Changelog

0.2.1 - 2026-11-08

• Add an explicit Claude manifest for claude.ai marketplace synchronization.

0.2.0 - 2026-09-04

• Bind the product thesis to a ranked evidence hierarchy, contradictions, assumptions, confidence, non-goals, and an explicit falsification step.
• Require every major product change to serve a user job and cite supporting evidence, expected observable behavior, and preserved invariants.
• Replace self-certified repair passes with a workflow ledger, same-journey before-and-after replay, thesis revision, and a cold-start gate.
• Add a safe-execution preflight for dirty worktrees, rollback boundaries, runtime endpoints, data classification, test accounts, and outbound effects.
• Restrict mutating browser flows to fixtures, disposable accounts, isolated data, or proven sandbox services. Missing proof now fails closed to read-only.
• Bound one primary product surface and report lower-priority residual work instead of expanding indefinitely.
• Add durable behavioral evaluations for triggers, non-triggers, stale evidence, generic redesign, real services, missing tools or credentials, cosmetic-only work, premature completion, authority expansion, and scope control.
• Add replayable fixtures for the highest-risk evidence, service, worktree, ambiguity, and scope-boundary evaluations.
• Guard both portable and OpenAI manifests against silently gaining hooks, MCP servers, or apps.
• Record two-family matrix sweeps and live disposable, read-only service, and large-application scope evaluations.
• Register Product Wrangler in the Codex and Cursor marketplace catalogs.
• Add explicit COMPLETE, PARTIAL, and BLOCKED result states.

0.1.0 - 2026-09-04

• Initial standalone Product Wrangler skill draft.