/next-move

Predict the best next move, present it clearly, and — on approval — hand it to the real WinDAGs execution path.

Two operating rules are mandatory:

1. Separate planning topology from runtime topology. Server execution is real for dag and workflow. team-loop, swarm, blackboard, team-builder, and recurring fall back to generic DAG execution. See references/runtime-honesty.md.

1. Prefer live WinDAGs visualization over ASCII when execution is going to happen. ASCII is fallback only. See references/live-execution-visualization.md.

Arguments: $ARGUMENTS

If the user passed --fresh, ignore conversation history and predict from project signals only.

Skill Layout

This skill is structured for on-demand loading. Don't bulk-load — pull only what the current stage needs.

Project Signals

These are preprocessed at skill-load time via !... shell calls. Treat them as ground truth.

Git State

!`git status --short 2>/dev/null || echo "Not a git repo"`

Branch: !git branch --show-current 2>/dev/null || echo "unknown"

Recent Commits

!`git log --oneline -8 2>/dev/null || echo "No commits"`

What Changed

!`git diff --stat 2>/dev/null || echo "No unstaged changes"`

Staged Changes

!`git diff --cached --stat 2>/dev/null || echo "Nothing staged"`

Recently Modified Files

!`git diff --name-only HEAD~3 2>/dev/null || echo "No recent file changes"`

CLAUDE.md

!`head -60 CLAUDE.md 2>/dev/null || echo "No CLAUDE.md found"`

Package / Project Info

!`cat package.json 2>/dev/null | head -40 || echo "No package.json"`

Port Daddy

!`pd find 2>/dev/null || echo "Port Daddy not installed. portdaddy.dev"`

!`pd notes 2>/dev/null || echo ""`

!`pd salvage 2>/dev/null || echo ""`

!`pd whoami 2>/dev/null || echo ""`

Prior WinDAGs Predictions

!`ls -1t .windags/triples/ 2>/dev/null | head -5 || echo "No prior predictions"`

Decision Flow

flowchart TD
  A[Gather signals + conversation context] --> B[Sensemaker]
  B --> C{Halt gate?}
  C -->|trip| D[Halt: name the contradiction\nload templates/halt-gate-response.template.md]
  C -->|pass| E[Decomposer]
  E --> F[Skill narrowing via SkillSearchService]
  F --> G[Skill Selector + PreMortem in parallel]
  G --> H[Synthesizer: planning + runtime topology]
  H --> I{Execution likely?}
  I -->|yes| J[Open live WinDAGs surface]
  I -->|no| K[Plan-only preview]
  J --> L[Accept / Modify / Change topology / Reject]
  K --> L
  L -->|Accept| M[Build skillful nodes\nload references/skillful-node-execution.md]
  L -->|Modify| N[Mutate vs restart\nload references/modify-vs-restart.md]
  L -->|Change topology| O[Re-evaluate runtime honesty\nload references/runtime-honesty.md]
  L -->|Reject| P[Record rejection in triple]
  M --> Q[Execute through real runtime]

For depth on any decision point, load the matching reference. Don't try to derive judgment-call rules from the diagram alone.

Core Workflow (Compact)

The full workflow is in the references; this is the lookup version.

1. Sensemaker → Load prompts/sensemaker.md. Validate output against schemas/sensemaker-output.schema.json.
2. Halt gate → Apply references/halt-gate-discipline.md. If trips: render templates/halt-gate-response.template.md, store no triple, stop.
3. Decomposer → Load prompts/decomposer.md. 3-7 subtasks. Validate against schemas/decomposer-output.schema.json.
4. Skill narrowing → Use mcpwindagswindagsskillsearch (routes through SkillSearchService). See references/skill-narrowing-cascade.md for internals.
5. Skill Selector + PreMortem → Load prompts/skill-selector.md and prompts/premortem.md. Run in parallel.
6. Synthesize → Apply references/topology-selection.md and references/runtime-honesty.md. Validate against schemas/predicted-dag.schema.json.
7. Present → Live WinDAGs UI preferred. See references/live-execution-visualization.md.
8. Execute on accept → Build skillful nodes. See references/skillful-node-execution.md and prompts/execution-node.md.
9. Modify / restart → See references/modify-vs-restart.md.
10. Store triple → Write to .windags/triples/. See references/triple-feedback-loop.md.

When to Refuse

/next-move halts on ambiguity. Halts are first-class output, not failure. See references/halt-gate-discipline.md. Worked example: examples/03-halt-gate-tripped.md.

When Things Look Wrong

Catalog of failure modes with detection signals and fixes: references/failure-modes.md.

Worked Examples

Load only when you need shape recognition end-to-end:

• examples/01-feature-delivery-happy-path.md — clean signals, native DAG
• examples/02-debugging-blackboard-shape.md — runtime honesty in practice
• examples/03-halt-gate-tripped.md — pipeline correctly refuses

Quality Gates

Before sending the prediction:

• Sensemaker, Decomposer, Skill Selector, PreMortem all returned valid structured output
• Halt gate enforced when ambiguity remained
• Planning topology and runtime topology both present and honest
• Every node has a primary skill, input contract, output contract, and commitment_level
• Approval steps modeled as real gates (human-gate-designer), not prose
• Live visualization attempted before falling back to ASCII
• Triple stored only on completion or user feedback (never on halt)

NOT-FOR Boundaries

• Creating skills → use skill-creator or skill-architect
• One-off bug debugging → use fullstack-debugger
• Promising native execution of unsupported topologies → see references/runtime-honesty.md
• ASCII theater when live UI is available → open the real surface
• Ad hoc subagent prompts → use prompts/execution-node.md and references/skillful-node-execution.md