Mermaid Mapper Workflow

Mermaid Mapper Workflow (Boss)

You are the Boss: orchestrate end-to-end mapping without stuffing an entire codebase into one context. Delegate discovery and per-file extraction to subagents; keep synthesis, state, and validation here.

This is a workflow-style skill: keep domain behavior in the component skills and use this file for sequencing, state, artifact paths, and recovery policy.

Scope

• Primary: object-oriented code, especially C# (classes, inheritance, composition, constructor injection, type families like *ViewModel).
• Extensible: other languages later; keep assignments file-and-type-bounded.

Artifact root (target repository)

All paths are relative to the target repository root (the project being mapped, not necessarily this tooling repo):

Create these directories on first run.

Skills (load these contracts)

Supporting references: skills/mermaid-scout/references/scoutmap-format.md, skills/mermaid-mapper/references/mermaid-conventions.md, and state transitions: references/state-model.md.

Invocation

The user passes the target in the message (no interactive scope prompts). Examples:

• /mermaid-mapper-workflow ResourceViewModel
• /mermaid-mapper-workflow current file
• /mermaid-mapper-workflow map RTS resource view models
• /mermaid-mapper-workflow map relationships from UnitViewModel

Resolve that text into concrete discovery instructions for the Scout (paths, globs, type names, feature keywords).

Model tiers

• Boss (you): use the default capable model for orchestration, recovery, and aggregation.
• Scout and Mapper subagents: spawn with model: "fast" when using the Task tool.

Execution loop

1. Init
   Ensure documentation/workflows/mermaid_mapper/, partials/, and logs/ exist on the target repo.

2. Resume check
   If scoutmap.md already exists and contains rows in progress, continue from persisted state (see references/state-model.md). Do not wipe partials unless the user asks for a clean run.

3. Scout
   Spawn a Scout subagent. Pass the full scout skill intent plus the resolved query and repo root. Scout writes or refreshes scoutmap.md with new rows starting at ⬜ per scoutmap format.

4. Assign
   Read scoutmap.md. Find rows with ⬜. For each batch you dispatch, set those rows to 🟩 and persist the file.

5. Map (parallel)
   Spawn one or more Mapper subagents with assignments: file path + type list per row (or batched rows if tiny). Mappers follow skills/mermaid-mapper/SKILL.md and write edge-only partials under partials/ with stable names.

6. Validate
   For each returned partial, apply mermaid-validation checks yourself (read skills/mermaid-validation/SKILL.md). Do not spawn a separate validator subagent unless fragments are huge.

7. State update

   • Pass: mark row ❎, keep partial.
   • Fail: mark row 🟥, append a short entry under logs/ with reasons.

8. Refill
   Repeat steps 4–7 until no ⬜ rows remain (iterative refill, not only one wave at startup).

9. Aggregate
   Run mermaid-aggregation (skills/mermaid-aggregation/SKILL.md) over all accepted partials. Write mermaid.md.

10. Done
    Tell the user where mermaid.md and scoutmap.md live.

Subagent dispatch

Use the Task tool for Scout and Mapper work. Each task prompt must include:

• The skill’s required inputs for that step.
• Exact paths on the target repo.
• Output path for partials when mapping.

Parallelism

The IDE may run several Mapper tasks at once. Cap concurrency only if the user or environment requires it; otherwise keep feeding ⬜ rows until the queue is clear.

Retry policy (v1)

🟥 is terminal. Optional future: requeue 🟥 → ⬜.

Standalone components

Users may run mermaid-mapper alone on one file, mermaid-scout alone for inventory, mermaid-validation on a fragment, or mermaid-aggregation on saved partials without running this full workflow.