New plugin enforces spec-first workflow for AI coding agents before code is written
A Show HN post introduces a plugin that forces AI coding agents to write a formal specification—goals, cited file-and-line facts, steps, and done criteria—before any code is generated. The spec is checked by a script for valid file references, then reviewed by a separate agent with no prior chat context before implementation begins. Only after tests and build pass does the tool push a branch and open a pull request containing the spec.
GoKawiil's interpretation of the reporting above, not reported fact.
The approach addresses a common complaint about AI coding agents: that they jump straight to code without grounding in verified facts about the codebase, leading to hallucinated references or unreviewed assumptions. By splitting the process into spec creation, independent review, and gated implementation, the tool's creator appears to be aiming to make agentic coding more auditable and trustworthy, though it's unclear from the post how this holds up on larger or more complex codebases.
- The plugin introduces a four-stage workflow: create-spec, review-spec, apply-spec, and deliver-spec.
- A validation script confirms cited file and line references exist before a second, context-free agent critiques the spec.
- Code only ships after tests and build pass, with decisions logged and the spec included in the resulting pull request.
Source: spec-first.dev, 2026-10-27
Published there as: “Show HN: Specification First Agentic Development Plugin and Prove”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.