YAML for Claude Code

View as Markdown

Run Claude Code AI assistant with Anthropic integration.

YAML parameters for Claude Code

PARAMETERS
integrationrequiredstring
Integration identifier for authenticating with the AI provider. It refers to an integration in the workspace by its identifier, not by its display name.
promptsrequired(string | object)[]
List of prompts for the agent. Each entry is either an inline string or a file reference (type INLINE or FILE).
typerequiredstring

Value: CLAUDE_CODE

actionrequiredstring
Unique identifier for the action within the pipeline.
effortstring
Reasoning effort: low, medium, high, xhigh, or max (default high)
claude_argsstring
Additional CLI arguments to pass to Claude Code
dangerously_skip_permissionsboolean
When set to true, bypasses permission checks in Claude Code
modelstring
The Claude model to use (e.g. sonnet, opus, haiku).
skillsstring[]
List of agent skills to install before the run. Each entry is an owner/repo, a repository URL, or a local path.
mcp_serversRecord<string, McpServerYaml>
MCP servers to make available to the agent, keyed by server name. Each server is either http/sse (url, headers) or stdio (command, args, env).
working_directorystring
The directory in which the pipeline filesystem will be mounted.
setup_commandsstring
The command that will be executed only on the first run. Several commands go in one multi-line string (a | block), one per line - not a YAML list.
main_service_namestring
The hostname of the container in which the action is run. The container will be available under this name in the docker network for services.
servicesServiceYaml[]
The list of services attached to the build environment.
trigger_timestring enum
Specifies when the action should be executed. ON_SUCCESS actions run after every successful run, also one ended by an EARLY_EXIT action.
Allowed enum:
ON_EVERY_EXECUTION,
ON_SUCCESS,
ON_FAILURE,
ON_BACK_TO_SUCCESS,
ON_WARNING,
ON_WAIT_FOR_APPROVE,
ON_TERMINATE
loopstring[]
Names of the variables to loop over (IMAGES, not $IMAGES and not a value). The action runs once for each item of the value, and once for each combination of items when several variables are listed. The variable value is split into items by new lines, or by commas when it has no new lines (a single-line JSON object or key=value pairs stays one item). An item that is a JSON object or key=value pairs also sets one variable per key, named <VARIABLE>_<key>; an item that starts with EXTRACT: sets them as plain <key>.
run_only_on_first_failureboolean
When true, the action runs only on the first failure instead of on each failure. Used only when trigger_time is ON_FAILURE.
disabledboolean
When set to true the action is disabled. By default it is set to false.
timeoutinteger
The timeout in seconds. It cannot exceed the action time limit of the workspace plan; a higher value is rejected.
ignore_errorsboolean
If set to true the execution will proceed, mark action as a warning and jump to the next action. Doesn't apply to deployment actions.
retry_intervalinteger
Delay time between auto retries in seconds.
retry_countinteger
Number of retries if the action fails.

Constraints: ≤ 10

run_nextstring enum
Defines whether the action should run in parallel with the next one.
Allowed enum:
WAIT_ON_SUCCESS,
IN_SOFT_PARALLEL,
IN_HARD_PARALLEL
trigger_conditionsTriggerConditionYaml[]
The list of trigger conditions to meet so that the action can be triggered.
variablesVariableYaml[]
The list of variables you can use in the action.

Outputs for Claude Code

OUTPUT VARIABLES
BUDDY_ACTION_CLAUDE_OUTPUT_FILE
Path to the main result file of this action, relative to the pipeline filesystem root.

Example: .buddy/output/claude/report_12_a1b2c3d.json

BUDDY_ACTION_CLAUDE_CONCLUSION
Conclusion of the agent run: success or failure. Exported also when the action fails.

Example: success

BUDDY_ACTION_CLAUDE_RESULT
Final message returned by the agent. Exported when the agent produced a result.

Example: Fixed the failing test in UserServiceTest and updated the changelog.

BUDDY_ACTION_CLAUDE_SESSION_ID
Identifier of the agent session, usable to resume the conversation in a later action. Exported when the agent reports a session.

Example: 9f8c1e2a-4b7d-4c31-8e5a-2d6f0b1c7e94

OUTPUT FILES
.buddy/output/claude/report_$BUDDY_RUN_ID_$BUDDY_ACTION_RUN_HASH.json
Session report of the Claude Code run. Written when the agent produces a report.

Last modified on Nov 19, 2026

YAML examples for Claude Code

Run Claude Code with all options

yaml
- action: "Run Claude Code"
  type: "CLAUDE_CODE"
  trigger_time: "ON_EVERY_EXECUTION"
  integration: "my_anthropic_integration"
  prompts:
    - "Review the code and suggest improvements"
    - file: ".buddy/review-prompt.md"
  model: "claude-sonnet-4-20250514"
  skills:
    - buddy-works/security-audit
    - "my-org/agent-skills/tree/main/linter"
  mcp_servers:
    github:
      type: "http"
      url: "https://api.githubcopilot.com/mcp/"
      headers:
        Authorization: "Bearer $GITHUB_TOKEN"
    filesystem:
      type: "stdio"
      command: "npx"
      args:
        - "-y"
        - "@modelcontextprotocol/server-filesystem"
        - "/buddy/my-repo"
      env:
        ROOT: "/buddy/my-repo"
  effort: "max"
  claude_args: "--verbose --max-turns 5"
  dangerously_skip_permissions: false
  working_directory: "/buddy/my-repo"

Minimal Claude Code action

yaml
- action: "Quick review"
  type: "CLAUDE_CODE"
  integration: "my_anthropic_integration"
  prompts:
    - "Review the code"

Claude Code with file-based prompts

yaml
- action: "Claude Code from prompt files"
  type: "CLAUDE_CODE"
  trigger_time: "ON_EVERY_EXECUTION"
  integration: "my_anthropic_integration"
  prompts:
    - file: ".buddy/code-review.md"
    - file: ".buddy/security-check.md"
  model: "claude-sonnet-4-20250514"

Claude Code with inline and file prompts

yaml
- action: "Claude Code mixed prompts"
  type: "CLAUDE_CODE"
  trigger_time: "ON_EVERY_EXECUTION"
  integration: "my_anthropic_integration"
  prompts:
    - "You are a code reviewer."
    - file: ".buddy/review-rules.md"
    - "Focus on security issues."
  model: "claude-sonnet-4-20250514"
  claude_args: "--max-turns 10"