| name | codex-runner-delegation |
| description | Experimental transport for requesting Deterministic Campaign Runner dry-run orchestration through the codex_runner_campaign_dry_run MCP tool. Use only to produce Runner-owned dry-run campaign artifacts for review; it does not authorize implementation, autonomous execution, Pi Loop, or Guardian operations. |
Codex Runner Campaign Dry Run
Experimental status — prototype. Do not advertise or release.
Use codex_runner_campaign_dry_run only for Deterministic Campaign Runner
dry-run orchestration. The adapter validates arguments and invokes the public
Runner CLI with --dry-run; it does not invoke providers directly.
Procedure
- Resolve the exact repository root.
- Obtain the required audit prompt, audit schema, compiler prompt, and campaign
set schema paths.
- Pass only documented tool arguments. Do not attempt to pass
--execute or
generic implementation, inspection, or autonomous modes.
- Call
codex_runner_campaign_dry_run.
- Report the process outcome as transport evidence only. Review the referenced
Runner-generated artifacts for campaign evidence.
The allowlist consists of the required repo_root, audit_prompt_file,
audit_schema_file, compiler_prompt_file, and campaign_set_schema_file;
plus optional Runner-supported provider, task_result_schema_file, passes,
base_ref, branch/discovery/verification/debug controls, provider model
selectors, codex_config, and claude_settings. The MCP tool schema is the
authoritative argument spelling and type reference.
The tool grants no implementation or mutation authority. Runner dry-run may
produce its normal campaign, state, audit, and receipt artifacts; those
Runner-generated artifacts remain the evidence surface. The MCP adapter creates
no authoritative receipt. A zero process exit is not approval, implementation
success, test success, or execution authorization.