[BUILD3R-Dev]

What changes

Projects can now preserve confirmed business rules and behavioral scenarios across agent sessions. Task-scoped handoffs deliver pinned context; corrections invalidate obsolete verification; review and atomic acceptance bind the exact source, knowledge, authority, and evidence. New web projects use this workflow, while existing legacy agency projects retain their behavior.

This adds the reference Outcomes, Knowledge, and Review experience; explicit handoff renewal; versioned history and portable bundles; real local Git/runner/previews; and release observations and historical source recovery under current requirements. The public implementation has no private agency dependency.

Validation

• Landing verification: typecheck 21 tasks passed; build 13 tasks passed; a fresh uncached full unit run passed all 21 tasks.
• Implementation completion record: 392 security passes (1 skip), 20 fault passes, and five real-local browser journeys plus canonical recovery verification. Historical aggregate unit totals included 44 duplicate emitted ledger tests; landing fixes exclude those generated tests, and the corrected ledger suite passes all 176 source tests.
• Actual fresh-agent continuity and isolated Cloudflare release/real freshness expiry/recovery are recorded in docs/agent-native/execution-status.md and sanitized evidence. Live verification exercised three deployments and 27 HTTPS behavioral cases.
• PR CI runs build/typecheck, unit, fault/security, fixture browser tests, the real-local G1 journey, and configured live-provider tests. Skipped tests do not establish live behavior.
• Focused landing review covered project acceptance, task scope, evidence binding, legacy bypass guards, and release binding. Changed text files were scanned for common credential patterns; the only match was a synthetic negative-test token.

PR CI exposed five-second default timeouts in the two-Worker integration test and the multi-megabyte bundle checks, plus duplicate execution of ledger tests emitted into dist. Landing fixes give these heavy tests explicit 30-second timeouts, exclude source tests from ledger builds, and exclude generated output from test discovery. All payload sizes and behavioral/isolation assertions remain intact. The seven-test rehearsal suite and corrected 176-test ledger suite pass locally. The real-local launcher now uses standard grep, waits for both API and preview readiness, allocates separate ephemeral inspector ports, and stops if either Worker exits. Its full five-browser-journey and canonical recovery rehearsal passes locally. CI must pass on the final head before merge.

Scope and limits

G1 is a bounded foundation: verification supports the documented synthetic Node and vars-only Worker runtimes, not arbitrary applications. Source and knowledge changes are accepted in two steps; source membership/shared staging are restricted; renewal starts a new attempt; bundles preserve history but do not automatically restore Git repositories. Evidence retention is conservative. No production deployment or package publication is part of this PR.

A real domain-expert usability pilot remains unperformed. The next goal is a real project pilot and then the private agency consumer, rather than expanding autonomous orchestration now.