gencersarp/Agent Sandbox
Run LLM agents safely inside CI with strict capability manifests, resource limits, and human approval.
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Activelast commit May 21, 2026
- License
- None
Pinned Snippet
uses: gencersarp/agent-sandbox@489f025ffe8401271599f2fdee9f84d76233d9af # no releases — HEAD as of 2026-07-11tags can be moved; commit SHAs can't. why a SHA?
Inputs
| name | description | required | default |
|---|---|---|---|
| manifest_path | Path to the .agent-sandbox.yml manifest file relative to the repo root. | no | .agent-sandbox.yml |
| llm_api_key | API key for the LLM endpoint (pass via secrets). | yes | — |
| llm_api_url | Override the LLM API URL (default: OpenAI chat completions). | no | https://api.openai.com/v1/chat/completions |
| llm_model | LLM model to use. | no | gpt-4o |
| create_pr | Whether to create a draft PR from the agent's changes (true/false). | no | false |
| base_branch | Base branch for the PR and diff context. | no | main |
| dry_run | Whether to run in dry run mode (true/false). Shows changes but does not generate patches/PRs. | no | false |
Outputs
| name | description |
|---|---|
| report_path | Path to the JSON report file. |
| patch_path | Path to the unified diff patch file. |
| pr_url | URL of the created draft PR (empty if create_pr is false or no changes). |