langchain-ai/Deep Agents Code
Run dcode in GitHub workflows
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Activelast commit Jul 10, 2026
- License
- MIT
Pinned Snippet
uses: langchain-ai/deepagents@7e70065200007896336f38fe905803e6763e8f85 # deepagents==0.6.12tags can be moved; commit SHAs can't. why a SHA?
Inputs
| name | description | required | default |
|---|---|---|---|
| prompt | The prompt/instruction to send to the agent | yes | — |
| model | Model to use, as provider:model (e.g. openai:gpt-5.5, anthropic:claude-opus-4-6) or a bare name (claude-*, gpt-*, gemini-*) with the provider auto-detected. | no | — |
| model_params | JSON model kwargs passed to --model-params | no | "" |
| max_retries | Maximum transient model retries passed to --max-retries | no | "" |
| profile_override | JSON model profile overrides passed to --profile-override | no | "" |
| anthropic_api_key | Anthropic API key | no | — |
| openai_api_key | OpenAI API key | no | — |
| google_api_key | Google API key | no | — |
| github_token | GitHub token for API access | no | ${{ github.token }} |
| working_directory | Working directory for the agent | no | . |
| cli_version | deepagents-code version (empty = latest) | no | "" |
| skills_repo | GitHub repo of skills to clone (e.g. owner/repo, owner/repo@ref, or full URL) | no | "" |
| enable_memory | Persist agent memory across workflow runs using actions/cache. When enabled, memory is keyed by agent_name + memory_scope so the agent can recall prior context. | no | true |
| memory_scope | Cache scope: pr (shared per PR), branch (shared per branch), repo (shared across repo) | no | repo |
| agent_name | Agent identity name — controls memory namespace (default: agent) | no | agent |
| shell_allow_list | Comma-separated shell allow list passed to --shell-allow-list (default: recommended,git,gh) | no | recommended,git,gh |
| skill | Skill name to invoke before running the prompt | no | "" |
| startup_cmd | Shell command to run at startup before the prompt | no | "" |
| sandbox | Sandbox provider passed to --sandbox (empty = local execution) | no | "" |
| sandbox_id | Existing sandbox ID passed to --sandbox-id | no | "" |
| sandbox_snapshot_name | Snapshot or blueprint name passed to --sandbox-snapshot-name | no | "" |
| sandbox_setup | Setup script path passed to --sandbox-setup | no | "" |
| mcp_config | MCP config path passed to --mcp-config | no | "" |
| no_mcp | Pass --no-mcp to disable MCP loading | no | false |
| trust_project_mcp | Pass --trust-project-mcp to skip project MCP approval | no | false |
| interpreter | Set to true for --interpreter, false for --no-interpreter, empty for dcode default | no | "" |
| interpreter_tools | PTC allowlist passed to --interpreter-tools | no | "" |
| max_turns | Maximum agentic turns passed to --max-turns | no | "" |
| task_timeout | dcode task timeout in seconds passed to --timeout | no | "" |
| rubric | Acceptance criteria passed to --rubric (literal text or @path) | no | "" |
| rubric_model | Model passed to --rubric-model | no | "" |
| rubric_max_iterations | Maximum rubric grader iterations passed to --rubric-max-iterations | no | "" |
| quiet | Pass --quiet for clean stdout | no | false |
| no_stream | Pass --no-stream to buffer the full response | no | false |
| json | Pass --json for machine-readable dcode output | no | false |
| stdin | Pass --stdin and pipe prompt as standard input instead of using --non-interactive | no | false |
| timeout | Maximum action runtime in minutes (default: 30) | no | 30 |
Outputs
| name | description |
|---|---|
| response | Full text response from the agent |
| exit_code | Exit code from the agent |
| cache_hit | Whether agent memory was restored from cache (empty if enable_memory is false) |