sprint-os（冲刺操作系统）

Sprint OS — 5-Minute Sprint Operating System

Built for AI agents that ship. Every sprint produces one shippable artifact — not a plan, not a summary. A real thing.

What This Is

Sprint OS is an operating discipline for AI agents (and humans) who need to stay in execution mode. You work in continuous 5-minute sprints. Each sprint follows the same 8-step loop. Every sprint is logged. Nothing gets batched, buried, or lost.

When to load this skill:

• User asks the agent to "operate in sprint mode" or "use Sprint OS"
• Starting a new project or work session and wanting structure
• Needing autonomous task execution with momentum tracking
• Wanting to log work to a Convex backend for tracking and deduplication

The Sprint Loop

Every sprint follows this exact sequence:

1. ASSESS

What is the current state? What is the gap to the target outcome?

• Read the active task list, relevant files, and recent sprint log
• Identify where things stand right now
• Name the gap: what's missing between current state and the outcome?

2. PLAN

What is the single highest-leverage action available right now?

• Pick ONE thing to do in this sprint
• Apply the prioritization hierarchy (see below)
• Do not batch or multi-task

3. SCOPE

Define "done" in ≤5 minutes.

• Name the specific artifact this sprint will produce
• If it can't be done in 5 minutes, break it into a smaller sprint
• No sprint ends without a concrete output

4. EXECUTE

Do the work. Produce the artifact.

• Execute the scoped task
• Focus entirely on the output — no scope creep
• If you discover the scope was wrong, stop, re-scope, and continue

5. MEASURE

Did it move the metric? What changed?

• State the concrete result: what artifact was produced
• Name the relevant metric and whether it moved
• Be honest: "completed" vs "partially completed" vs "blocked"

6. ADAPT

Reprioritize. Kill what's not working.

• Based on the result, what should the NEXT sprint be?
• If 3 consecutive sprints produced no measurable movement: switch workstream or angle
• Never keep grinding on a dead approach — adapt immediately

7. LOG

Record to sprint log + (if configured) Convex.

Write a sprint log entry (see format below) to the sprint log file, and optionally POST to the Convex endpoint.

8. NEXT

Immediately begin the next sprint.

No gaps. No reflection breaks longer than 30 seconds. Momentum is the goal.

Sprint Rules

• Every sprint MUST produce a shippable artifact
• If >5 minutes, break into smaller sprints
• Never batch-plan more than 3 sprints ahead
• Bias toward momentum over perfection
• Every sprint must connect to an active outcome
• If blocked, log the blocker and skip to the next available sprint — never idle

Prioritization Hierarchy

Before every sprint, ask:

"If I could only do ONE thing in the next 5 minutes to move closer to the outcome, what would it be?"

1. Fix what's broken → Actively losing money or trust? Fix it first.
2. Optimize what's working → Something converting? Double down before exploring new.
3. Test new angles → Small experiments to find the next lever.
4. Build infrastructure → Only when 1–3 are humming.

Pivot Triggers

Stop the current workstream and pivot when:

• 3 consecutive sprints with no measurable movement → switch workstream or angle
• Channel hitting diminishing returns → reduce allocation, test alternatives
• Unexpected win (viral, press, referral spike) → drop lower-priority, capitalize immediately
• Customer feedback pattern emerging → elevate to top of sprint queue

Sprint Log Format

Write one entry per sprint to sprint-log.md in the working directory:

## Sprint [N] — [YYYY-MM-DD HH:MM]

**Project:** [project name]
**Workstream:** [marketing / development / content / research / etc.]
**Task:** [what you did]
**Artifact:** [what was produced — link or one-line description]
**Metric:** [what moved, or "no movement"]
**Status:** completed | partial | blocked
**Blocker:** [only if blocked — what's stopping you]
**Next sprint:** [what comes next]

Convex Integration (Optional)

If CONVEX_SPRINT_URL is set, POST every sprint log entry to the Convex HTTP endpoint. This enables:

• Sprint history across sessions
• Workstream breakdown reports
• Content deduplication (check before creating)
• Metric trend tracking

Setup

1. Deploy the Convex backend in scripts/convex-setup.md
2. Set CONVEX_SPRINT_URL to your Convex HTTP site URL (e.g., https://your-deployment.convex.site)
3. Sprints will auto-log on step 7 of each loop

Endpoints

Sprint Log Payload

curl -X POST $CONVEX_SPRINT_URL/sprints/log \
  -H "Content-Type: application/json" \
  -d '{
    "sprintId": 1,
    "project": "my-project",
    "workstream": "marketing",
    "task": "Write homepage headline variants",
    "artifact": "3 headline variants in headlines.md",
    "metric": "no movement yet",
    "status": "completed",
    "owner": "agent",
    "timestamp": 1740000000000
  }'

Script

Use scripts/log-sprint.sh for quick CLI logging:

./scripts/log-sprint.sh \
  --project "my-project" \
  --workstream "development" \
  --task "Fix checkout redirect bug" \
  --artifact "PR #42 opened" \
  --metric "checkout CVR: TBD pending deploy" \
  --status "completed"

Daily Rhythm

Morning

• Read active task list
• ASSESS the current state of all outcomes
• Set today's #1 priority
• Begin sprint 1

Continuous

• Sprint back-to-back, 5 minutes each
• Log every sprint (file + Convex if configured)
• Spawn sub-agents for heavy execution work
• Never stop between sprints for more than 30 seconds

End of Day

• Complete the sprint log
• Update active task list with what moved
• Set tomorrow's #1 priority
• Run scripts/log-sprint.sh --daily-summary if Convex is configured

Weekly (Friday)

• Review: which workstream had the most impact?
• Which sprints were wasted? Why?
• Biggest bottleneck assessment
• Restack priorities for next week

Reporting Formats

Daily Status

📊 DAY [X] — [DATE]
SPRINTS: [completed today] | TOP WIN: [best result]
BLOCKER: [biggest obstacle]
METRICS: [key metric] → [current value]
TOMORROW: [1–2 sentences]

Weekly Review

📈 WEEK [X] — [DATE RANGE]
SPRINTS: [total] (by workstream breakdown)
WINS: [top 3 with metrics]
MISSES: [top 3 with root cause]
LESSONS: [top 3]
NEXT WEEK: [top 3 priorities]
ESCALATIONS: [decisions needed from human]

Usage Examples

# Start sprint operating mode
"Enter sprint mode. My project is [X]. Target outcome: [Y]."

# Run a sprint
"Run sprint on: write 3 email subject line variants for the welcome sequence."

# Review recent sprints
"Show my sprint log for today."

# Weekly review
"Generate weekly sprint review."

# With Convex logging
"Log sprint: task=wrote homepage copy, artifact=homepage-v2.md, metric=awaiting test, status=completed"

File Structure

sprint-os/
├── SKILL.md                    ← This file
├── README.md                   ← Human-readable overview
└── scripts/
    ├── log-sprint.sh           ← CLI sprint logger (Convex optional)
    └── convex-setup.md         ← Instructions for Convex backend setup

*Sprint OS v1.0 — February 2026* *A product by Carson Jarvis (@CarsonJarvisAI)*