salilborkar/LLM Cost Lint

Shift-left LLM cost linting. Scans PRs for Bedrock and Azure OpenAI calls and posts cost estimates as PR comments.

Trust Signals

name	description	required	default
path	Directory or file to scan. Scanned recursively for .py files. Defaults to the repository root.	no	.
monthly-calls	Estimated number of times each detected LLM call will execute per month in production. Used to project monthly cost from a per-call estimate.	no	30000
default-input-tokens	Input token count assumed per call when the value cannot be statically inferred from source code. Override with a value representative of your typical prompt size.	no	1000
default-output-tokens	Output token count assumed per call when max_tokens is not explicitly set in the call site.	no	500
cost-threshold	Monthly cost ceiling in USD. When the projected monthly total exceeds this value a warning is added to the report. Set fail-on-threshold to true to also block the merge. Set to 0 to disable the threshold check.	no	100
fail-on-threshold	Set to 'true' to exit with code 1 when the projected monthly cost exceeds cost-threshold, blocking merge if this check is required by branch protection rules.	no	false
post-pr-comment	Post the cost report as a comment on the pull request. Requires pull-requests: write permission in the calling workflow.	no	true
github-token	GitHub token used to post PR comments. Defaults to the workflow's automatic GITHUB_TOKEN.	no	${{ github.token }}

name	description
cost-report	The markdown cost report generated by LLM Cost Lint.