icaire-whats-next

Provide a concise snapshot of what should be done next from in-progress and open ICAIRE task pages exposed through the ICAIRE Cortex MCP task tools. Use when the user asks what's next, wants a short t

SKILL.md

GitHub

ICAIRE: What's Next

Use this skill when the user wants a short, decision-ready snapshot of active and ready ICAIRE task pages. ICAIRE tasks are page-backed records under tasks/*.md; do not read or reconstruct old ops/tasks.md task lists.

This skill is related to icaire-fanout-tasks, but it does not create handoff threads by default. It summarizes the next work first, then offers to fan out Codex threads if the user wants to start the work.

If the user answers questions from the input-needed section or provides additional task context after a snapshot, treat that as task-enrichment input. Do not start executing the task in the same thread unless the user explicitly asks you to work on it here. Working on tasks in the same thread where What's Next ran is an anti-pattern by default; enrich or clarify the task record, then wait for an explicit fanout or execution request.

Workflow

Use the ICAIRE Cortex MCP task endpoint as the source of truth:

  1. Call tasks/list twice by default, scoped to the authenticated member: first with status: "in_progress", assignee: "me", then with status: "open", assignee: "me".
  2. If tasks/list is not visible, use targeted Codex tool discovery for the ICAIRE tasks/list tool before falling back to runtime or code inspection.
  3. Resolve assignee scope before listing tasks:
    • For a generic request such as "what's next", and for "my tasks" or "assigned to me", pass assignee: "me".
    • For "for people/name" or "assigned to people/name", pass that assignee slug.
    • Omit assignee only when the user explicitly asks for all-team, everyone's, or unassigned-across-the-team work.
    • For a named priority such as p0, p1, p2, or p3, pass priority.
    • For a named status, pass that status; otherwise use both default calls.
  4. Use the returned task title, body, priority, assignees, source, and slug to identify the most useful next work. Use slugs internally for lookup and continuity, but do not normally show them in the snapshot output.
  5. Within the resolved assignee scope, prefer in_progress tasks first, then high-priority open tasks. Keep waiting tasks separate unless the user asks for them.
  6. Treat readiness and execution_mode as useful hints, then read the task body. If ## Open Questions contains substantive questions, put the task in the input-needed section even when frontmatter says readiness: "ready", unless the task is clearly an interactive guided session whose purpose is to answer those questions with the user.
  7. Classify the remaining tasks with readiness and execution_mode:
    • readiness: "ready" plus execution_mode: "agent" means the task can appear in the normal next-work numbered list and can be fanned out as an autonomous prompt.
    • execution_mode: "user" means the user must personally do the work, even if the task is otherwise ready. Keep these in a separate user-action section.
    • execution_mode: "interactive" means an agent can help only by walking the user through input, review, or decisions. These still belong in the main next-work numbered list and can be fanned out as guided step-by-step prompts.
    • readiness: "underspecified" means the task needs user input before it should be fanned out or treated as an executable handoff.

Respond in chat with the snapshot. Do not write the result to a file, do not return only a file link, and do not make the user open an artifact to see what is next.

Output Requirements

Default output is capped at 8 numbered items. Keep the snapshot short:

  • Show a What's Next section first.
  • This section should contain only tasks with readiness: "ready" and execution_mode: "agent" or execution_mode: "interactive" after the open questions override above.
  • Format ready tasks as a numbered list, not bullets.
  • Each numbered item should lead with a human-readable task title or action, not the task slug. Include priority only when it helps ranking or urgency.
  • Do not include task slugs in the What's Next numbered list unless the user explicitly asks for paths/slugs or two tasks would otherwise be ambiguous.
  • Do not format the numbered list as copyable prompt blocks.
  • Do not include boilerplate about reading files, preserving changes, verification, or commits.
  • Only when there are matching execution_mode: "user" tasks, append exactly: There are a few things I can't physically help with:
  • Under that line, show a numbered list of user-only tasks with the concrete real-world user action required.
  • If there are no matching user-only tasks, omit this heading entirely and do not mention that no such tasks were found.
  • Only when there are matching readiness: "underspecified" tasks, or tasks whose body contains substantive open questions, append exactly: I also need your input on a few tasks:
  • Under that line, show a numbered list of input-needed tasks. For each task, include indented bullet questions or the missing context required.
  • Make clear that answers to these questions are for clarifying or enriching the task, not for starting execution in this What's Next thread.
  • If there are no matching input-needed tasks, omit the input-needed heading entirely and do not mention that no such tasks were found.
  • Name input-needed tasks by human-readable title or action, not slug, unless the user explicitly asks for paths/slugs.
  • Prefer questions from the task page's ## Open Questions section. If that section is absent or incomplete, add a small number of inferred blocking questions on the spot.
  • Do not include input-needed or user-only tasks in the main What's Next numbered list.
  • End by asking whether the user wants Codex threads launched with handoff prompts for the ready agent-executable or interactive tasks.

If the user agrees to launch threads or receive prompts, immediately use the ICAIRE Fanout Tasks workflow on the same task scope and filters, preserving the resolved assignee filter. Do not ask the user to repeat the scope.

If no actionable ICAIRE task pages match the requested filters, say that directly and offer to run ICAIRE: Roadmap Tasks only if the user wants new tasks proposed from current brain evidence.

Quality Rules

  • Treat MCP task data as authoritative for task status and assignees.
  • Never issue an unscoped tasks/list call by default. Omit assignee only when the user explicitly requests an all-team view.
  • Do not use local TODO.md discovery for this skill.
  • Do not mutate tasks while producing the snapshot.
  • If the user follows up with answers or additional context, use that context to enrich the relevant task record when an update workflow is available; do not treat the answer as permission to start working on the task.
  • Do not create summaries for tasks with status: "done" or status: "archived" unless the user explicitly asks for those statuses.
  • Do not promote an underspecified task to the ready list just because it sounds important.
  • Keep each numbered item scoped to one task; put multi-task or ambiguous records in the input-needed list rather than guessing hidden subtasks.