anupamchugh

Building my dev workspace into an operating system. Not metaphorically — structurally.

  10 MCP servers as device drivers (exchange APIs, browser automation, Apple docs, issue tracking).
  200+ skills as prose runbooks that compose system calls. Agent-mail for IPC between parallel
  agents. A drift detector called "wobble" that scores skill stability using bias/variance analysis.

The interesting failure mode isn’t just “one bad actor slips through”, it’s provenance: if you want to “denounce the tree rooted at a bad actor”, you need to record where a vouch came from (maintainer X, imported list Y, date, reason), otherwise revocation turns into manual whack-a-mole. > Keeping the file format minimal is good, but I’d want at least optional provenance in the details field (or a sidecar) so you can do bulk revocations and audits.

Pinning exists, but the interesting part is signal quality: macOS gets consistent “urgency” signals (QoS) from a lot of frameworks/apps, so scheduling on heterogeneous cores is less guessy than infer from runtime behavior.

Bad code crashes. You fix crashes. Acceptable code fails by doing nothing. You don't fix nothing.

Good code isn't dying. The cost of bad code just went up.

This is the real story buried under the simulation angle. If you can generate reliable 3D LiDAR from 2D video, every dashcam on earth becomes training data. Every YouTube driving video, every GoPro clip, every security camera feed.

Waymo's fleet is ~700 cars. The internet has millions of hours of driving footage. This technique turns the entire internet into a sensor suite. That's a bigger deal than the simulation itself.

"Stack Overflow that reads your codebase" — perfect. But Stack Overflow is stateless. Agent sessions aren't.

One session's scaffold assumes one pattern. Second session scaffold contradicts it. You reviewed both in isolation. Both looked fine. Neither knows about the other.

Reviewing AI code per-session is like proofreading individual chapters of a novel nobody's reading front to back. Each chapter is fine. The plot makes no sense.

Agent teams nuke your tmux layout. The fix is one line: new-window instead of split-pane. Filed as a bug.

I have that in my project... comments to track work as specs.

This is a very early research prototype with no other inter-agent communication methods or high-level goal management processes."

The lock file approach (current_tasks/parse_if_statement.txt) prevents two agents from claiming the same task, but it can't prevent convergent wasted work. When all 16 agents hit the same Linux kernel bug, the lock files didn't help — the problem wasn't task collision, it was that the agents couldn't see they were all solving the same downstream failure. The GCC oracle workaround was clever, but it was a human inventing a new harness mid-flight because the coordination primitive wasn't enough.

Similarly, "Claude frequently broke existing functionality implementing new features" isn't a model capability problem — it's an input stability problem. Agent N builds against an interface that agent M just changed. Without gating on whether your inputs have changed since you started, you get phantom regressions

  Agent teams in this release is mcp-agent-mail [1] built into
  the runtime. Mailbox, task list, file locking — zero config,
  just works. I forked agent-mail [2], added heartbeat/presence
  tracking, had a PR upstream [3] when agent teams dropped. For
  coordinating Claude Code instances within a session, the
  built-in version wins on friction alone.

  Where it stops: agent teams is session-scoped. I run Claude
  Code during the day, hand off to Codex overnight, pick up in
  the morning. Different runtimes, async, persistent. Agent
  teams dies when you close the terminal — no cross-tool
  messaging, no file leases, no audit trail that outlives the
  session.

  What survives sherlocking is whatever crosses the runtime
  boundary. The built-in version will always win inside its own
  walls — less friction, zero setup. The cross-tool layer is
  where community tooling still has room. Until that gets
  absorbed too.

  [1] https://github.com/Dicklesworthstone/mcp_agent_mail
  [2] https://github.com/anupamchugh/mcp_agent_mail
  [3]
  https://github.com/Dicklesworthstone/mcp_agent_mail/pull/77

i would like to see a mobile app for this to vibe code on the fly. currently the DIY options are good but clumsy UX wise as they’re workarounds. if i can open worktrees from an ios app it would be great.

When do you actually need to open Xcode if you have XcodeBuildMCP [0]?

I haven't opened Xcode in months. My terminal: Claude writes code. build_sim. launch_app_sim. screenshot describe_ui.

