v0.1.1 — multi-language coverage, unified CLI, UX

Usability + multi-language accuracy over v0.1.0.

Highlights

• Coverage-based confidence for all supported languages — confidence=full now
  requires the changed lines to be exercised by the suite: Python (sys.settrace),
  JavaScript (V8 coverage), Go (go test -coverprofile), Rust (cargo-llvm-cov).
  Untested changes degrade → reject. Fail-open when coverage can't be measured.
• Precision parity via tree-sitter — comments, blank/brace lines, declaration
  signatures and imports don't require coverage (no false degrade on comment-only edits).
• Unified greenlock CLI — gate, check (gate git changes, no manual diff:
  --staged/--against), doctor, harden, init, mcp, --version, --apply.
• greenlock doctor — reports what Greenlock can verify in your repo.
• Actionable REJECT hints — tells you how to fix (run harden, --trust, …).
• GitHub Action posts the verdict — job summary + PR comment on reject.
• pre-commit framework support (.pre-commit-hooks.yaml).
• PyPI publishing workflow (Trusted Publishing) — once configured, pip install greenlock.

Notes / limits

• Go/Rust coverage parsers are unit-tested; end-to-end not yet validated against a live
  toolchain (fail-open keeps it safe).
• JS/Python coverage validated end-to-end. Not a sandbox unless --isolated.

Full details: CHANGELOG.md.

v0.1.0 — first public alpha

Greenlock is a deterministic verify-gate for AI-generated code changes: a patch is
applied only if the oracle is green — closed-world ✔, the project's own tests pass,
no regression — otherwise it is refused, untouched. Headline: WRONG-APPLY = 0.

Highlights

• gate (verify-only) + generate-and-gate; exit 0/1 for CI & pre-commit.
• execution isolation (--isolated / GREENLOCK_DOCKER): whole gate in a locked
  Docker container (network-off, read-only, non-root, limits), fail-closed.
• honest coverage-based confidence: changed lines must actually be exercised by the
  suite, else degrade → reject (no false MERGE). Python/pytest only in 0.1.0.
• testgen (characterization tests), danger-check (+ --trust advisory mode),
  MCP server, GitHub Action.

Known limitations (read before relying on it)

• Coverage-based confidence is Python/pytest only — node/go/rust treat a green suite
  as full confidence (multi-language coverage is the next milestone).
• Coverage uses in-process tracing → execution in test-spawned subprocesses is invisible
  (fail-open: never blocks a green patch on a measurement gap).
• Not a sandbox unless --isolated; the verifier executes your tests. See SECURITY.md.
• Alpha; install from source (not yet on PyPI).

Full notes: see CHANGELOG.md.