v1.1.0

• Fan out subagents for the socratic-review skill's silent assessment: large or unfamiliar targets (multi-file PRs, SHAs, inherited code) are now explored by parallel subagents — one per problem space — whose findings are merged into the private ranked list, while small targets are still read inline.
• Strengthen the test-driven-development skill's outside-in workflow: run the affected test after every change and let the failure dictate the next move, and write a new failing test at each layer you drop into (e.g. a request spec before building a controller) rather than treating lower layers as covered by the feature spec.