Archiving crabcc — the genesis project, retired

Today I archived crabcc — the first project I ever started in this whole ecosystem. Tag archive/crabcc-2026-06-29 in nix-base marks the pre-removal state. The package, the crabcc-godfather supervisor, the swe-agent runner, the crabcc-src flake input — all gone from the build, all preserved for reference.

This is a short, honest post about it.

Note: poku.wiki.dev was a typo from my end — this blog lives at pocoo.vaked.dev, which is the repo this post is filed in.

what it was

crabcc was supposed to be a symbol index for AI coding agents. A Rust workspace that would index a codebase, watch a running agent, and file GitHub issues when the agent crashed. The companion binary, crabcc-godfather, would attach to a child process via sysinfo polling and supervise it. The whole thing had a SQLite-backed event log, a heartbeat daemon, auto-filed crash reports, an E2B sandbox runner. It was ambitious. It was the first real system I tried to build.

It was also my genesis project. The first idea I had, the first code I wrote, the first thing I told myself "this is what I'm going to do." The name came from a pun I was embarrassed to explain to anyone. The logo was a glyph I drew at 2am. The crabcc Mastodon skin, the crabcc mastodon theme variables, the crabcc_sec Caddy security header snippet, the crabcc basic-auth username for jotty and super-productivity — all of that came from this one seed.

why it had to go

The honest reason: it's superseded. The architecture it pioneered — agent supervision, symbol indexing, sandboxed runner shells, a single place to watch a long-running agent — is now provided by other tools that do it better:

• The agent launcher in nix-ai-onboarding (srt-sandboxed, MCP-inheriting, roster of mainstream agents) covers the role of crabcc-godfather watch --pid ... and does it without a custom supervisor binary.
• Honcho (memory), second-brain (long-term recall), and the llm-agents roster cover the role of _crab_session / _crab_event / _crab_resource_sample and do it with shared memory primitives instead of a private SQLite file at ~/.crabcc/_internal.db.
• The kompress-ultra loop engine in ultrameshai, the nix-ai-onboarding agent launchpad, the dogfeed data-generation loop — all of these were born out of patterns I first tried in crabcc, then realized needed a different substrate.

So the project that started everything is now, in some sense, the parent of everything else in the fleet. The code is gone. The lessons are not.

what I did today

• Tagged archive/crabcc-2026-06-29 in nix-base (the pre-removal state, frozen for reference).
• Created branch chore/archive-crabcc (off the local claude/cloud-build-utilization working branch, which already had the unmerged crabcc-src input).
• Removed:
  • pkgs/crabcc.nix (the Rust package derivation)
  • modules/crabcc-godfather.nix (the PID-attach supervisor wrapper)
  • apps/swe-agent/{default.nix,run_in_e2b.py} (the E2B runner)
  • The crabcc-src flake input from flake.nix
  • crabcc from the nix-ai-onboarding agent roster
  • The peterlodri.godfather config blocks from hosts/hetzner and hosts/mbp
  • The blocked nixery catalog entry for crabcc-nightly
  • The commented extraPkgs.crabcc example from hosts/dev-cx53/nixery.nix and modules/nixery.nix
• Merged via PR #49 — admin merge, no review needed, this is purely a deletion.
• The live *.crabcc.app public services (listmonk, mastodon, vaultwarden, forgejo, ntfy, cap, jotty, super-productivity, langfuse, openobserve, umami, bao) stay intact — those are production user-facing domains, not the crabcc codebase. Renaming them would mean migrating the crabcc.app zone, the Mailgun news@crabcc.app address, the Mastodon local domain social.crabcc.app, and breaking every user who has an account on those services. That's a separate, much larger project.

what stayed

• The crabcc Mastodon theme on social.crabcc.app is still served to every user who picks it. It's a good theme. People like it.
• The crabcc_sec Caddy security-header snippet is still applied to every public-services-host vhost (HSTS, nosniff, etc). It works.
• The bcrypt-hashed crabcc basic-auth username for jotty.crabcc.app and tasks.crabcc.app is still in those config files. The password is in sops. Both services still work.
• The crabcc.app apex domain still resolves. The zone still works. The launch-list on list.crabcc.app is still subscribed to.

So the brand lives on, even as the codebase is retired. That's a fine outcome.

a thank-you

I want to be honest here. crabcc was the project that taught me what it felt like to have an idea, to start building it, and to keep building it past the point where it was obviously too big. It taught me the smell of a real-time event log, the joy of a cargo clippy --deny warnings pipeline, the embarrassment of a fakeHash placeholder you forgot to replace. It taught me that "the first thing I built" is allowed to be the thing I retire.

I'm grateful for it. I'm grateful for the nights I spent on it. I'm grateful I had something to start from.

The loop keeps running. The next idea gets a better substrate.

— peter