Topic Guide
Agent Tooling on Your Own Machine
Every post in this hub is a wall someone hit running AI agent tooling on a real Windows dev machine, not a demo laptop. Claude Code will eat your C: drive with a 12.4 GB bundle you never asked for and can't find without knowing where to look. A dozen isolated Chrome profiles, each one the right call in isolation, will quietly hoard 48 GB until the machine freezes and forces a reboot. Windows will run your agent under a service identity that has never seen a desktop, and the tool that worked perfectly in your terminal will fail silently the moment something else launches it.
This is the cluster closest to the reader we're actually writing for: someone who has Claude Code open right now and just hit something that doesn't make sense. The posts below are grouped by what kind of wall it was. Some are pure hardware, disk space, junctions, temp directories that parallel agent sessions depend on more than you'd guess. Some are Windows being Windows: scheduled tasks, service accounts, session isolation, the kind of quirks that have nothing to do with AI and everything to do with the OS underneath it. Others are the tooling itself: session sprawl with no map, memory that doesn't survive a folder change, a browser that needs to be treated as infrastructure once you're running more than one agent at a time.
None of this is theoretical. Every fix below shipped because the alternative was a frozen machine, a silently failing service, or a session that forgot everything it knew the moment you closed the terminal.
33 posts in this guide, by DX
Start here
Moving Claude Code Off Your C: Drive: The 12.4 GB VM Bundle and the UWP Permission Wall
Claude Code ate my C: drive. Not a slow creep of logs and caches either. One file: a 12.4 GB claudevm.bundle sitting in AppData, the sandbox VM the desktop build spins up to run tools.
The machine fills up: disk, temp, junctions
-
You can't repoint a junction under live file handles (why I had to stage the .claude move)
My C: drive dropped to 230 MB free out of 231 GB, and one of the fattest things I could move off it was a 1.7 GB .claude directory full of transcripts and a chrome-debug profile.
-
Clearing %TEMP% on a machine running parallel Claude sessions is friendly fire
My C: drive hit 230 MB free out of 231 GB. I cleared the obvious junk and got 18 GB back.
-
My agents quietly hoarded 48 GB of Chrome profiles until the machine froze
Per-session browser isolation was the right call. Per-session isolation with no reaper filled my system drive, froze the machine, and forced a reboot. The fix was a fixed pool.
-
Measure the Bottleneck Before Blaming the Visible Workload
When a workstation slowed under concurrent work, the visible AI workload was an easy explanation. The useful lesson was to measure the whole system, test the hypothesis, and make reversible infrastructure changes that protect data and account for the actual workload.
Windows wants to be a person, not a service
-
claude --resume worked in my terminal and failed in my web app: the LocalSystem homedir trap
claude --resume <sid> worked when I ran it in my terminal and returned No conversation found when my web app ran the exact same command, on the same machine, against a session file that demonstrabl...
-
The console window that flashed every 30 minutes: Interactive vs S4U scheduled tasks
A PowerShell window flashed up on my desktop every 30 minutes, on the dot. Not an error, not a crash, just a black console that appeared and vanished while I was in the middle of something else.
-
Cloudflare Tunnel as a Windows scheduled task (and the visible-console bug I shipped)
I needed a Cloudflare tunnel to survive reboots without me babysitting it, so I wired it up to autostart and walked away.
-
My pinned app kept vanishing after every reboot, and it wasn't Windows being flaky
I pin Cursor to my taskbar. I reboot. It's gone. Not minimized, not hiding behind another window, gone.
-
My Agent's Chrome Was Running but Invisible: A Windows Session 0 Isolation Ghost Story
Chrome was up, CDP answered, automation worked fine, but the window was nowhere. The process was running in Session 0, a Windows desktop no human is ever logged into.
-
\"My internet feels unstable\" was a half-disabled IPv6 stack, not the ISP
The pings were clean. Zero packet loss to anywhere I tested, DNS resolved in milliseconds, and a TCP handshake to port 443 completed fast.
Running Claude Code at scale: sessions, memory, models
-
35 sessions and no map
I hit 35 Claude Code sessions against one repo in a day with no way to tell which were live. A SessionStart manifest and a generated dashboard fixed that.
-
My AI Agent's Memory Was Per-Folder, So My Clones Never Learned From Each Other
Claude Code's local memory is keyed to the working directory, so separate clones did not automatically share useful context. Durable project knowledge needed a reviewed, Git-tracked home in the repository.
