agenthub-task-manager

User-invoked AgentHub workflow for planning a multi-part request and delegating its independent subtasks to parallel AgentHub agent sessions. Use only when the user explicitly invokes this skill to plan, create, list, or clean up AgentHub task sessions.

Jump to install

Repository: jamesrochabrun/AgentHub
Last source activity: June 22, 2026 at 06:29
Detected SKILL.md language: English
Stars: 483
Forks: 35

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

A direct command skips the review prompt. Inspect the source before running it.

npx skills add https://github.com/jamesrochabrun/AgentHub --skill agenthub-task-manager

The command stays on one line. Scroll horizontally to inspect it before copying.

Prefer a local copy? Download the files currently available to SkillsMP.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

File Explorer

2 files

SKILL.md

Source instructions · Read-only preview

Back to search Back to installation

agenthub-studio

Render UI variants or any visual — mockups, reports, diagrams, dashboards — into AgentHub's Studio panel beside this session, on a scratch surface that never touches project files. Use when the user asks to see, compare, mock up, or explore options, versions, or variants of a component or screen, asks "what would it look like", or wants a rendered document. Also invoked explicitly as /agenthub-studio.

2026-08-19

View all from this repository

AgentHub Task Manager

Use this workflow only because the user explicitly invoked this skill to plan and delegate work to parallel AgentHub agent sessions.

Plan And Create Task Sessions

Read the user's request and infer the independent subtasks yourself. Preserve explicit numbered or bulleted work streams as separate subtasks; do not split ordinary prose just because of punctuation, commas, semicolons, or conjunctions.

Call agent_hub_planning, passing the original prompt and the subtasks you inferred as subtasks. AgentHub does not split for you — with no subtasks it plans the whole prompt as a single task.

Review the returned assignments and the harnessCapabilities block — each harness's real installed skills and MCP servers. AgentHub launches a harness (Claude Code or Codex), not a model, so present only the harness, never a model name. When multiple harnesses are installed, each subtask comes with a suggested harness plus the matchedCapabilities that drove it. Confirm or override every suggestion by matching the subtask to the harness whose actual tools fit — e.g. a profiling subtask → the harness that has an xctrace/profiler tool; a SwiftUI subtask → the harness with a swiftui skill. Decide from the listed capabilities, never from a harness's general reputation, and cite the matching skill/tool in your rationale.

Present the final assignments (task, chosen harness + why, branch, launch prompt) and wait for explicit user approval.

Call agenthub_create_worktree_sessions only for approved assignments. Pass one task per assignment with explicit provider, branch, and prompt. Include startPath when the task belongs to a subdirectory; AgentHub uses nearby project markers to infer a sparse worktree and then launch the agent from that same relative path. Use sparseProfile only for extra required paths and fullCheckout: true only when the task genuinely needs the whole repository materialized.

For a single-task request, still call agent_hub_planning with one subtask so the agent and branch are explicit before approval.

List Or Delete Task Sessions

Use agenthub_list_worktrees before destructive cleanup when the target or session impact is unclear. Use agenthub_delete_worktree only after the user identifies the session to delete.

Boundaries

Do not use these AgentHub task tools for generic planning, fan-out, background work, or subagent requests unless the user explicitly invoked this skill. Preserve the current harness's native subagent and background-task capabilities.