links: site | repo | project page
AI Blueprint 1.2 is about making “another agent reviewed it” mean something specific.
The new path is /audit independent current. The builder has to verify a clean
checkpoint before handoff. The reviewer starts fresh, audits the complete
branch delta against an allowed base, records commands and evidence, and writes
a receipt tied to that checkpoint. Then /complete checks the receipt before
it will close the feature.
what shipped:
create-ai-blueprint@1.2.0on npm- independent-review policies: manual, when-sensitive, always
- review receipt metadata for target, merge base, spec hash, reviewer, commands, Check result, findings, and remaining risk
- staleness detection when code, config, specs, tests, or base changes after review
- completion blocking for missing, stale, malformed, changes-requested, or serious unresolved review evidence
- public docs across Audit, configuration, adapters, findings, completion, automation, status, and What’s New
The important constraint is honesty. Blueprint can record the requested reviewer model and the actual model the reviewer declares, but it is not pretending that a local workflow can cryptographically prove fresh-session identity. The gate is still useful because it makes review evidence concrete, repeatable, and attached to the exact work being merged.