fleet-sprint-planning
How to plan sprints with capacity awareness, velocity history, and realistic commitment — the art of saying no to protect what you said yes to
SKILL.md.claude/skills/fleet-sprint-planning/SKILL.md
GitHubSprint Planning
The Core Tension
The PO wants everything. The sprint has finite capacity. Your job is to create a realistic plan that DELIVERS, not a fantasy plan that impresses.
A sprint where 8 of 10 stories complete is better than a sprint where 6 of 15 complete. Predictability > ambition.
Step 1: Calculate Capacity
Agent Availability
Each agent has a heartbeat interval and a maximum parallel task capacity:
| Agent | Heartbeat | Capacity | Notes |
|---|---|---|---|
| software-engineer | 65m | 1 task | Implementation is focused work |
| architect | 60m | 2 tasks | Design can overlap (review + new) |
| qa-engineer | 70m | 2 tasks | Predefinition + validation overlap |
| devsecops-expert | 55m | 1 task | Security review is deep |
| devops | 75m | 1 task | Infrastructure is sequential |
| technical-writer | 80m | 2 tasks | Documentation can overlap |
| ux-designer | 85m | 1 task | UX spec is deep thinking |
Sprint Points Budget
- Sprint length in days x points-per-day-per-agent
- Conservative: 3 points/day for implementation agents
- Reduce by 20% for sprint overhead (standups, reviews, blockers)
- Reduce by contribution load (architect + QA + devsecops time on contribution tasks for OTHER agents' stories)
Contribution Tax
For every story the engineer works on, architect/QA/devsecops each spend time on contribution tasks. Budget this explicitly:
Story (5pts engineer) requires:
+ 2pts architect (design_input)
+ 1pt QA (test_definition)
+ 1pt devsecops (security_requirement)
= 9pts total fleet capacity consumed
A 5-point story costs 9 points of fleet capacity. Plan for this.
Step 2: Prioritize
Use the PO's priority ordering. Don't re-prioritize.
But within that ordering, sequence for dependency efficiency:
- Foundation stories first (models, migrations, shared utilities)
- Stories that unblock the most other stories
- Stories with the most contribution requirements (start early so contributions have time to complete)
- Independent stories last (can fill gaps)
Step 3: Commit Realistically
The Commitment Protocol
- List all candidate stories by PO priority
- Calculate total points
- Compare against sprint capacity (with contribution tax)
- Draw the line where capacity runs out
- Everything below the line is backlog, not sprint
Negotiation Signals
If the PO pushes for more:
- Show the math: "10 agents, 2-week sprint, 180 available points, stories total 220 points including contribution tax"
- Offer trades: "We can add story X if we move story Y to next sprint"
- Never commit to what can't be done — the PM's credibility is the fleet's credibility
Step 4: Create the Sprint Board
For each committed story:
- Set sprint assignment in Mission Control
- Verify contribution tasks exist
- Verify dependencies are ordered
- Set initial stage and readiness
Sprint Kickoff Checklist
- All stories have ALL fields set (type, stage, readiness, points, agent, verbatim, phase)
- Contribution tasks created for required roles
- Dependencies mapped and ordered
- Capacity math documented in board memory
- PO has approved the sprint commitment
Velocity Tracking
After each sprint:
- Points committed vs. points completed
- Stories committed vs. stories completed
- Blocker count and duration
- Contribution completion rate
Use velocity to improve NEXT sprint's commitment. Three sprints of data beats any estimation technique.
Group Calls
| Call | When |
|---|---|
pm_sprint_standup(sprint_id) |
Every heartbeat — velocity, blockers, gaps |
pm_contribution_check(task_id) |
Before advancing to work stage |
pm_gate_route(task_id, "sprint_boundary") |
At sprint boundaries |