Superplan (Single-Skill Hierarchical Planner)

Single-skill planner. /superplan executes the selected tier directly in one invocation.

Progressive Loading Contract (Skill Graph)

For each invocation, load in order:

1. graph/index.md
2. graph/mocs/router-flow.md
3. Required nodes under graph/nodes/
4. Tier playbook docs only after tier selection:
   • big: tiers/big/graph/index.md
   • medium: tiers/medium/graph/index.md
   • small: tiers/small/graph/index.md

Hard Rules

1. Resolve valid target_directory before tier work.
2. Execute exactly one tier (big or medium or small) per invocation.
3. Keep all generated files inside the ticket/project folder tree.
4. Do not chain-run another tier automatically in the same invocation.
5. Preserve existing files via read/update (no blind overwrite).

Hierarchy Model

• big (ticket root): main.md + phase-*.md
• medium (expand one phase): phase-<N>/main.md + item-*.md
• small (expand one item): phase-<N>/item-<N>/main.md + task-*.md + plan.md

Use lazy expansion: only selected phase/item is expanded.

Flow

Tier Execution Mapping

• big -> execute big-tier playbook (tiers/big)
• medium -> execute medium-tier playbook (tiers/medium)
• small -> execute small-tier playbook (tiers/small)

Always pass/use:

• target_directory
• selected tier
• ticket/project metadata
• user goals/constraints
• phase_file / item_file when available

Return Contract

Return:

• selected tier
• target directory
• created/updated artifact paths
• expansion context used (phase_file, item_file when present)
• recommended next /superplan ... command when useful