Skip to content

v0.56.0

Choose a tag to compare

@github-actions github-actions released this 28 Jun 06:34
· 1411 commits to main since this release

Added

  • Added first-class semantic selector ergonomics to synapse git-claim:
    --module, --symbol, --api, --source, --test, --generated, and
    --migration resolve locally into ordinary claim paths, while
    --semantic-evidence-json writes receipt-ready selector evidence.
  • Added dashboard-local bearer authentication for synapse dashboard: loopback
    dashboards remain unauthenticated by default, explicit --dashboard-token
    protects browser and JSON requests, and non-loopback dashboard binds receive
    a generated startup token when the operator does not provide one.
  • Added synapse hub --paranoid as a fail-closed local hub profile that
    requires token-protected access, a durable event log, metrics bearer-token auth
    when metrics are enabled, disables relaxed metrics/off-loopback switches, and
    prints the missing hardening hooks it does not implement.
  • Added the official read-only Go client for ops and CI tools to fetch dashboard
    JSON snapshots without implementing WebSocket mutation flows.
  • Added a committed five-agent coding fleet benchmark that measures local claim
    conflict rate, claim latency, release cleanup, and replay recovery evidence.
  • Added branch-conflict candidates to the dashboard fleet view, derived from
    active git-scoped claim metadata without running git from the dashboard.
  • Added a read-only dashboard task-dependency graph derived from blackboard task
    edges and exposed through both HTML and /snapshot.json.
  • Added fleet visibility to synapse dashboard and /snapshot.json, including
    live agents, -rx waiters, missing waiters, active and stale claims, ready
    and blocked tasks, release receipt notes, and optional persisted A2A task
    counts via --a2a-state-file.
  • Added the public agent trust graph design for evidence-linked routing review
    over reliability signals, release receipts, capability observations, handoff
    outcomes, conflict history, provenance references, decay windows, policy
    inputs, and explicit non-scoring boundaries before any graph runtime ships.
  • Added the public differential-privacy blackboard design for redacted and noisy
    multi-organisation board projections, privacy budgets, cohort thresholds,
    privacy-ledger audit evidence, and explicit raw-log/encryption/authorization
    boundaries before any privacy runtime ships.
  • Added the public signed capability cards design for tamper-evident capability
    advertisements, manifest/card digests, verification results, replay controls,
    credential rotation, revocation, trust bundles, and advisory-discovery
    boundaries before any signing runtime ships.