Repository navigation
v0.2.1 — Trigger Efficiency
Highlights
- Shorter starter prompts and trigger-only descriptions across all 12 published skills.
- A 36-case routing corpus plus three synthetic/sanitized plugin-eval scenarios with fail-closed evidence checks.
- Structural release-gate coverage for trigger efficiency, workspace integrity, and benchmark evidence boundaries.
- Synchronized Codex, Claude, OpenCode, and generic skill mirrors.
Static comparison
- Trigger cost: 1,161 → 682 (-41.26%)
- Invoke cost: 15,365 → 14,886 (-479)
- Deferred cost: 161,228 → 161,248 (+20)
The measurements are estimated-static. Synthetic/generated benchmark outputs do not prove real Excel task success; observed usage and representative workbook behavior remain separate evidence.
Validation
- 102 Python tests passed from a cache-clean workspace.
- Structural release gate and GitHub Actions checks passed.
- Published skill IDs and Excel feature scope are unchanged.