Review — Workflow

Overview

How the review skill works, step by step.

Source Workflow

Claude skill workflow.

Step-by-step Workflow

review — Workflow

Auto-generated from SKILL.md. The skill itself is the source of truth; this is a human-readable map of what it does and how it runs.

What it does & when to use

Pre-landing PR review. Analyzes diff against the base branch for SQL safety, LLM trust boundary violations, conditional side effects, and other structural issues. Use when asked to "review this PR", "code review", "pre-landing review", or "check my diff". Proactively suggest when the user is about to merge or land code changes. (gstack)

Triggers

  • review this PR
  • code review
  • pre-landing review
  • check my diff

How it works

Workflow outline (sections of SKILL.md):

  1. Preamble (run first)
  2. Plan Mode Safe Operations
  3. Skill Invocation During Plan Mode
  4. Skill routing
  5. AskUserQuestion Format
  6. Artifacts Sync (skill start)
  7. Model-Specific Behavioral Patch (claude)
  8. Voice
  9. Context Recovery
  10. Writing Style (skip entirely if EXPLAIN_LEVEL: terse appears in the preamble echo OR the user's current message explicitly requests terse / no-explanations output)
  11. Completeness Principle — Boil the Lake
  12. Confusion Protocol
  13. Continuous Checkpoint Mode
  14. Context Health (soft directive)
  15. Question Tuning (skip entirely if QUESTION_TUNING: false)
  16. Repo Ownership — See Something, Say Something
  17. Search Before Building
  18. Completion Status Protocol
  19. Operational Self-Improvement
  20. Telemetry (run last)
  21. Plan Status Footer
  22. Step 0: Detect platform and base branch
  23. Step 1: Check branch
  24. Step 1.5: Scope Drift Detection
  25. Implementation Items
  26. Test Items
  27. Migration Items
  28. Cross-Repo / External Items
  29. Step 2: Read the checklist
  30. Step 2.5: Check for Greptile review comments
  31. Step 3: Get the diff
  32. Step 3.4: Workspace-aware queue status (advisory)
  33. Step 3.5: Slop scan (advisory)
  34. Prior Learnings
  35. Step 4: Critical pass (core review)
  36. Confidence Calibration
  37. Step 4.5: Review Army — Specialist Dispatch
  38. Step 5: Fix-First Review
  39. Step 5.5: TODOS cross-reference
  40. Step 5.6: Documentation staleness check
  41. Step 5.7: Adversarial review (always-on)
  42. Step 5.8: Persist Eng Review result
  43. Capture Learnings
  44. Important Rules

Components

Single-file skill — all instructions live in SKILL.md.

Invoke

  • Slash: /review
  • Or a natural-language request matching the triggers above.

Source: SKILL.md in this directory.

Execution Logic

The skill executes when its trigger fires (slash command, natural-language match, or direct invocation). It reads its references, applies its rules, and produces the documented outputs.

Edge Cases

See the source skill's references/ and scripts/ folders for edge-case handling.

Failure Handling

A skill failure surfaces as a tool error or a partial output; never a silent skip. Re-run with --verbose (where applicable) for diagnostics.

Integration Notes

  • Claude — invoked via the Skill tool with skill: "review".
  • Codex — referenced from AGENTS.md if mirrored.
  • Antigravity — referenced from the workspace agent rules if mirrored.
  • HQ Project — listed on the landing page Skills section + post-login sidebar.
  • MD Project (md.sgnk.ai) — file rendered from Skills/Review/workflow.md.
  • Obsidian — file rendered with frontmatter + tags.

Usage Examples

Invoke via slash command or natural language matching the skill description.

Source: ~/.claude/skills/review/WORKFLOW.md