Squadra 3+ 3.9.0

Squadra 3+ 3.9.0: system skills that activate on their own, faster proofs, and releases that leave on time.

What changes

• System skills. Three built-in skills activate on their own and the agent follows their guides without asking:
  • errata: after a failed proof, a read that asked for changes or a closed defect fix, the agent records a lesson with its trigger; matching lessons come back at brief and start, and a lesson that did not prevent a repeat can be marked ineffective (squadra lesson).
  • rasoio: a lean-code lens for code reads (review --lens rasoio --net-lines N); it adds observations and never changes a verdict.
  • previsione: squadra forecast --task ID compares a task with similar finished ones and suggests a route; it is computed on demand and records nothing.
    errata and rasoio record each activation in the registry.
• System skills page. A dashboard page lists the three skills with what they do, their recorded activity and the result of the security scan on their files, read from a bundled audit record. If a file changed after the scan, the page says the skill is not audited for this version.
• Reviewer pool and a single gate for small tasks. task --reviewer can be repeated (up to four); the first non-author to acknowledge takes the read. A small light task registered --read-at-gate closes without its own read and shows as closed, not yet read, until a release gate (--release-gate) covers it or the coordinator explicitly defers it.
• Quick checks before proof and delivery. A repository can declare fast checks in .squadra-precheck.json; prove and deliver run them on the exact files first, and delivery refuses if the files changed after the checks.
• Parallel whole-suite proof. tools/parallel_tests.py runs the suite in separate processes and fails if any collected test did not run. On our test machine: 953 tests in about 105 seconds.
• Release rules. Rule 9: a release has a cut-off two hours after its scope is fixed; final reads look only at the integration seams when every lane was already read; the parallel runner is the release proof; a read nobody picks up moves to a free reader. Rule 10: a fix round carries only the fix and is read only by whoever raised the problem.
• A lighter README with the same sections, and a «What's new» section after the video.
• Plugin versions: memoranda 1.0.3, Playwright 1.0.4, Sentry 0.1.1 (their bundled registry reader follows this release).

Compatibility

Existing tasks keep working; single-reviewer tasks behave as before. A squad already at work keeps its version until every writer is upgraded: older CLIs ignore the new lesson, pool and gate metadata, so mixed-version enforcement is not promised.

Updating from 3.8.0

Claude Code: /plugin marketplace update squadra, then /plugin update squadra@squadra, then restart Claude Code.
Codex: codex plugin marketplace upgrade, then codex plugin add squadra@squadra again.
Standalone installer: git pull, then run scripts/install.py again with a new backup folder.
If you installed memoranda, Playwright or Sentry from the Store, stop their producer and consumer, then run squadra plugin install <name> again; settings, keys and data are kept.