José Miguel Gómez

Back to the home page

Forgeline

From requirement to reviewed PR, governed end to end

The code lives in a private repository; this sheet and the demo, when there is one, show the work.

Draft. Written from the program notes and pending the author’s review.

Status
In production
Type
Agentic workflow
Running since
March 2026
Role
Architect and sole engineer
Domains
software delivery, agents
Cast
Python, Claude Code, GitHub Actions, LangGraph

Context and problem

Turns a feature request into a reviewed pull request: DDD spec, hexagonal code, tests and a security audit.

Turning an approved requirement into a reviewed pull request still takes days of hand-offs between people and tools (assumption documented in the repository).

Screens

How it was built

Built in 3.3 h · USD 72.54 of AI · 110 tests · 4 audits

Source: Trysac/03-forgeline docs/development-report.md (lines 10, 12, 65), dated 21 November 2026.

Architecture and interfaces

Provides
events pr.*, api forgeline-api
Consumes
mcp specgate (specgate)
Shared keys
spec_ref, pr_ref

Results with dated evidence

Outcomes as stated in the manifest; the measured figures are the ones with a date.

  • Requirement to reviewed pull request without a manual hand-off, behind human approval

Case study in preparation

The full case study adds these sections when it is published:

  • Constraints
  • Architecture diagrams (C4 context and containers)
  • Three key decisions
  • What went wrong
  • Demo video
  • What I would do differently

In the project mesh

  • Specgate, zero effort: interface specgate provided by one and consumed by the other
  • Resolvo, low effort: shared resources llm-gateway; a standard interface exists
  • Revela, low effort: shared resources llm-gateway; a standard interface exists
  • Sublingua, low effort: shared resources llm-gateway; a standard interface exists
  • Vigilo, low effort: shared resources github; a standard interface exists

Highlights

durable agents, human in the loop, spec driven