v0.3.1

Added

• Public local-browser reuse path — agentyc mcp --reuse-local-browser and AGENTYC_REUSE_LOCAL_BROWSER=1 let agents in other projects attach to the latest locally launched Agentyc browser without manually copying a CDP URL, and agentyc browser --detach now registers that browser for later reuse.
• Canonical browser-automation skill and guide validation tooling — .agents/skills/agentyc-browser-automation/ now carries the shared-browser/browser-tool playbook with an eval rubric and cases, and the repo now includes skill validation/eval helpers plus a distributed-guide drift test.

Changed

• Packaged skills guide refreshed — agentyc init now emits guidance that explicitly covers shared-browser reuse, same-browser-process/profile versus same-tab safety, tool-selection rules, cookies and storage workflows, richer interaction recipes, JS evaluation, and a fuller quick-reference tool list across the 66-tool browser MCP surface.
• Cold frame enumeration path is leaner — browser_list_frames now skips stale unattached page targets before attempting a CDP session attach, which materially removes the old cold frame-path stall and improves the confirmed two-run headless stdio median to 24140.0 ms total with 43.3 ms average / 161.6 ms p95 tool latency.
• Release benchmark baseline refreshed — the current confirmed two-run headless medians are runtime import_ms=194.4, session_init_ms=1027.4, avg_auto_payload_reduction_pct=10.2, and collaboration_latency_ms=1283.1, with recall/action metrics remaining 1.0.

Tests

• New shared-browser and frame-path coverage — tests/ci/test_mcp_cli.py, tests/ci/test_mcp_runtime_optimizations.py, and tests/ci/browser/test_frame_target_filtering.py.
• New distributed-guide and skill-quality coverage — tests/ci/infrastructure/test_distributed_skill_guide.py, tests/ci/infrastructure/test_skill_quality.py, scripts/validate_skills.py, and scripts/run_skill_evals.py.