Code Investigate Skill

SkillAI & models

Dual-perspective code investigation. Use when: deep code analysis needing both Claude and Codex perspectives. Not for: quick exploration (use code-explore), code review (use codex-code-review). Output: integrated findings from dual analysis.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Code Investigate Skill skill

What this skill tells your AI

The instructions your AI receives, as published by sd0xdev/sd0x-harness in skills/code-investigate/SKILL.md and read by ahel’s review.

Trigger

  • Keywords: investigate code, how feature works, trace implementation, dual confirmation, deep dive, how code works, what this code does, code research

When NOT to Use

  • Just need quick lookup (use Grep/Glob directly)
  • Code review (use codex-review)
  • System verification (use feature-verify)
  • Git history tracking (use git-investigate)

Core Principle

Codex must explore independently. Feeding Claude's conclusions to Codex is prohibited.
┌─────────────────┐     ┌─────────────────┐
│ Claude Explores  │     │ Codex Explores   │
│ Independently    │     │ Independently    │
│   (Phase 1-2)   │     │    (Phase 3)     │
└────────┬────────┘     └────────┬────────┘
         │                       │
         ▼                       ▼
   ┌───────────┐           ┌───────────┐
   │ Claude    │           │ Codex     │
   │ Conclusion│           │ Conclusion│
   └─────┬─────┘           └─────┬─────┘
         │                       │
         └───────────┬───────────┘
                     ▼
              ┌─────────────┐
              │ Consolidated│
              │   Report    │
              │  (Phase 4)  │
              └─────────────┘

Workflow

PhaseNameActionOutput
1Claude ExploreGrep/Glob/Read to search codeRelated files list
2Claude ConcludeAnalyze logic, form understandingInitial conclusion (internal)
3Codex ExploreInvoke Codex exec to explore independentlyCodex analysis report
4IntegrateCompare both perspectives, mark differencesConsolidated report

Codex Invocation Rules

Dispatch Phase 3 per @skills/codex-code-review/references/codex-transport.md § Start with the template in references/prompts.md. The transport pins the sandbox and the approval policy, and it derives the working directory from git rev-parse --show-toplevel — none of the three is chosen here, and this skill no longer restates them.

Bind every placeholder before writing prompt.md. The template is body-only, so nothing evaluates an expression inside it: ${USER_QUESTION} and ${PROJECT_PATH} must carry real values by the time the file is written.

Correct Approach

Code Investigation Task

Question

${USER_QUESTION}

Project Info

  • Path: ${PROJECT_PATH}
  • Tech Stack: ${TECH_STACK}

Please independently explore the codebase and answer:

  1. What files are related?
  2. How does the core logic work?
  3. What is the data flow?
  4. What are the key dependencies?

Please grep/read and explore on your own, then provide your analysis.

Prohibited Approaches

PatternProblemExample
Feeding conclusionClaude's findings leak to CodexClaude found these files: ${findings}
Leading questionPresupposes answerI think the problem is in cache, verify
Scope restrictionPrevents independent explorationOnly look at src/service/

Output

## Investigation Report
- **Claude findings**: <independent analysis>
- **Codex findings**: <independent analysis>
- **Integrated conclusion**: <merged findings>
- **Confidence**: High / Medium / Low

Verification Checklist

CheckStandard
Claude independent conclusionPhase 2 forms conclusion, not output to user
Codex prompt is cleanContains only question + project path, no Claude findings
Report perspectives separatedClaude / Codex conclusions presented separately
Integration is completeMarks agreement, differences, possible gaps

References

FilePurposeWhen to Read
references/prompts.mdCodex prompt templatesBefore Phase 3
references/output-template.mdReport formatDuring Phase 4

Examples

Feature Investigation

Input: Investigate how order processing works
Phase 1: Grep "processOrder" -> Read src/service/order/*.ts
Phase 2: Form understanding: Controller -> Service -> Repository write
Phase 3: Codex explores independently (only given question + path)
Phase 4: Consolidated report -> mark both perspectives

Mechanism Understanding

Input: How does the API caching mechanism work?
Phase 1: Grep "cache" + "portfolio" -> Read related files
Phase 2: Understand Redis TTL + fallback mechanism
Phase 3: Codex investigates independently
Phase 4: Compare differences -> output consolidated report

Problem Diagnosis

Input: Why is token price sometimes null?
Phase 1: Search price-related logic + error handling
Phase 2: Identify possible fallback paths
Phase 3: Codex diagnoses independently
Phase 4: Synthesize both findings -> list possible causes

Signals

GitHub stars
188
Forks
24
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
code-investigate
Source
github.com/sd0xdev/sd0x-harness