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.mdskills/icaire-whats-next/SKILL.md
GitHubICAIRE: 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:
- Call
tasks/listtwice by default, scoped to the authenticated member: first withstatus: "in_progress", assignee: "me", then withstatus: "open", assignee: "me". - If
tasks/listis not visible, use targeted Codex tool discovery for the ICAIREtasks/listtool before falling back to runtime or code inspection. - 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
assigneeonly 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, orp3, passpriority. - For a named status, pass that
status; otherwise use both default calls.
- For a generic request such as "what's next", and for "my tasks" or
"assigned to me", pass
- 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.
- Within the resolved assignee scope, prefer
in_progresstasks first, then high-priorityopentasks. Keepwaitingtasks separate unless the user asks for them. - Treat
readinessandexecution_modeas useful hints, then read the task body. If## Open Questionscontains substantive questions, put the task in the input-needed section even when frontmatter saysreadiness: "ready", unless the task is clearly an interactive guided session whose purpose is to answer those questions with the user. - Classify the remaining tasks with
readinessandexecution_mode:readiness: "ready"plusexecution_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 Nextsection first. - This section should contain only tasks with
readiness: "ready"andexecution_mode: "agent"orexecution_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 Nextnumbered 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 Nextthread. - 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 Questionssection. 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 Nextnumbered 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/listcall by default. Omitassigneeonly when the user explicitly requests an all-team view. - Do not use local
TODO.mddiscovery 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"orstatus: "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.