AIM v3.2.0 — Fewer approvals. Clearer next steps.

Fewer approvals. Clearer next steps. AIM 3.2 brings direction and first-plan approval together, then delivery acceptance and verified Epic closure together. A straightforward single-Increment Strict Epic has two ordinary decision moments, with the same review and whole-Epic verification requirements.

• Present the Epic goal, first Increment, remaining work and risks before a combined start decision.
• Prepare the whole-Epic recommendation before offering delivery acceptance. Users can accept only the Increment, or also approve the explicitly offered close, continuation or scope split.
• Plan the next Increment after accepted continuation; Strict still requires approval of its plan before implementation.
• Share versioned proposals and next-step information across chat, CLI and AIM UI.
• Register decisions with immutable receipts, a workspace process lock, an atomic state commit and idempotent recovery from interrupted calls.
• Check closure readiness without fabricating acceptance; reject changed proof, changed product inputs and stale proposals.
• Preserve committed acceptance, legacy gate semantics and bounded Auto/Portfolio authority.
• Fix Continue so saved pauses and blockers take precedence over change requests, and an accepted split leads to its recorded work rather than another request for the same disposition.

Compatibility: runtime contract 2.0, runtime-state schema 1.0 and installer manifest 1.0 remain unchanged. The generated public skill package revision is 17. Runtime-schema decision fields are optional; existing runs and gate responses retain their original meaning.

Validation: 497 regression tests, including two connected headless CLI scenarios using isolated fixtures and fresh processes; the external review's two reproduced Continue issues are covered by regression tests. Release gates also check generated package integrity, official skills CLI discovery/installations, documentation, engineering evidence and deterministic publication artifacts.

Migration: run npx skills update agile-iteration-method --yes, reload the skill session and restart AIM UI. Adaptive installations use /aim upgrade. No runtime-state migration is required; older approvals never silently acquire combined-decision authority.

Known limitations: end-to-end runtime savings and user understanding have not yet been measured through real coding-agent sessions. The local registration microbenchmark measures only the helper function, not the full agent turn. Native execution and specialist support depend on the host; AIM UI remains Beta. Review and whole-Epic verification remain required. See the combined-decision contract and implementation evidence.

Get started · Full changelog