trellis-exec

2024-06-23 ~ 2024-07-12 · 19 days

Ambition Burnout

Suffocated by its own sophisticated design for AI agents, built at an unsustainable human pace.

“An AI orchestrator, built by a human at superhuman speed.”

Death Type

Self-inflicted AI Overlord Syndrome

Designed to orchestrate AI agents for perfect code, this project's human author pushed development at an inhuman pace. With `dist/` explicitly checked into the repo and Docker images building `playwright` and `@anthropic-ai/claude-code`, the infrastructure was ready for an AI revolution. The human, however, could not sustain 13.11 commits/day indefinitely, leading to its abandonment 8 days after its final commit.

Cause of Death

1. Hyper-accelerated human development

A single contributor made 249 commits in just 19 days, averaging an unsustainable 13.11 commits per day. This pace suggests a project driven by sheer will, not sustainable development.

2. Relentless refactoring of core logic

`src/runner/phaseRunner.ts` experienced a significant rewrite, with -639 lines removed and +286 lines added, indicating constant architectural churn rather than stable feature growth.

3. Infrastructure built for an empire, for an audience of none

The `docker/Dockerfile` defined `slim` and `browser` build stages, globally installing `playwright` and `@anthropic-ai/claude-code`, demonstrating an advanced deployment strategy for a project that died 8 days after its last commit.

Vibe Score

27/ 100

AI-assisted but human-driven

What They Did

A sophisticated 'phased execution harness for coding agents' designed to orchestrate Claude Code, it promised a 'judge-fix verification loop' and iterative refinement to prevent 'context rot'. It aspired to be both an npm CLI and a Claude Code plugin, all while supporting Docker with Playwright for browser automation, as detailed in its `CLAUDE.md` and `docker/Dockerfile`.

zod@types/nodeagnixmarkdownlint-cli2oxlintplaywrighttsxtypescriptvitevitest@anthropic-ai/claude-codegitca-certificates

Burnout Analysis

The single author, Rob McLarty, maintained an extraordinary pace of 249 commits in just 19 days, averaging 13.11 commits per day. Commit messages primarily focused on 'chore' (version bumps) and 'fix' (e.g., 'fix(container)'), indicating a relentless cycle of maintenance rather than creative expansion. Despite a reported 'low' burnout score of 0/100, this intensity suggests a project driven by sheer will, not sustainable development. The project's lifespan of 19 days itself is a testament to its rapid creation and equally rapid cessation.

Dependency Archaeology

A robust 10 development dependencies, including `zod` for schema validation and `vitest` for extensive testing, were installed. The Docker images globally installed `@anthropic-ai/claude-code` and `playwright` from the get-go. This suggested a full-fledged, battle-ready AI agent platform, yet it died 8 days after its final commit, having solved 0 real-world problems for 0 human users.

Autopsy: File Structure

├──CLAUDE.mdThe true README, detailing the AI's grand architecture and proving the AI knew more about the project than anyone else.

├──docker/DockerfileTwo build stages, `slim` and `browser`, for an app that didn't live long enough to need a browser, complete with `playwright`.

├──src/runner/phaseRunner.tsThe epicentre of refactoring, losing 639 lines only to gain 286, a net reduction in complexity perhaps, but at what human cost?

├──src/container/__tests__/containerLauncher.test.tsWhere 345 lines of new tests were added, ensuring perfection for a system few would ever run, a testament to thoroughness.

├──src/cli.tsThe human interface, updated with 279 new lines, only to be abandoned 8 days later, its commands forever unheard.

├──package.jsonBoasting 10 dev dependencies for code quality and testing, a true testament to building things 'the right way', even if briefly.

├──dist/The compiled output, explicitly committed to the repo, ensuring its existence even after the source repo's eventual demise.

└──src/runner/judgeRunner.ts326 lines of new code for the 'judge-fix verification loop' that ultimately couldn't save itself from abandonment.

Eulogy Stats

Total Commits: 249
Ambitious Adjectives: 5
Deploy Config: Yes
Estimated Users: 0 (unless you count the Claude Code agents it was designed to serve)

Last Words

“The final commit was a 'fix(container)', addressing an issue in the very system designed for perfection, rather than a triumphant 'release' for a project intended to be distributed as an npm CLI and Claude Code plugin.”