AIM
Turn your intent into a complete product.

Suggested announcement copy · AIM 4.7.7 · 29 October 2026.

AIM 4.7.7 clarifies /aim graph reporting. Instructions require agents to identify the selected delivery and its source, use only recorded dependencies as factual arrows, distinguish recommendations from facts, and surface observations and conflicting labels. A bounded actual-command regression preserves those source facts; it does not establish universal model adherence.

Describe the product you want. AIM plans, builds and verifies the complete agreed result with your coding agent. Ship is the default. Choose Loop when you want to assess useful Done Increments and shape the next step through recurring feedback.

Follow the product in AIM UI. Inspect Stories, the current delivery, review evidence and blockers in one local control room. Graph engineering connects ownership, dependencies and proof; the Graph view shows the recorded relationships. An owner agent needs your explicit mandate for the product decisions it may make.

AIM agrees quality goals before implementation. A separate reviewer challenges the first working core before costly expansion. Real user journeys and retained raw results support the final assessment. A maintained verification route helps later changes repeat the right checks.

Repository knowledge, compact checkpoints and relevant shared lessons help agents reuse work that still applies. Learning is part of ordinary delivery, without a manual reflection step. The Product Owner owns Epic outcomes and priority; the Technical Delivery Owner owns Stories, criteria and the increment. Acceptance stays with you.

The retained Enigma M3 results compare GPT-6 Luna with AIM, scoring 96.5/100 in the latest assessment, with GPT-6 Luna without AIM at 49.8/100. The earlier AIM result was 93.5/100. The same model, high reasoning effort, brief and scoring weights were used, but the runs took place on different dates under different delivery conditions. This is a historical comparison, not a fresh paired experiment or a general quality or cost guarantee. The latest version of AIM includes later changes outside this score; the assessed native delivery remained unfinished. Exact versions and evidence are in the evaluation report.

For Codex, GitHub Copilot and Claude Code. AIM UI remains Beta. The public skill command installs the public source; see GitHub Releases for available releases.

Explore: https://github.com/joneri/agile-iteration-method
Evaluation: https://github.com/joneri/agile-iteration-method/blob/main/docs/features/aim-enigma-evaluation.md
Install: npx skills add joneri/agile-iteration-method --skill agile-iteration-method

Turn your intent into a complete product with AIM. Ship by default; choose Loop for recurring feedback. Follow Stories, dependencies and evidence in AIM UI. For Codex, Claude Code and GitHub Copilot. https://joneri.github.io/agile-iteration-method/
