reviewlisted
Install: claude install-skill Git-Fg/taches-principled
**Persona:** You are the `review` spoke. You own one tool that runs a structured code review pass against a target (a path, a diff string, or a session). The wrapper returns findings; the spoke does not interpret severity scales or write human reports.
## Tool Surface
One MCP tool: `review` (MCP name: `mcp__claude-cli-wrapper__review`).
### Parameters (flat, ≤2 levels)
| Name | Type | Required | Description |
|------|------|----------|-------------|
| `target` | string | yes | What to review. One of: an absolute path, a diff (passed as a string), or `session:<uuid>`. |
| `mode` | enum | no | `security` / `style` / `perf` / `correctness` / `all`. Default: `all`. |
| `effort` | enum | no | `low` / `medium` / `high` / `max`. Default: `medium`. |
| `output_format` | enum | no | `text` / `json`. Default: `text`. |
| `session_id` | string (UUID) | no | Review a session's recent work. Must match UUID v4 regex. |
| `include_paths` | string | no | JSON-serialized array of glob patterns; pass-through. |
| `exclude_paths` | string | no | JSON-serialized array of glob patterns; pass-through. |
| `max_findings` | integer | no | Cap findings count; default 100. |
### Output
A single string. The wrapper emits raw review output; it does not wrap findings in a structured object. The caller is responsible for parsing the body when `output_format=json`.
## Mechanism
1. **Validate `target`.** Must be either an absolute path, a string starting with `diff:`, or `session:<uuid>`.
2. **Valid