You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Local packages/agent CLI with private user-level model configuration and browser-consent PKCE login through @nocoo/base-cli
One Pi Durable Harness, one Astra planner, one shared Jev decision client, and persistent Sol specialists for repository/global Issues, PRs, CI and CD
Authenticated read/write report, job and heartbeat publication with stable IDs, revision conflicts, restart recovery and bounded remote history
Basalt analysis workbench with four domains, repository/portfolio scope, evidence versions and age, Jev distributions, runner presence and queued requests
Browser API access to Agent resources under existing Access and same-origin checks
Changed
Read GitHub facts from the existing Giraffe collector instead of duplicating collection in the Agent
Separate current source time, report-generation time and model judgments; stale or incomplete evidence remains unknown
Bound each model run to eight responses and retain current workflow outcomes ahead of historical failure samples
Include the Node Agent workspace in root typecheck, lint, build and coverage gates
Removed
Cloud AI model settings, provider execution, assessment endpoints and old report consumers; no compatibility fallback remains
Unused cloud AI dependencies; legacy D1 tables remain intact and frozen collection skips only retired AI stages
Verification Boundary
Automated acceptance uses isolated SQLite/Worker/GitHub fixtures; the real model pipeline was exercised visibly using synthetic evidence
Production browser consent and authenticated Agent publication require the owner to complete giraffe login; model success alone does not establish that production acceptance