prompts-gpt run-batch

Execute multiple prompt files in sequence using a manifest or explicit file list.

Usage

prompts-gpt run-batch [--manifest <path> | --prompt-files <a,b,c>] [--agent <name>] [--model <name>]

When to Use

Use run-batch when you have multiple independent prompt files and want to execute them back-to-back with the same provider settings. Each prompt runs as an isolated single-shot execution — no context is shared between prompts.

Options

Flag	Description
`--manifest <path>`	Prompt manifest JSON file. Overrides config defaults.
`--prompt-files <list>`	Comma-separated prompt file paths.
`--agent <name>`	Orchestration profile. Default from config or `router`.
`--model <name>`	Model override for all runs.
`--timeout <seconds>`	Timeout per prompt in seconds.
`--artifacts-dir <path>`	Override run artifacts directory.
`--json`	Print machine-readable JSON output.
`--cwd <path>`	Project directory.

Prompt Resolution Order

--prompt-files flag (highest priority)
--manifest flag
batchPromptFiles in .prompts-gpt/config.json
batchManifest in .prompts-gpt/config.json
Auto-detect .prompts-gpt/manifest.json

Examples

From manifest

prompts-gpt run-batch --manifest .prompts-gpt/manifest.json

Explicit files

prompts-gpt run-batch --prompt-files .prompts-gpt/review.md,.prompts-gpt/tests.md,.prompts-gpt/docs.md

With provider and model

prompts-gpt run-batch --prompt-files a.md,b.md --agent claude --model claude-sonnet-4-20250514

Output

Returns a RunBatchResult with results for each prompt:

{
  results: RunPromptResult[];
  totalDurationMs: number;
  succeeded: number;
  failed: number;
}

How It Differs from Sweep

Aspect	run-batch	sweep
Input	Multiple files	Single file
Iterations	One per file	N per file
Context	No sharing between runs	Previous iteration summary injected
Lock management	None	Sweep lock prevents concurrent runs
Self-evaluation	Not available	Available (scoring each iteration)