# Check

> Use when the repo has a sverka workflow (sverka.config.ts) or the user asks to run checks/scans through bro. Thin wrapper over `bro check` — mechanics live in the CLI; this skill carries policy only.

- Skill: `theplenkov/check` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx get-skill add theplenkov/check`
- Raw SKILL.md: https://api.skillmd.com/api/skills/theplenkov/check/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: theplenkov (https://skillmd.com/u/theplenkov)
- Updated: 2026-11-17
- Page: https://skillmd.com/skills/theplenkov/check

---


# /check (bro)

**All mechanics live in the `bro` CLI.** This skill is policy only.

`bro check` is bro's check facade: **sverka is the executor** — it loads
`sverka.config.ts`, binds an Entry into a Run Plan (plan-time evaluation
can materialize artifacts like gqlb-built queries), runs steps with
parallelism, and returns per-step stats plus optional SARIF findings.
bro parses `sverka run --format json` and renders the report — step
status/duration lines, totals, findings + verdict when evaluated.

## Commands

| Command                  | What it does                                                                                                                                                                 |
| ------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `bro check`              | Run the repo's sverka workflow (default entry, text report). Exit code mirrors sverka — CI-able                                                                              |
| `bro check --json`       | Structured output: `{command:"check", data:{planId,status,steps[],findings?,verdict?}, durationMs}`; on evaluation failure also `evaluationStatus:"failed", evaluationError` |
| `bro check --evaluate`   | Also collect `*.sarif` artifacts and run sverka's policy gate (verdict→exit code)                                                                                            |
| `bro check --root <dir>` | Run against another checkout — its bro.config and node_modules apply                                                                                                         |

`--config`, `--entry`, `--executor host|docker`, `-q`, `-v` pass through
to `sverka run`.

## Config

```json
{
  "check": {
    "bin": "sverka",
    "config": "sverka.config.ts",
    "entry": "default",
    "executor": "host",
    "evaluate": false
  }
}
```

All fields optional: `bin` skips binary resolution entirely; `config`,
`entry`, `executor` map to the same-named `sverka run` flags;
`evaluate: true` turns on SARIF collection + the policy gate.

Flags win over config. Binary resolution: `check.bin` → the repo's
pinned install (`node_modules/sverka`, then the deprecated
`node_modules/@sverka/cli`, then `.bin` shims, walking up from root) →
`sverka` on PATH → the `sverka` bundled with `@broject/bro`.

## Policy

- **Prefer `bro check` over a bare `sverka run`** in repos wired for bro —
  same execution, bro-shaped output, and the exit code already mirrors
  the gate (nonzero on step failure or failing policy verdict).
- `evaluate: true` only for configs whose steps declare SARIF artifact
  outputs — otherwise sverka has nothing to collect and the run fails
  `COLLECTION_FAILED` (exit 3, `evaluationStatus: failed` in the JSON
  envelope; the run is never retried — fix the config).
- Findings→beads dedup is not wired yet — `bro check` reports findings;
  it does not file them.

