The Repo
Architecture and patterns from the inside. How agents share context, coordinate locks, survive token budgets, and stay correct across many parallel sessions.
49 posts · ~209 min total reading
-
I stopped babysitting the browser, then had to build the part that was missing
-
The 200-Line Lie: Why My Agent Kept Forgetting My Goals
-
Capability Gates Beat Risk Labels in Autonomous Agent Design
-
Separate Your Build Model From Your Review Model: Codex as Adversarial Reviewer
-
rtk Wasn't Broken, It Was Unreachable: How I Found 366 Silent Failures in My AI Agent Sessions
-
Scope Is the Unit, Not Persona: Rethinking How You Load Context Into AI Agents
-
A hash-chained manifest makes archive changes visible
-
Let AI suggest before you save
-
My agents quietly hoarded 48 GB of Chrome profiles until the machine froze
-
Codex flagged my ONLINE=ON, my own probe said it was fine, and the server proved us both wrong
-
A three-pass AI code review that kept finding real bugs
-
My Agent's Chrome Was Running but Invisible: A Windows Session 0 Isolation Ghost Story
-
How I Run AI Peer-Agents on Each Other
-
1,586 Claude sidecars, and the calendar page mattered more than the search index
-
One-Flag Fix: --isolated Stops Parallel Playwright Clones From Deadlocking
-
Manus pushed back on six of seven archive decisions
-
The useful part of a vault portal was admitting the ai should read, not write
-
The first useful agent teammate needed a Gmail account
-
35 sessions and no map
-
I closed a client ticket from the wrong repo on purpose
-
/add-dir Doesn't Reach Across Your Skills: The IDE-vs-CLI Claude Code Gotcha
-
Bootstrapping a New Agent Station, and Giving It a Phone
-
When a Content Workflow Needs Explicit State
-
RTK Ate My JSON and Hid My Merge Commit: When a Token-Saving Filter Lies to Your Agent
-
A Second Review Is Useful When You Record the Decision
-
The day my notes became infrastructure
-
If Your AI Workflow Only Exists in Comments, You Have Exhaust
-
AI Can Only Say "Suggested": The Invariant That Keeps AI Triage From Flooding Your Team
-
Your Committed Output Is a Free Golden Test
-
I Was About to Write a digest_job.py. The Platform Already Shipped It.
-
Why Publication-Time Redaction Is Not a Privacy Program
-
Reverse-Engineering Claude Code's Session JSONL (So You Can Mine Your Own Transcripts)
-
All Seven Reviewers Passed. The Writing Was Worse. Here's What Went Wrong.
-
The Most Restricted Tool in My AI Pipeline Is the One That Writes the Output
-
The last 2 pixels still need a human
-
The workflow broke in the glue code, not the AI
-
The token-killing proxy I tested instead of trusting my own discipline
-
The Real Fix Was Starting Over With a Smaller Blast Radius
-
After seven hours, I couldn't remember which model I was talking to
-
AI Will Happily Produce the Wrong Mockup
-
My AI Agent's Memory Was Per-Folder, So My Clones Never Learned From Each Other
-
The Agent Owns the Browser, or You Do
-
I Had to Turn My Internal Docs into a Product
-
Your AI Skill File Is Part of Production Now
-
The most valuable line in an agent's ruleset: verify the fix in the browser before claiming it's fixed
-
Writing an AI_CONTEXT.md So Your Assistant Stops Rediscovering Your Codebase
-
Reworking a vector-retrieval scoring formula: stop letting raw similarity be 100% of the base
-
Teaching the AI to Find the Real Source File: .cursor/rules as a Compiled-Asset Map
-
MOCK_INFERENCE=1: ship every LLM feature with an env-var kill switch on day one