-
Writing an AI_CONTEXT.md So Your Assistant Stops Rediscovering Your Codebase
A committed AI context file turns repeated codebase explanations and product guardrails into a reviewable project artifact that sessions can be instructed to read first.
-
After seven hours, I couldn't remember which model I was talking to
A 414-prompt pairing day where the active model was never surfaced. Session continuity and model-identity continuity are different things, and a footer line fixes it.
-
The token-killing proxy I tested instead of trusting my own discipline
After a single audit session burned 365,257 output tokens and 37,945,352 cache reads, I tested RTK, a proxy that compresses common command output before it reaches the model. The post is about why an IO-layer fix beats personal discipline once you are running parallel sessions.
-
RTK Ate My JSON and Hid My Merge Commit: When a Token-Saving Filter Lies to Your Agent
My token-saving CLI filter turned a JIRA payload into a schema sketch and hid a merge commit from my agent. The rule: lossless on any path that verifies state.
The browser is infrastructure, not a tab
-
I stopped babysitting the browser, then had to build the part that was missing
One agent's browser is a config change. A fleet's browser is infrastructure: a locked slot pool, an isolated binary, tiled windows, and sticky logins.
-
The Agent Owns the Browser, or You Do
Manual setup can leave an unseen human in an agent workflow. A lifecycle-ownership audit for every dependency, starting with the browser, reveals which steps are not yet safe to run unattended.
-
One-Flag Fix: --isolated Stops Parallel Playwright Clones From Deadlocking
Intermittent Playwright failures across parallel Claude Code clones traced to one shared Chromium profile lock. The --isolated flag removed the contention.
Tooling gaps the platform exposes
-
/add-dir Doesn't Reach Across Your Skills: The IDE-vs-CLI Claude Code Gotcha
In the Claude Code CLI, /add-dir grants file access but not skills: discovery stays rooted at the session cwd, so cross-repo skills belong at the user level.
-
I Was About to Write a digest_job.py. The Platform Already Shipped It.
Three custom scripts died in one review because the platform already shipped each one. The five-minute 'is this native yet' check that finds out before you build.
-
Your AI Skill File Is Part of Production Now
A merge conflict in an agent skill file delayed a hotfix and exposed an important change-control lesson: files that shape repeatable automated workflows need intentional review.
-
Hotfix branches need different rules than develop
Running /sync from a hotfix branch tried to merge develop in. On a legacy ASP-classic codebase with an active release in flight, that would have dragged unfinished work toward master. Tooling gap caught during a hotfix session, April 2026.
-
The workflow broke in the glue code, not the AI
I spent a morning fixing five workflow gaps I'd been compensating for by hand. None were model failures. All lived in the glue code between tools.
-
Reverse-Engineering Claude Code's Session JSONL (So You Can Mine Your Own Transcripts)
Claude Code logs every session to undocumented JSONL. Three parser traps: launch-cwd directory names, tool results labeled as user turns, and the nearly absent aiTitle.
-
The CAPTCHA lockout you trigger yourself: when a wrong API endpoint hammers you to a 403
A routine JIRA ticket-create came back 403 Basic Authentication Failure - AUTHENTICATIONDENIED with credentials I'd been using all session.
Working with Claude Code day to day
-
rtk Wasn't Broken, It Was Unreachable: How I Found 366 Silent Failures in My AI Agent Sessions
A silent PATH misconfiguration meant my token-optimization tool had actually run 10 times across 4,446 AI agent sessions. Here's how grepping my own transcripts like a flaky test suite found it, and two other live bugs.
-
I closed a client ticket from the wrong repo on purpose
Closing a client ticket from outside its own repo proved that shared agent skills belong one level above any checkout, where every session can call them.
-
The day my notes became infrastructure
Three AI sessions across three repos shared no memory, so every switch cost ten minutes of restated context. A markdown vault wired into session start fixed it.
-
Your Best Coder Will Be Your Lowest AI Adopter
The strongest coder I onboarded to Claude Code is my lowest-usage adopter. Capability and conviction are separate axes, and a second demo can't close the gap.
-
Closing a Ticket From a Machine That Never Cloned the Repo
A release workflow can use a remote control plane rather than a local checkout, but only when protected branches, approval evidence, provenance, deployment checks, and post-deployment verification remain explicit.
-
Six months of AI, measured in git instead of vibes
Everyone has a story about AI making their team faster. I audited mine: commits per person per year, tickets resolved, review comments. The team doubled its output in six months, and every person's gain landed in a different currency.
Hitting one of these walls in your own codebase or your own machine? Talk it through with us, or read the rest of the Build Log.