AIWG 2026.10.18 - Undo a setup and get answers you can trust

AIWG 2026.10.18 helps you undo a framework setup, see the same pinned rules wherever you ask, and judge whether resource numbers are fresh enough to use. It also explains why a worker request was refused, so you know what needs attention.

Undo a framework you added

“Remove the framework I added. Show me what will go, and keep anything I edited.” Your assistant can preview the removal, then remove the framework you name:

aiwg remove <framework> --dry-run
aiwg remove <framework>

Removal now covers the files, settings and hooks recorded when you deployed the framework. It keeps your edits and anything another installed framework still uses, and tells you what it preserved. If some changes cannot be undone, AIWG keeps their record so a later retry can continue.

That includes comments you changed in YAML or TOML settings. You can review the preview before removing anything. The scope and recovery guide is the place to start when deciding what to keep.

Get the same pinned rules wherever you ask

“Find the skill for this task, then show me the rules it requires.” Start with your task, then inspect the skill returned:

aiwg discover "<your task>"
aiwg show skill <skill-name>

A rule that a skill, agent or flow says it needs now shows up the same way in search results and in its details, no matter which way your assistant looks it up. You can choose a capability and inspect its requirements without getting conflicting answers. The discovery guide explains how to find and inspect what you need.

See when resource numbers were collected

“Check whether this machine has room for a helper. Tell me when the numbers were collected and explain any refusal.” Your assistant can run:

aiwg doctor --resources

The report now shows when its machine readings were taken, when they were checked and how old they are. By default, snapshots older than five minutes cannot admit a worker launch. Missing or invalid readings also produce a reason instead of an answer you might mistake for a current one.

Know why a helper request was refused

When your assistant asks for a helper and the request is refused, it now gets the reason, such as a premium model needing your confirmation. Before, it could get a refusal with nothing to go on. Ask your assistant to explain the reason before trying again. The orchestration guide covers the workflow.

Smaller changes you will notice

Every project now tells your assistant never to add AI co-author lines or "Generated with" footers to your commits and pull requests, even when the assistant's own settings suggest it. The VS Code sidebar refreshes when `.aiwg/aiwg.config` changes. More component guides are collected into the documentation site, with example requests before commands.

For researchers, paper lookup accepts one strong match and joins a draft with its published version, keeping both links. A conflicting year still needs your review. Mission users can run `aiwg mc summary --ledger <path>` to see which work was accepted, finished but not yet accepted, failed, or can't be judged yet. Work that was only planned never counts as done.

If you are upgrading

• Run `aiwg use all` so AIWG has a fresh list of what it installed. That list is what makes a later removal clean.
• Deployments made before this release cannot be fully undone. Removal keeps and reports what it cannot verify.

See the CHANGELOG entry for the full technical detail.