render-recipe

Render Recipe

Produce a compact, structured overview of an AutoSkillit recipe. Reads the recipe YAML (provided in the prompt or loaded via load_recipe), analyzes the step graph, and writes a formatted Markdown file to temp/render-recipe/.

When to Use

• User explicitly says "render recipe", "show recipe overview", or invokes /render-recipe
• Do NOT auto-invoke when a recipe is loaded — only on explicit user request

Critical Constraints

NEVER:

• Modify any source code or recipe files
• Invent steps, ingredients, or routing not in the YAML
• Add decorative flair, emoji, or unnecessary commentary
• Create files outside temp/render-recipe/ and recipes/diagrams/
• Use Unicode characters, emoji, or non-ASCII symbols. ASCII only — use * for pass, x for fail, ? for confirm, | for spine, + for joins. No box-drawing characters, no arrows like → or ↑, no check marks, no crosses. Write -> and (up) instead.
• Include HTML comments, hash markers, format version markers, or any metadata in the output. No <!-- ... --> lines.
• Include "Agent-managed" lines listing internal context variables. The user doesn't need to see plumbing state.

ALWAYS:

• Read the recipe YAML carefully and render exactly what exists
• Write output to temp/render-recipe/{recipe-name}_{YYYY-MM-DD_HHMMSS}.md
• Print the rendered content to terminal after writing the file

---

Rendering Specification

The output has exactly two sections. Follow this structure precisely.

Note: The inputs table is NOT part of the diagram output. It is generated at runtime from the recipe YAML by _format_ingredients_table (single source of truth). Do not include an inputs table in the diagram file.

Section 1: Header

## {name}

Just the name. No description — the user already chose the recipe from the picker.

Section 2: Flow Diagram

Build an ASCII flow diagram showing the step graph. This is the core visual.

Infrastructure steps to hide: The diagram shows ONLY the skill work — run_skill invocations and test_check. Everything else is plumbing. Hide:

• All run_cmd steps — value capture, shell one-liners, branch computation. All plumbing.
• All push_to_remote steps — mechanical transport.
• All clone_repo steps — setup plumbing.
• All issue lifecycle steps — get_issue_title, claim_issue, release_issue. Plumbing.
• All branch creation steps — create_unique_branch, create_branch. Plumbing.
• All CI/CD steps — wait_for_ci, wait_for_merge_queue, enable_auto_merge, route_queue_mode, reenter_merge_queue, queue_ejected_fix, diagnose_ci. Plumbing.
• All merge_worktree steps — worktree-back-to-branch merge. Plumbing.
• All cleanup steps — delete_clone, cleanup_failure, confirm_cleanup. Teardown.
• All post-PR steps — review_pr, resolve_review, diagnose_ci, resolve_ci. Post-delivery plumbing.
• All action: route steps — pure routing nodes with no work.