How Planwright works

Your agent turns what you already have, such as a meeting transcript, a spec, a feature request or a Jira epic, into initiatives and objectives. You steer the conversation and approve the result.

An initiative sets a goal and a target date. Each objective states one outcome, not a task, with acceptance criteria a reviewer can answer yes or no to. Planwright scores every draft's clarity and flags vague ones before anyone claims them. Context files carry the project knowledge every agent should read first.

An initiative's objectives and target date are also what the roadmap and its forecast are drawn from, so approving the plan is what gives you a date.

A coding agent claims a scheduled objective and gets a mission brief: the objective, how often recent work in the same strategic area passed review first time, the most common reasons it was rejected, and whether to work on a branch or directly.

It posts a plan, confirms the plan matches the objective before editing anything, builds and tests against the acceptance criteria, and attaches its diffs and test results. It can't move work on the board itself: it asks, and the server checks every transition.

Finished work waits in the acceptance queue. Each change carries risk signals, such as sensitive paths touched, new dependencies and churn hotspots, so reviewers can triage the queue and look hardest where it matters.

The reviewer judges the work against its acceptance criteria and accepts it, or rejects it with a reason. Agents can never approve their own work.

When an accepted objective's pull request merges to your production branch, Planwright releases it. An initiative counts as shipped once all its objectives have been released. Release notes and the public changelog are written from completed objectives, so nobody writes the weekly summary.

Every result feeds back. Rejection reasons are grouped into six categories, and the most common ones for an area go into the next agent's mission brief. Rejection notes become suggested updates to context files. Planwright tracks first-pass acceptance by agent and area, where work waits, and how far delivery drifted from the original intent. Every finished run and review also updates the averages the roadmap forecast uses, so its dates track how the team really works.