S-Agent Core v0.2.0
GitHub Action and benchmark infrastructure release.
Highlights
- Added a root composite GitHub Action that runs the existing S-Agent CLI.
- Added Action inputs for project path, rules path, output format, and blocking behavior.
- Added GitHub Action usage docs and a copy-paste workflow example.
- Added
pnpm benchmarkfor canonical evaluation fixtures. - Added benchmark JSON summary and Markdown leaderboard output.
- Added benchmark metrics:
- precision
- recall
- false positive rate
- PROVEN rate
- runtime
- clean blocking rate
- Added shared evaluation fixture manifest.
- Renamed the future commercial platform references to AxiomGuard.
Validation
- Build: passed
- Tests: 25 files / 40 tests passed
- Lint: passed
- Clean demo: passed, no findings
- Broken demo: passed as expected by producing a PROVEN critical blocking finding
- Benchmark: passed
Current benchmark
| Fixtures | Precision | Recall | False positive rate | PROVEN rate | Runtime | Clean blocking rate |
|---|---|---|---|---|---|---|
| 6 | 1.000 | 1.000 | 0.000 | 0.500 | 21 ms | 0.000 |
Note: this is a small canonical fixture benchmark intended to track regressions in the OSS core, not a production-quality claim.
Known limitations
- Pull request comments are not implemented yet.
- No hosted service or dashboard.
- TypeScript only.
- Import resolution is intentionally simple.
- Side-effect checks are heuristic and name-based.
- No analyzer behavior changes in this release.