Skip to content

v0.2.0 - ClickUp Agent feedback pass

Latest

Choose a tag to compare

@zenzenzen zenzenzen released this 04 Jun 19:30
· 30 commits to main since this release
62668e8

Latest Rollout: ClickUp Agent Feedback Pass

This release turns the successful end-to-end ClickUp workflow trial into a clearer, safer, less chatty CLI and MCP surface.

Features

  • Added explicit command family metadata so every run result identifies whether it came from a curated wrapper or a generated OpenAPI operation.
  • Made exact OpenAPI operation IDs win: UpdateTask now runs the raw generated operation, while update-task runs the curated wrapper.
  • Added compact task fetch workflows:
    • clickup-agent run get-task --summary
    • clickup-agent run get-task --fields ...
  • Added status discovery and wrapper validation:
    • clickup-agent run task-statuses --task-id ...
    • clickup-agent run task-statuses --list-id ...
    • update-task --status now uses wrapper status validation before live mutation.
  • Added bulk checklist workflows:
    • create-checklist --items ... --resolved true
    • sync-checklist --task-id ... --name ... --items ... --resolve-all
  • Added normalized checklist response summaries so callers can find checklist.id, checklist_item.id, created_items, and matched_items without digging through raw ClickUp response shapes.
  • Expanded MCP tools for get-task, task-statuses, sync-checklist, bulk checklist creation, and status-aware task updates.

Fixes and Quality Improvements

  • Fixed the confusing update-task vs UpdateTask behavior by separating exact generated IDs from curated wrapper names.
  • Improved wrapper guidance with notes such as: For full API fields, use generated operation UpdateTask.
  • Improved dry-run output with source/family metadata so agents can verify whether they are previewing a wrapper or raw generated operation.
  • Improved checklist-item 404 errors with corrective guidance when a checklist ID may have been passed where a checklist item ID was required.
  • Reduced task fetch verbosity through summary and projection modes.
  • Documented that markdown_content is preferred for rich descriptions and that ClickUp may normalize stored/rendered markdown text.
  • Added repo-local CONTEXT.md and UBIQUITOUS_LANGUAGE.md so future agents share the same project intent and terminology.
  • Added concise intent comments/docstrings across the codebase without changing behavior.

Validation

  • uv run pytest -q -> 72 passed
  • git diff --check -> clean before merge
  • Full reinstall completed with uv tool install . --python 3.12 --reinstall
  • Installed CLI reports clickup-agent 0.2.0
  • clickup-agent doctor --live-auth confirmed user and workspace authorization without exposing secrets
  • Installed CLI smoke checks confirmed the new wrapper surface

PR

Merged in #11: #11