Agent Skills: schedule-burn

Create a visual implementation plan with mermaid when delegating a single burn to a sub-agent. Includes decision diamonds with exit ramps for early abort when unexpected issues arise. The sub-agent does not know about orbit - use implementation language only.

UncategorizedID: viktorvan/dot-files/schedule-burn

Install this agent skill to your local

pnpm dlx add-skill https://github.com/viktorvan/dot-files/tree/HEAD/opencode/skill/schedule-burn

Skill Files

Browse the full folder contents for schedule-burn.

Download Skill

Loading file tree…

opencode/skill/schedule-burn/SKILL.md

Skill Metadata

Name
schedule-burn
Description
Create a visual implementation plan with mermaid when delegating a single burn to a sub-agent. Includes decision diamonds with exit ramps for early abort when unexpected issues arise. The sub-agent does not know about orbit - use implementation language only.

GOAL Create a Mermaid flowchart implementation plan for delegating a single burn to a sub-agent. The plan must include clear decision points with "exit ramps" - paths where the sub-agent should STOP and report back to the captain rather than attempting to solve unexpected issues on its own.

IMPORTANT: The sub-agent (pilot) does NOT know about orbit, waypoints, or burns. Your diagram must use plain implementation language only. Describe what to do, not which burn it is.

WHEN TO USE

  • Before delegating a burn to the pilot agent
  • For any burn that has potential failure modes or requires specific conditions to be met
  • When you want the sub-agent to abort early if assumptions don't hold

SCOPE CHECK - BEFORE BUILDING THE DIAGRAM

One burn = one narrow behavior slice with 3-8 action steps.

When in doubt, make the burn smaller. A too-small burn has zero cost. A too-large burn causes repeated aborts and wasted cycles.

A burn should do one thing — e.g., extract one piece of shared logic, migrate one caller, add tests for one behavior, or fix one defect — not combinations of these.

If the description uses "and" between independent actions, split it.

Each implementation burn follows TDD: write tests before the code they test, not after. A burn may contain multiple red → green cycles. Tests and the implementation they verify must be in the same burn — never split across separate burns.

MERMAID SYNTAX

flowchart TD
    Start[Start: Brief description of what to implement] --> Step1[First action step]
    Step1 --> Check1{Step 1 successful?}
    Check1 -->|yes| Step2[Second action step]
    Check1 -->|no| Abort1[[ABORT: Report - step 1 failed because X]]
    Step2 --> Check2{Step 2 successful?}
    Check2 -->|yes| Step3[Third action step]
    Check2 -->|no| Abort2[[ABORT: Report - step 2 failed because Y]]
    Step3 --> Check3{Step 3 successful?}
    Check3 -->|yes| End[End: Expected outcome]
    Check3 -->|no| Abort3[[ABORT: Report - step 3 failed because Z]]

NODE SHAPES

  • Actions/steps: [square brackets]
  • Decisions/conditions: {curly braces}
  • ABORT points: [[double brackets]] - sub-agent must stop and report back
  • End states: [square brackets]

CRITICAL RULES

  1. EVERY action step MUST be followed by a decision diamond with success/failure paths

    • BAD: Step1 --> Step2 --> Step3 (no verification)
    • GOOD: Step1 --> Check1{OK?} --> Step2 --> Check2{OK?} --> Step3
  2. Every decision diamond has two paths:

    • -->|yes| continues to next step
    • -->|no| leads to [[ABORT: Report - <specific reason>]]
  3. NO orbit terminology in the diagram (no "burn", "waypoint", "orbit", "trajectory")

    • Use plain implementation language: "step", "action", "task"

OUTPUT FORMAT After creating the mermaid diagram, schedule the burn by passing the diagram via stdin using heredoc syntax (do NOT save to a file first):

orbit burn schedule <orbit-id> <waypoint> <description> <instructions> [--context "..."] <<'EOF'

[your diagram]

EOF

Where <description> is a concise summary (1-2 sentences), <instructions> are detailed steps, or a bulleted list for the implementation, and --context should contain e.g. related files, files to create, rationales, background.

Then delegate to the pilot agent with YAML front-matter only:

---
orbit-id: MED-7620
burn-number: 3
---

The pilot will automatically fetch the burn plan and mark it as initiated using orbit burn initiate.

AFTER DELEGATION

When the sub-agent returns:

  • On success: orbit burn complete
  • On abort: analyze reason, then orbit burn abort and schedule a corrected burn, or consult user if fundamental

EXAMPLE - Adding an API endpoint

flowchart TD
    Start[Start: Add CSV export endpoint to PatientController] --> S1[Find existing controller and identify endpoint patterns]
    S1 --> C1{Controller found with consistent patterns?}
    C1 -->|yes| S2[Write integration test for GET /patients/export endpoint]
    C1 -->|no| Abort1[[ABORT: Report - controller not found or inconsistent patterns]]
    S2 --> C2{Test written and fails as expected?}
    C2 -->|yes| S3[Implement export endpoint following existing patterns]
    C2 -->|no| Abort2[[ABORT: Report - test does not compile or does not fail as expected]]
    S3 --> C3{Test passes?}
    C3 -->|yes| S4[Run full test suite to verify no regressions]
    C3 -->|no| Abort3[[ABORT: Report - implementation does not make test pass, describe errors]]
    S4 --> C4{All tests pass?}
    C4 -->|yes| End[End: Export endpoint added with passing test]
    C4 -->|no| Abort4[[ABORT: Report - existing tests failing, describe failures]]