0.2.0

Machine-readable output and an opt-in CI gate for richter:detect-changes, plus stricter config validation. Advisory-by-default is unchanged: no flags still means human-readable text and exit 0.

Added

• --json on richter:detect-changes and richter:impact. JSON mode emits a single parseable document on stdout — the full, uncapped report — for scripting and CI. Any failure is expressed as {"error": "…"} on stdout rather than a leaked stack trace, so the output is always valid JSON.
• Opt-in CI gating on richter:detect-changes. --fail-on=<low|medium|high> exits non-zero when the reported risk is at least the given level; --fail-on-unresolved exits non-zero when any changed file is UNRESOLVED, independent of the risk threshold. Both fail closed: a missing or invalid threshold (--fail-on, --fail-on=, --fail-on=bogus) is a usage error, and an un-assessable diff (a broken or invalid base ref) fails under a gate rather than passing as "no impact". With --json, the report carries a gate object recording the verdict.
• README "Gating in CI" section with a copy-paste GitHub Actions recipe. No Action ships with the package — detect-changes is a plain Artisan command.

Changed

• Config is validated on read. A mis-shaped richter.* value now throws instead of being silently dropped, and a base ref shaped like an option (leading -) is rejected before it can reach a git argument. A misconfiguration surfaces loudly rather than degrading into a falsely-empty report.
• MCP tool names pinned (impact, detect-changes) so agent integrations stay stable across releases.

Notes

Richter remains dev/CI tooling and advisory by default; a low or empty result is a signal, not a guarantee of no impact. Gating is strictly opt-in.

Full Changelog: v0.1.0...v0.2.0