bhaveshpabnani/humungousaur/focus-and-priority-review

Focus And Priority Review

Purpose

Help the assistant and user choose the next best action from current evidence. This skill maps priority-review concepts to Humungousaur's cognitive priority tools.

When To Use

Use when the user asks "what should I do next", "continue", "prioritize this", "what is important", or when autonomous work needs a bounded next action.

Inputs And Evidence

• Active goals, tasks, and focus from cognitive_state.
• Commitments, wakeups, and recent briefings.
• Environment constraints and current system status.
• User urgency, deadlines, and interruption tolerance.

Tool Map

• cognitive_priority_review
• cognitive_priority_status
• cognitive_state
• cognitive_environment_status
• cognitive_commitment_status
• automation_daemon_tick

Workflow

1. Gather active work and explicit user constraints.
2. Identify hard deadlines, blockers, dependencies, and high-risk unresolved items.
3. Ask the model to rank next actions from evidence instead of task title similarity.
4. Choose one primary focus and at most a few secondary items.
5. If autonomous mode is active, queue only one interruptible next action unless explicitly allowed.
6. Communicate the reasoning and what will be deferred.

Safety And Boundaries

• Do not bury user-explicit priorities under inferred priorities.
• Do not trigger high-risk work without approval.
• Do not keep cycling priority reviews when the next action is already clear.

Safety And Approval

• Priority review can recommend work, but it cannot authorize high-risk tools, external-visible actions, or durable state changes by itself.
• Preserve explicit user priorities over inferred urgency unless current evidence shows a blocker or safety issue.
• Do not queue autonomous work unless the user allowed initiative or the active goal requires bounded continuation.
• Avoid repeated priority loops that consume time without creating a concrete next action.

Native Implementation Boundaries

• Use Humungousaur priority review/status, cognitive state, environment, commitment, and automation daemon tools.
• The model ranks actions from evidence; deterministic code must not rank by recency, keyword, or static urgency lists.
• Queue/status changes must be visible in native records before claiming the assistant will act later.

Verification

• The recommendation should map to active records or user input.
• Deferrals should name why they are deferred.
• If a next action is queued, verify it appears in the relevant queue/status.

Failure Modes

• Producing many "top" priorities.
• Ignoring blockers and choosing an impossible action.
• Ranking from recency alone.

References

• Shortlist item: focus-and-priority-review.
• Humungousaur tools: priority review and cognitive state.