Skip to content

v0.8.0

Latest

Choose a tag to compare

@tt-a1i tt-a1i released this 17 Mar 14:32
· 5 commits to main since this release
2c34285

What's New

Chain Mode (--chain)

Run providers sequentially — each provider's output is fed as context to the next, enabling "challenge and supplement" workflows where later agents critically review earlier agents' findings.

Per-Provider Perspectives (--perspectives-json)

Assign different review focus areas to each provider. For example, Claude focuses on security while Codex focuses on performance. Configurable via CLI flag or .mcorc.json.

Consensus Badges

Findings now show cross-agent agreement indicators:

  • Parallel mode: [2/3 agree] / [1 agent only]
  • Chain mode: [confirmed by 2/3] / [unconfirmed]

Session Bug Fixes

  • Critical: Fixed --no-wait data loss — daemon no longer writes to closed socket; results stored for later retrieval via new session result command
  • Broader exception handling: adapter.cancel() catches all exceptions, preventing worker thread death
  • Provider validation: resume_session() validates provider matches

Session Retry with Error Classification

Session dispatch now uses the orchestrator's RetryPolicy with exponential backoff. Errors are classified (timeout, rate limit, network → retryable; auth, invalid input → non-retryable). Partial output preserved on timeout.

New CLI Commands

  • mco session result <name> <request_id> — retrieve results of async (--no-wait) requests

Strict --perspectives-json Validation

Invalid JSON, non-object root, or non-string values now raise clear errors instead of being silently ignored.


558 tests passing.