Platform Tool for Scheduler: Allow Goose to Manage Its Own Schedule

[tlongwell-block]

This pull request introduces several updates to the Goose server and agent modules, focusing on enhancing scheduler functionality, improving error handling, and adding test utilities. The key changes include providing scheduler access to agents, refining error responses for scheduling operations, introducing a new tool for managing schedules, and adding test support utilities.

Scheduler Enhancements:

• crates/goose-server/src/commands/agent.rs: Added functionality to provide scheduler access to agents by cloning the scheduler instance and setting it in agent_ref.

Error Handling Improvements:

• crates/goose-server/src/routes/schedule.rs: Enhanced error handling in create_schedule by mapping specific scheduler errors (e.g., JobNotFound, CronParseError, RecipeLoadError, JobIdExists) to corresponding HTTP status codes.
• crates/goose-server/src/routes/schedule.rs: Updated API responses to include status codes for 400 (Bad Request) and 409 (Conflict) for invalid cron expressions and duplicate job IDs, respectively.

New Tool for Schedule Management:

• crates/goose/src/agents/platform_tools.rs: Introduced the manage_schedule_tool function and its associated constant PLATFORM_MANAGE_SCHEDULE_TOOL_NAME. This tool enables managing scheduled jobs with actions like create, delete, pause, and inspect. [1] [2]

Test Utilities:

• crates/goose/src/scheduler_test_support.rs: Added a test-only utility for managing a default provider (TEST_PROVIDER) during scheduler job executions in tests. Includes set_test_provider and get_test_provider functions for setup and retrieval.
• crates/goose/src/agents/mod.rs: Added test modules private_tests and test_support for improved test coverage and organization.

[michaelneale]

can you explain why agents need scheduler access, not quite following what this is about.

[tlongwell-block]

can you explain why agents need scheduler access, not quite following what this is about.

Enabling goose to run recipes in isolated agents on demand opens up some interesting use cases around agents effectively managing agents. And recipes are going to be a major part of the UX. So having goose manage its own library of them seems right.