stelow-workflow-execution-critique

Execution Critique

sem diff is the primary diff tool. When sem is available, always use sem diff first for entity-level change detection (functions, types, methods). Use git diff only as fallback when sem is absent. The entire skill follows this convention — the fallback table in Tool Availability covers every sem command's git equivalent.

Standalone awareness: when inside stelow, reads scope + mode from .stelow/*/plans/spec-tech*.md and stelow.json. When standalone, auto-detects input type (plan path, directory, URL, or nothing). The HAS_WORKFLOW_DIR flag gates stelow-specific features; all audit criteria work in both modes.

Overview

Run a structured audit after any implementation — whether it followed the full stelow or was done ad-hoc. Every evaluation criterion runs in every mode; only the source of truth differs based on what input is available.

Tools: See ../stelow-workflow-orchestrator/references/cli-tools/subagents.md for subagent patterns. For large outputs, use bash with output truncation, or read with offset/limit.

🔀 Input Detection

Detect the input type before proceeding:

Input received:
  ├── Is a path ending in spec-tech*.md?
  │   └→ 🗺️ Mode: Workflow Audit (source of truth = plan scopes)
  ├── Is a path ending in spec-product*.md or spec-*.md?
  │   └→ 📋 Mode: Plan Audit (source of truth = product spec)
  ├── Is a directory path or URL?
  │   └→ 📁 Mode: Context Audit (source of truth = sem diff + git + session)
  ├── User described what changed verbally?
  │   └→ Use description to scope the audit. Still run sem diff / git diff
  │      to auto-discover changes, but use inline text as the primary
  │      scope anchor for what to evaluate.
  └── NOTHING / no input?
      └→ 💻 Mode: Standalone Audit (source of truth = auto-discovered)
         Run: sem diff HEAD~1 (or git fallback), git diff, session file list
         If nothing found: ask "What changed?"