> ## Documentation Index
> Fetch the complete documentation index at: https://docs.declaw.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Code reviewer agent

> Agent clones a public repo, runs ruff, applies auto-fixes, and emits a structured review inside a declaw ai-agent sandbox.

## Use case

Agent-driven code review. Clone a repo, lint it, let the agent
propose fixes via the `apply_patch` tool, summarize findings. The
review never runs on your laptop — the clone, the linter, and the
diff all live in a throwaway VM.

## Template

`ai-agent` — large template with common agent-framework deps
pre-installed (langchain, crewai, autogen, plus git/python
tooling). Good choice when the agent needs to run Python code
with a rich set of imports without a `pip install` delay.

## Run it

```bash theme={null}
export DECLAW_API_KEY=dcl_...
export DECLAW_DOMAIN=api.declaw.ai
export OPENAI_API_KEY=sk-...
# Optional: review a different repo
export TARGET_REPO=https://github.com/<owner>/<repo>.git

python cookbook/examples/openai-agents-code-reviewer/main.py
```

## Security policy

```python theme={null}
SecurityPolicy(
    injection_defense=InjectionDefenseConfig(enabled=True, sensitivity="medium", domains=["api.openai.com"]),
    network=NetworkPolicy(
        allow_out=[
            "api.openai.com",
            "pypi.org",
            "files.pythonhosted.org",
            "github.com",
            "codeload.github.com",
            "objects.githubusercontent.com",
        ],
    ),
)
```

Injection defense matters because an adversarial `README.md` could
try to override the reviewer's system prompt. The scanner runs on
the request body before the LLM call — the agent code doesn't have
to implement any defense itself.

## Env isolation in practice

```python theme={null}
envs={
    "REVIEWER_ID": "rev-001",
    "REVIEW_DEPTH": "standard",
    "TARGET_REPO": target_repo,
}
```

The agent reads these with `printenv` rather than having them in
the system prompt. This means:

* No secret values in model traces or guardrails logs.
* Rotating a value doesn't need a prompt change.
* Per-reviewer customization (depth, id) stays structured.

## What the agent does

1. `printenv REVIEWER_ID REVIEW_DEPTH TARGET_REPO`
2. `git clone --depth 1 $TARGET_REPO /workspace/repo`
3. `pip install -q ruff`
4. `ruff check /workspace/repo`
5. `ruff check --fix /workspace/repo && git -C /workspace/repo diff`
6. Write `/workspace/review.md` with sections for metadata,
   findings, auto-fixes, and remaining action items.

## Expected output

```
== /workspace/review.md ==
## Reviewer metadata
- id: rev-001 · depth: standard · repo: https://...

## Lint findings
- flask/examples/.../app.py:12 E501 line too long
...

## Auto-fixes applied
- 3 files modified by `ruff check --fix`
...
```

## Full source

See `cookbook/examples/openai-agents-code-reviewer/main.py` in the repo.