v0.5.0

botler-agent v0.5.0

Highlights

• Scheduler can run an in-project script directly (Plan A). A schedule entry may now carry a script (path relative to DATA_ROOT) instead of a message, executing the script directly on fire — bypassing the LLM — while reusing the run tool's security boundary (path allowlist, fixed interpreter, no shell, 60s timeout). The script fires on the dispatcher's sequential queue (never races chat writes) and its auto-commit is scoped to the script's own subproject. Script tasks retry only on error, never on validation-failed.

Follow-up fixes

• The schedule tool now accepts retryMax / retryBackoffMs via chat, with enforced upper bounds.
• Retries are non-blocking (setTimeout instead of await sleep) so the single-threaded scheduler loop is never stalled.
• Pending retries are cancelled if the entry's action / trigger / project changes during the backoff window.
• Updating one retry field preserves the other; switching the action type clears the inherited retry (avoids re-running a non-idempotent script with a stale retry).
• Precondition failures now return cwd so history logs are attributed to the correct project.

Other

• WebUI schedule editor: Action radio (Message / Run script), script fields, action-aware save guard, and script label in lists.
• System prompt documents the script action plus idempotency / retry guidance.
• npm test 225/225 pass; npm run typecheck clean.

Full Changelog: v0.4.1...v0.5.0