v1.3.0 — Task queue for work distribution

Bridge grows from a message bus toward an agent-orchestration platform. A channel can now act as an exclusive work queue: an orchestrator enqueues tasks and a fleet of worker agents atomically claim them — no two workers ever process the same task.

Added — 5 new MCP tools (13 total)

• bridge_enqueue — add a task (structured payload or string content) with priority, max_attempts, delay_seconds, and an optional idempotency_key that dedupes a retried enqueue.
• bridge_claim — atomically claim the oldest eligible task; holds a lease for lease_seconds (a visibility timeout) and returns a lease_token. Set wait_seconds to long-poll for work.
• bridge_complete / bridge_fail — resolve a claimed task, fenced by the lease_token so a worker whose lease expired and was reclaimed can't clobber the result. fail requeues (with retry_delay_seconds backoff) until max_attempts is exhausted, then dead-letters.
• bridge_tasks — inspect a channel's queue (per-status counts + a task list).

At-least-once semantics: an unacknowledged lease is requeued (or dead-lettered past max_attempts) automatically, so a crashed worker's task is retried — make task handlers idempotent.

No breaking changes; every existing tool and transport is unchanged. Shipped after a multi-lens adversarial code review (3 low findings fixed).

Install: pip install claude-code-bridge · uvx claude-code-bridge --stdio
Full notes in CHANGELOG.md; usage in README and CLAUDE.md.