Argus 0.6 turns exploratory testing scope into an explicit, evidence-backed coverage contract while reducing the default MCP context footprint.
What changed
- Add natural-language testing goals, user constraints, and an advisory time budget to browser sessions.
- Add evidence-required coverage updates for exercised and blocked goals; Argus never guesses semantic completion.
- Accumulate visited and discovered-page coverage across observations.
- Preserve completed, blocked, in-progress, and untested coverage in final summaries plus HTML and JSON reports.
- Move the full QA protocol into the one-time session start response, reducing the estimated default MCP description/schema footprint from about 58,589 to 33,522 characters.
- Update the public tool profiles to core 30, screen 14, and full 77 tools.
Verification
- 243 local tests passed.
- GitHub CI passed on Python 3.10 and 3.12 for the feature and release PRs.
- The deterministic capability bench remained 34/34 with 3/3 false-positive resistance.
- The 0.6.0 wheel and source distribution passed Twine checks.