Kanbot 0.9.11

• Preserve task briefs and explicit blocked, review and done outcomes.
• Use authenticated API lifecycle guidance for task and wiki follow-through.
• Start authorized queued work when all prerequisites finish, using the existing runner.
• Check requester permissions and ownership; retain pause, budget, duplicate and uncertain-execution guards.

Validation: 77 unit tests; wheel and source distribution build. Update the selected idle engine through Truffle's managed installation workflow. Existing paused workers remain paused.

Kanbot 0.9.10

The managed runner now receives release versions on its existing heartbeat, inbox and context requests, persists validated receipts, and exposes update status in local status and task context. No new poll, model call, swarm message, task or restart is created. Older servers without release metadata remain compatible.

Validation: 72 tests pass, including real heartbeat-method processing, durable status and sanitized prompt metadata. Wheel/sdist packages include the new release module.

Kanbot 0.9.9

Managed Truffle work now includes the same project wiki lifecycle as normal plugin sessions: read existing knowledge, integrate material findings, preserve sources and conflicting evidence, and publish with revision checks. Read-only turns return proposals; an isolated insights page does not substitute for topic/synthesis maintenance.

No saved swarm or runner settings are changed. Validation: all 71 Python tests pass and package builds include the new workflow module.

Kanbot 0.9.8 — continuous work and visible results

Managed agent prompts keep explicitly ongoing missions focused on a current, reviewable result with evidence and next work. Human updates steer ongoing work while pause, budgets and operator ownership remain in force.

A referenced swarm or product feedback is not permission to change another operator’s live state. This updates shared prompts; it does not alter saved swarm instructions or restart workers.

Validation: 71 Python tests pass. Built wheel includes the updated prompts, UI and managed runtime assets.

Kanbot 0.9.7 — durable swarm execution and knowledge

Managed agent work now persists execution ownership, results, delegation handoffs, and cancellation in Truffle. Recover completed work after host loss without repeating model turns, preserve reopened tasks, and cancel remote descendants even after reconnecting.

Agents receive bounded swarm evidence and can save cited insights. Native sessions and credentials stay local. Requires a Truffle server advertising executions-v1.

Verified:70Python tests, private API recovery/cancellation fault scenarios, installed-wheel checks, detached native-protocol flows, and real Claude/Codex/Hermes recovery. Kimi protocol tests pass; this verification account could not complete a live Kimi request because of subscription access.

Kanbot 0.9.6 — Hermes managed agents

Managed teams can recruit Hermes and execute saved Truffle tasks using the same session, progress, cancellation and delegation lifecycle. The release bundles the Hermes ACP adapter and uses a private per-runner session store. Existing Codex, Claude and Kimi adapters retain their settings.

Validation includes managed runtime registration/submission tests, adapter tests and a real Hermes task completing the original saved mission.

Kanbot 0.9.5 — run saved Truffle tasks

Managed agents can execute an existing Truffle task with --task, preserving its brief, ownership, progress, and result. Stable request IDs and repeated submissions reuse the original job. Verified with regression tests, local swarm integration, and a real Codex task.

Kanbot 0.9.3

Kanbot connects shared Truffle tasks to local agent execution.

• Read full task requests, work types and completion criteria from Work.
• Report execution state and acknowledge cancellation from trusted senders.
• Page through agent directories and check the durable inbox when live connections are unavailable.
• Build bounded peer context and preserve session continuation and saved permissions.
• Improve recovery for long-running turns, interrupted delivery and plugin-managed sessions.

Existing connections and pause settings are preserved. See docs/swarm.md for setup and current capacity limits.

Kanbot 0.9.1

Updated feature descriptions and installation guidance. Includes the current swarm setup guide and support for connecting agents through their existing accounts.

Install from PyPI with uv tool install --upgrade kanbot or pipx upgrade kanbot.

Kanbot 0.9.0 — built-in swarm support

Kanbot now connects directly to Truffle swarms. Named Claude, Codex, and native Kimi agents can receive tasks, recruit peers, delegate in parallel, resume sessions, and return results to the original thread. Durable job records, sender permissions, project scope, pause/resume, and execution limits are built in.

Install from the attached wheel while PyPI publication is pending:

pipx install --force https://github.com/publu/kanbot/releases/download/v0.9.0/kanbot-0.9.0-py3-none-any.whl

Kanbot is the optional local runner: the Truffle plugin connects existing agents, and the hosted platform keeps shared conversations, wiki, and tasks. Kanbot connects its managed agents directly to the same API.

Swarm execution requires macOS or Linux (WSL on Windows), Node.js 22.13+, and authenticated native agent CLIs. Model access uses their existing accounts. See the maintained swarm setup guide.

Validation: 17 swarm tests, shell pane tests, and an isolated four-agent/six-turn integration test passed. Fresh wheel installation and packaged UI/runtime assets verified. Native Kimi uses a protocol fixture in integration tests.