Forespec Adds Domain-Aware Checks to AI Coding Workflows

Summary

Forespec is an early-build GitHub project that codifies the domain-specific requirements a senior engineer might catch when AI coding tools miss non-obvious issues. It currently supports five software archetypes and provides a reasoning verifier, planning and interrogation commands, a pull-request gate, and a greenfield on-ramp. The tool can generate a gotcha-aware build specification before implementation, grade the resulting code against backbone checkpoints, identify required pieces that are missing, and compare checkpoint movement between runs. It is intended to run inside the coding agent, with Claude Code as the first supported integration; that path uses the user’s existing Claude Code subscription and does not require a separate API key. A direct API path is also available with Anthropic credentials, while the mock keyword adapter cannot certify a merge. Forespec stores calibration data locally behind a pattern/instance boundary, keeping project-specific data on the user’s machine according to the repository description. The project reports 0 false-greens across 152 critical-bad plugin trials, with a 95% rule-of-three upper bound of 2.0% and complete agreement across two independent runs, but notes that this measures a snippet grading contract rather than the agent’s repository-navigation advantage. In an audit of eight previously unseen public repositories, the plugin produced 147 verdicts, including 18 findings, and the project says every verdict cited a file and line; 12 of the 18 findings were hand-verified and none were fabricated. The repository also describes a production ecommerce audit that found a missing Stripe idempotency key and a refund nonce issue. Newer SaaS, AI-app, and BaaS archetypes remain first-pass validated, with full validation still pending.