sego-review

name	sego-review
version	0.1.0
description	AI coding engineering trust review. After the user generates or modifies code with any AI tool, invoke Sego to perform a structured review. Sego returns findings with severity/file/line/evidence/risk/suggestion and persists a review artifact to .sego/reviews/. Use this skill whenever the user wants to check, review, or validate AI-generated code changes before committing or shipping.

Sego Review Skill

When to use this skill

Use this skill when the user:

• Asks to "review", "check", or "validate" code changes
• Wants to know if AI-generated code is safe to commit
• Asks "is this code okay?" or "should I commit this?"
• Wants a second opinion on AI-generated diffs

Do NOT use this skill for:

• General coding questions (use normal tools)
• Running tests (use cargo test / npm test directly)
• Git operations (use git directly)

How to invoke

Call the Sego sidecar via the platform-appropriate script:

Unix / macOS

echo '{"schema_version":1,"action":"review","cwd":"'"$PWD"'","scope":"staged"}' \
  | bash skills/sego-review/scripts/sego-review.sh

Windows (PowerShell)

$request = @{ schema_version = 1; action = "review"; cwd = (Get-Location).Path; scope = "staged" } | ConvertTo-Json
$request | powershell -File skills/sego-review/scripts/sego-review.ps1

Request format

{
  "schema_version": 1,
  "action": "review",
  "cwd": "/absolute/path/to/project",
  "scope": "staged",
  "options": { "model": null },
  "context": { "user_intent": "optional description of what the user wants" }
}

• scope: "staged" (git staged changes), "workspace" (all uncommitted changes), or a path like "src/auth".
• options.model: override the review model (default: auto-selected).

Response format

{
  "schema_version": 1,
  "status": "ok",
  "review_id": "rev-20260616-001",
  "diff_hash": "sha256:...",
  "artifact_path": ".sego/reviews/rev-20260616-001.json",
  "findings": [
    {
      "id": "f-001",
      "severity": "high",
      "file": "src/auth/login.rs",
      "line": 42,
      "title": "Missing error handling",
      "evidence": "...",
      "risk": "...",
      "suggestion": "...",
      "confidence": 0.85
    }
  ],
  "parse_status": "structured"
}

Severity levels: critical > high > medium > low > info.

Interpreting results

After receiving the response:

1. Summarize the findings for the user in plain language
2. Highlight any critical or high severity findings first
3. If status is "error", report the error message to the user
4. Mention the artifact_path so the user can review the full report later

Requirements

• Sego must be installed and on PATH, or built from source (cargo build -p rusty-claude-cli)
• The project must be a git repository (review uses git diff)