[rybosome]

Executive summary

Sets the mic-evals version to 0.3.0 so the maintainer can cut the v0.3.0 release. The main change since 0.2.0 is declared result sinks (@mic.eval(sinks=...) and mic run --no-sinks).

Why

Downstream users (intent2auth's results sink, per TEN-1188) need a published version that includes declared sinks. Per docs/releasing.md, releasing starts by merging a reviewed version bump to main.

Design and implementation

• pyproject.toml: version = "0.3.0".
• uv.lock: only the mic-evals version line changes. My local uv (0.13.1) also wanted to raise the lockfile format revision from 3 to 5. I kept revision 3 so the diff matches the 0.2.0 bump and older uv installs can still read the lock; uv lock --check passes.
• skills/mic/SKILL.md: the version note now reads "Mic 0.3's CLI", because the CLI gained --no-sinks and sink names in mic list.
• The 0.2.0 strings in tests/packaging/test_release.py are self-contained fixture versions, not the project version, so they are unchanged.

Verification and confidence

The risk is a release-identity mismatch: the version in pyproject.toml, uv.lock and the built metadata not matching the tag.

• uv build produced mic_evals-0.3.0-py3-none-any.whl and mic_evals-0.3.0.tar.gz.
• scripts/verify_packaging.py: PASSED.
• scripts/verify_release.py artifacts: verified, run on a directory holding exactly the wheel and sdist. It fails on raw dist/ because uv build writes a .gitignore there; this is expected.
• scripts/verify.py: tests, coverage gates, typing, lint and format all PASS.
• uv sync --frozen --all-extras succeeds with the edited lock.

Compatibility and risks

None beyond the version change itself. The code, formats and dependencies are unchanged from main. For the release notes: 0.3.0 adds declared sinks. That is additive, but it is a new implicit effect: running an eval that declares a remote sink now writes to it with no flag, and --no-sinks turns this off. The mic list --json output gains a sinks key on eval rows.