Skip to main content

​
The Story

We built a system where GitHub issues become running code — automatically. FORGE reads a project board, picks the next unblocked work, writes implementation prompts. TOMMY takes those prompts, SSHes into a build server, and runs Claude Code CLI to write the actual code. On success, it closes the GitHub issues. The catch: the 34-node TOMMY pipeline had never been tested end-to-end. Phase 2 verification — a simple dry run — took 15 webhook iterations to get right. Each iteration peeled back a new layer: type coercion bugs, credential mismatches, architectural anti-patterns, a billing surprise, and an undocumented SplitInBatches v3 behavior change. The entire debugging session was done without opening the n8n UI once. Every diagnosis: curl to fire the webhook, curl to pull execution logs, python3 to parse the trace. When a fix required the UI (node settings, rewiring), the instructions were precise: “change this dropdown”, “delete this wire”.

​
Key Messages

  • Autonomous coding pipelines are real — not a demo, not a prototype. FORGE×TOMMY operates on a live GitHub project with 22 blocks and 94 tasks.
  • The hard part isn’t the AI — it’s the orchestration plumbing. 11 of 15 bugs were n8n wiring issues, not LLM failures.
  • API-driven debugging scales better than UI clicking. The execution trace contains everything — inputs, outputs, errors, timing — in structured JSON.
  • Cost tracking from day one — each dry run: 0.02−0.04.Totaldebuggingsession: 0.02-0.04. Total debugging session: ~0.16. The goal is near-zero autonomous operation cost.

​
Content Angles

Twitter/X thread: “I just debugged a 34-node autonomous coding pipeline through 15 iterations in one session. The LLM worked perfectly. Here’s what actually broke: [thread of the 15 bugs]” LinkedIn post: “Building autonomous dev tools means 80% plumbing, 20% AI. Here’s what 15 iterations of end-to-end testing taught me about n8n workflow architecture.” Technical blog: “n8n Anti-Patterns: What I Learned Debugging a 34-Node Pipeline Via REST API” — the debugging methodology as a standalone reference piece. Dev community (HN/Reddit): “Show HN: FORGE×TOMMY — GitHub issues → running code, zero human intervention” with the iteration story as the honesty hook.

​
Status

  • Phase 1 (FORGE solo): Complete, verified
  • Phase 2 (TOMMY dry run): 15/15 bugs fixed, pipeline runs end-to-end
  • Phases 3-7: Pending (live execution, integration, edge cases)