agent-desk

Explicitly specified only: use this skill only when the user writes `$claim-agentdesk-task` or `claim-agentdesk-task` by name. Never infer or proactively trigger it from task content, workflow fit, or requests to claim AgentDesk work alone. When explicitly requested by name, atomically claim one checklist item, implement only that item, and complete it through AgentDesk.

Explicitly specified only: use this skill only when the user writes `$generate-agentdesk-task` or `generate-agentdesk-task` by name. Never infer or proactively trigger it from task content, workflow fit, or task-generation requests alone. When explicitly requested by name, create or prepare an AgentDesk task for a project before subagent execution.

Use when the user explicitly writes `$codexapp-direct-subagents` or `codexapp-direct-subagents`, or explicitly asks to dispatch Codex App subagents directly without creating or writing an AgentDesk task. This skill bypasses AgentDesk task/session generation and uses Codex App subagents directly, defaulting to model `o4-mini` and reasoning `low`.

Explicitly specified only: use this skill only when the user writes `$run-agentdesk-subagents` or `run-agentdesk-subagents` by name. Never infer or proactively trigger it from task content, workflow fit, or subagent-related requests alone. When explicitly requested by name, run or coordinate an existing AgentDesk task with parallel Codex CLI or Codex App subagents.

Explicitly specified only: use this skill only when the user writes `$review-agentdesk-task` or `review-agentdesk-task` by name. Never infer this skill from task content, generic review, planning, task-generation, or implementation requests alone. When explicitly requested by name, perform a read-only pre-implementation review of an AgentDesk task or task.md to find ambiguity, missing acceptance details, scope mismatches, and likely ways future agents could implement something different from the user's intent.