jasonrve/agentic-runner

Call an OpenAI-compatible LLM with injected context, render a concise PR comment, and optionally upsert it as a PR comment.

Trust Signals

name	description	required	default
prompt	Primary instruction for the LLM workflow.	yes	—
context	Optional injected context to include with the prompt.	no	""
llm_base_url	OpenAI-compatible LLM base URL. The action normalizes root and `/v1` endpoints to chat completions.	no	""
llm_api_key	OpenAI-compatible LLM API key.	no	""
github_token	GitHub token used to upsert PR comments.	no	${{ github.token }}
model	Model name to use.	no	openai/gpt-4o-mini
pr_number	Pull request number to comment on. Defaults to the current PR context when available.	no	""
post_comment	Whether to upsert a PR comment.	no	true
fail_on_findings	Fail the step when the generated response signals attention or blocking issues.	no	false
comment_marker	Stable HTML marker used to update one comment instead of creating duplicates.	no	<!-- agentic-runner -->
dry_run	Skip posting a PR comment and only emit outputs.	no	false
mock_response_file	Optional path to a file containing a raw JSON response payload for tests and local validation.	no	""
context_mode	Context strategy to use. diff, full, hybrid, or agentic.	no	diff
focus_paths	Comma or newline separated list of files to focus the scan on.	no	""
extra_context_paths	Comma or newline separated list of file paths to always include.	no	""
max_file_chars	Maximum characters to load per file.	no	12000
max_follow_up_rounds	Maximum number of additional file-request rounds in agentic mode.	no	1

name	description
signal	Signal from the model response.
answer	Direct answer text rendered into the comment.
comment_body	Rendered Markdown comment body.
comment_url	URL of the upserted PR comment, if any.