u2i/Laocoön — Supply Chain Guard

Diffs changed dependency lockfiles (Hex/npm/PyPI) against a soaked baseline and uses an LLM to flag freshly-injected supply chain trojans.

Trust Signals

name	description	required	default
gemini-api-key	Google Gemini API key (for google: models). Maps to GEMINI_API_KEY.	no	""
anthropic-api-key	Anthropic API key (for anthropic: models). Maps to ANTHROPIC_API_KEY.	no	""
openai-api-key	OpenAI API key (for openai: models). Maps to OPENAI_API_KEY.	no	""
llm-api-key	API key for an OpenAI-compatible endpoint (compatible: models). Maps to LLM_API_KEY.	no	""
llm-base-url	Base URL for an OpenAI-compatible endpoint (OpenRouter/Together/Groq/Gateway/Ollama). Required for compatible: models.	no	""
github-token	GitHub token used to read the PR diff and post a comment.	no	${{ github.token }}
triage-model	Cheap model run on every PR, as 'provider:model'. Bare model defaults to google:.	no	google:gemini-3.1-flash-lite
deep-model	Stronger model used only on escalation, as 'provider:model'.	no	google:gemini-3.5-flash
soak-days	A release older than this many days is treated as a presumed-clean 'soaked' baseline to diff against.	no	60
ecosystems	Comma-separated ecosystem ids to analyze (e.g. 'hex,npm,pypi'). Empty = auto-detect all.	no	""
lockfile	Restrict analysis to a single lockfile path (overrides auto-detection).	no	""
base-ref	Git ref to diff against (the 'before' state). Defaults to the PR base.	no	""
fail-on	Minimum risk that fails the check: critical \| high \| medium \| low \| none.	no	high
comment	Whether to post / update a PR comment with the findings.	no	true
max-diff-bytes	Overall byte cap on the artifact diff sent to the LLM per ecosystem.	no	60000

name	description
risk-level	Highest risk level found (none\|low\|medium\|high\|critical, or 'skipped').
findings-json	Structured findings as a JSON string.