Occluded Windows on Windows
-
One registry, two modes: declarative remote-host access for you and your agent
-
You Can Only Optimize What You Can See: Measuring an Agent's Context Footprint
-
A warm proxy in front of a remote MCP server: the reentrant-client singleton
-
Stop your agent repeating the same web search
-
The $/1M-ctx-growth meter: measuring what a coding agent's context is actually costing you
-
Clickable file paths in your terminal: why they work in your AI agent but not your shell
-
When your agent's auto-learned skills start competing with each other
-
Running commands in WSL from Windows: stop fighting the quoting
-
Placing windows by title on Windows: nircmd lands them wrong, and why the fix is DPI awareness
-
Building a quiet /pause for OMP — and discovering how much of the feature set I didn't know
-
Fixing a Windows→WSL shim for multi-line scripts (and a Python detour I shouldn't have taken)
-
A terminal theme bug that kept coming back — and why the real fix was two directories away from where I was looking
-
session-reminder-omp: periodic reminders inside a coding-agent session
-
Auto-AFK: detecting real-world idle instead of typing /afk
-
When your AI agent's Slack search forgets the link: building a deterministic MCP CLI
-
A TZ environment variable silently corrupts every local timestamp Python computes on Windows
-
Diagnosing a delayed Wi-Fi failure: correlating power events with NCSI in netsh wlan show wlanreport
-
Restart-safe extension state: from persisted checkpoints to pure recomputation
-
The OMP memory feature that quietly cost me 33,000 tokens of context
-
The next OMP + Sonnet failures: replay, legacy thinking shapes, and poisoned sessions
-
Teaching an AI coding agent to know what time it is (the Jyotish way)
-
Clock-time vs elapsed-time triggers for OMP's retro extension
-
Fixing OMP turn-2 replay failures without disabling reasoning
-
One ~/.claude, two environments: making Claude Code hooks work in WSL and Windows Git Bash
-
Why I removed per-session goal tracking (six weeks later)
-
Auto-sync agent repos at session start
-
Porting OMP productivity mechanisms to Grok — AFK, suggester, retro
-
Clock-aligned retrospection in Claude Code — no WezTerm required
-
agent-suggester-omp: periodic Stoic/Gita/first-principles nudges for OMP
-
Three traps in one PowerShell profile edit: OneDrive redirect, wrong path, and the UTF-8 BOM cliff
-
Making terminal URLs clickable in an AI coding agent — four approaches, one that works
-
Building a /screenshot command for Claude Code on Windows — what doesn't work and what does
-
Model fit beats model rank
-
Extending claude-code-timestamps: three things that didn't work and what to do instead
-
Building a daily goals widget for OMP — a full extension walkthrough
-
Debugging a BLE mouse that pairs but won't work on Windows 11
-
Building a 234-page PDF book from a 795-file mkdocs site
-
Mining AI agent session logs for skill candidates
-
Scanner to Agent: integrating a flatbed scanner as an AI coding agent extension on Windows
-
Claude Code timestamps: 12 lifecycle events, one script
-
Building a /screenshot command for OMP — one option away from working
-
AFK toggle for AI coding agents
-
No Broken Windows for AI coding agents
-
Switching to WezTerm from Windows Terminal
-
Session review for AI coding agents
-
Slow-motion mode for AI coding agents
-
Per-session goal tracking for AI coding agents
-
Personal Agentic Process — engineering discipline for human-agent pairs
-
Migrating 40 repos to ghq on Windows: what went wrong and what fixed it
-
Clock-aligned retrospection for AI coding agents
-
Adding a bare-metal worker node to OpenShift SNO on Supermicro: 6 gotchas that cost me 4 hours
-
Adding timestamps to AI coding agent conversations
-
Break-time retrospection for AI coding agents
-
Running pi-dashboard with Oh My Pi on Windows
-
Fast SSH on Windows via PuTTY connection sharing
-
Fixed-time breaks with Stretchly on Windows
-
Patterns of Agentic AI
-
The Complete Guide to Git Symlinks on Windows: Why They Break and How to Fix Them
-
Publishing to dev.to Programmatically in 2026: What Actually Works
-
Getting Unix Tools to Work in PowerShell: A Debugging War Story
-
Welcome to Jekyll!
subscribe via RSS