What still requires Xcode: Instruments profiling, Signing/provisioning

For UI iteration, describe_ui returning the accessibility tree might actually be more useful to an agent than a preview screenshot.

Following up - I built a tool "wobble"[1] to measure this: parses ~/.claude/projects/*.jsonl session transcripts, extracts skill invocations + actual commands executed, calculates Bias/Variance per the paper's formula.

Ran it on my sessions. Result: none of skills scored STABLE. The structural predictors of high variance: Numbered steps without clear default, Options without (default) marker, Content >4k chars (overthinking zone), Missing constraint language

[1] https://github.com/anupamchugh/shadowbook (bd wobble)

This solves distribution well. Curious about the change propagation story though - what happens when you update your .ai/ source and tools have cached/transformed versions?

I ran into this building a spec/skill sync system [1] - the "sync once" model breaks down when you need to track whether downstream consumers are aware of upstream changes.

  [1] https://github.com/anupamchugh/shadowbook

Not a big use case outside of criminals and refugees" - the refugee use case alone matters to hundreds of millions living under capital controls or currency collapse. Argentinians escaping peso devaluation aren't criminals. Ukrainians moving value during invasion aren't edge cases.

Whether Bitcoin is good at this is debatable. Whether it's a real use case isn't.

Atleast they're explicit about having a SOUL.md. Humans call it personality, and hide behind it thinking they can't change.

Wow. And just like that fliki.ai and similar products have been sherlocked. Great time to be a creator, not the best time to be a product developer, production designer

I’m more curious how the journal suggestions API will play out and change our relationship with third party apps.

There’s lots of conflicts of interests beyond Adam and his Poe AI. Yes, he was building a commerical bot using OpenAI APIs, but Sam was apparently working on other side ventures too. And Sam was the person who invested in Quora during his YC tenure, and must have had a say in bringing him onboard. At this point, the spotlight is on most members of the nonprofit board

FWIW, the board structure is more in the spotlight, especially Adam's, since Quora's Poe chatbot uses OpenAI's APIs. And on top of that, Sam was an investor in Quora (2018).

But the Board members of the non-profit company have been leveraging OpenAI's APIs in their other business endeavors -- like, Adam's Poe chatbot for Quora. More details on board's timeline:

https://loeber.substack.com/p/a-timeline-of-the-openai-board

Related read: https://www.techtris.co.uk/p/openai-set-out-to-show-a-differ...

Interesting that Sam is also an investor in Quora: https://blog.samaltman.com/quora

It’s convulated

At least there's no accurate way for a search engine to check for originality. It's like asking a machine to evaluate other machines.

Here's the top-most featured snippet when I google if programming languages had honest slogans: https://medium.com/nerd-for-tech/if-your-favourite-programmi...

Half of the above post is plagiarised from my 2020 post: https://betterprogramming.pub/if-programming-languages-had-h...

This reminds me of the current state of blogs vs platforms. The issue with streaming platforms in various formats is that they rake in more money over their lifetime than the creators, simply by acquiring copyrights and hosting it.

There is a “sign up for membership” soft paywall on all stories.

Does something like a blink, wink tracker for swiping UI pique your interest? I built a PoC a while back: https://github.com/anupamchugh/BlinkPoseAndSwipeiOSMLKit

This guy is spamming Medium everyday. Has been banned on a few occasions, but creates a new account, writes nonsense and then shares it on HN and Reddit

Correction: I used ChatGPT to discover new words and rephrase the story myself.

Here's the link to my non-paywall story for references:

https://medium.com/humor-bytes/i-your-nba-highlights-broadca...

On the topic of assistance, where do detectors draw the line between AI-augmented, aided, and generated writing?

I tried this on my recent AI satire article. While ChatGPT detected my story as AI (because I wrote it like a machine), GPTZero gladly says it's human written.

ChatGPT's response: https://imgur.com/a/u7iBBaX

GPTZero's response: https://imgur.com/a/89icz2X

To be clear, my story was ChatGPT-assisted. But I wonder why ChatGPT couldn't detect it correctly like GPTZero?

That’s the state of Web 2.0 I guess? Google also plans to rank AI-aided good content lower than thin blog posts that have contributed to SEO spam.

It’s like, earlier we had ads and subscriptions. Now AI is a new business model. But I think it needs Web3.0