Ops Prioritize

Given a project's active plans, backlog, and goals, select 2-3 tasks for immediate autonomous execution. The output is a ranked shortlist with execution order — this skill does NOT execute them.

Context Required

You need the full project overview loaded (MEMORY.md, state, backlog, active plans) plus any newly identified work items from

find-tasks

.

Selection Criteria

Score each candidate task on these dimensions:

Impact (weight: high)

How much does completing this task move the project toward its goals?

• High: directly advances a primary goal, unblocks other work, or fills a critical knowledge gap
• Medium: advances a secondary goal, improves project quality, or creates useful reference material
• Low: nice-to-have, minor cleanup, or distant from current priorities

Feasibility (weight: high)

Can an autonomous agent complete this task without human input or real-world actions?

• High: pure research, writing, analysis, or knowledge management — all inputs are available
• Medium: mostly feasible, but may need assumptions about preferences or priorities
• Low: requires external information the agent can't access, real-world actions, or significant subjective judgment

Tasks with low feasibility should be skipped — they'll end up in the follow-up file anyway.

Independence (weight: critical)

Can this task be completed without depending on or conflicting with other selected tasks?

• Tasks that modify the same files are NOT independent
• Tasks where one's output is the other's input are NOT independent
• Tasks touching different backlog areas, state files, or reference topics are usually independent

Independence is a hard constraint, not a gradient. If two high-impact tasks conflict, pick the higher-impact one and find an independent alternative for the second slot.

Effort (weight: moderate)

Prefer a mix of effort levels. A good batch might be:

• 1 medium task (the main contribution of this cycle)
• 1-2 light tasks (quick wins that keep momentum)

Avoid selecting only heavy tasks — the cycle should produce visible progress. Avoid selecting only light tasks — the cycle should produce meaningful progress.

Process

Step 1: Inventory

List all candidate tasks from:

1. Active plans in

   plans/

   — these are already approved work
2. Backlog items — particularly those marked as high priority or quick wins
3. Newly identified items from the latest

   find-tasks

   run

For each, note: description, source file, estimated effort, and which files it would touch.

Step 2: Score

Rate each candidate on impact and feasibility. Discard anything with low feasibility.

Step 3: Select

Pick 2-3 tasks that:

1. Have the highest combined impact × feasibility score
2. Are mutually independent (hard constraint)
3. Include a mix of effort levels (soft preference)
4. Together represent a productive cycle of work

If active plans exist, prioritize them — they represent work the user already approved. Backlog items and newly identified work fill remaining slots.

Step 4: Prepare

For each selected task, prepare an execution brief:

## Selected Tasks

### Task 1: [Title]
- **Source**: [active plan file or backlog file + item]
- **What**: [1-2 sentence description of what the sub-agent should do]
- **Scope**: [which files will be created/modified]
- **Effort**: light / medium / heavy
- **Plan exists**: yes/no — if no, the sub-agent should create one first

### Task 2: [Title]
...

If a selected task doesn't have an existing plan file, note that the sub-agent should create one as its first step (following

plan

format).

Step 5: Order

Determine execution order. Preferences:

1. Tasks with existing plans go first (less setup overhead)
2. Light tasks before heavy ones (quick wins build momentum and context)
3. If a task might inform another (even if independent), put it first

Rules

• Always select at least 2 tasks unless the backlog is nearly empty
• Never select more than 3 — better to complete a focused batch and reassess
• Skip tasks that require real-world actions — no emails, form submissions, API calls to external services, purchases, or social media posts
• Prefer breadth over depth — touching multiple project areas in one cycle is better than deep-diving one area, unless one area is clearly the bottleneck
• Respect the user's priorities — if MEMORY.md or state files indicate urgency on a topic, weight those tasks higher