Foreman
Most "AI does my ticket" tools stop at generating a diff and hoping it's right. Foreman, described by its creators as an eve Software Factory, instead builds an assembly line of four specialized agents around a single work item, and deliberately stops short of merging anything without a human decision.
Work can enter the pipeline from six places: a labeled GitHub issue, an @-mention, a Linear ticket, a failed CI check, a submitted pull request, or a local development session. Once triggered, four agents run in sequence. A Classifier triages the work item and determines its scope. An Analyst turns that scope into a plan with testable acceptance criteria — concrete, checkable conditions rather than a vague description. An Implementer executes the plan inside its own sandbox, runs the verification checks itself, and pushes a branch. A Reviewer then independently checks each acceptance criterion against the actual resulting code, rather than trusting the Implementer's own account of what it did.
The pipeline's output is always a draft pull request. Nothing merges automatically — the two decisions left to the developer are what to hand the pipeline in the first place, and whether the result is good enough to ship. Foreman is open source and is distributed as a template repository deployable on Vercel.
Foreman sits in Assess. The four-stage classify/plan/implement/review pipeline and the hard stop at draft-PR are a coherent answer to the usual complaint about autonomous coding agents — that they ship confidently wrong changes — but the project itself is six days old at the time of this entry, with no track record yet. Outer position reflects both that immaturity and the setup cost: getting real signal requires deploying the template and wiring it into an actual issue tracker and CI pipeline, not just reading the docs. No first-person use exists, which is the gate to Trial.
