jbovet/oxidized-agentic-audit

Security scanning for AI skills and agents — prompt injection, bash patterns, frontmatter and agentic security checks

Trust Signals

workflow.ymlSHA-pinned

uses: jbovet/oxidized-agentic-audit@39d0f23233d8225bf31d103b8542ed6d88e83291 # v0.6.0

tags can be moved; commit SHAs can't. why a SHA?

name	description	required	default
path	Path to a single skill/agent directory or a collection directory containing multiple subdirectories.	no	.
skills-path	Deprecated: use `path` instead. Path to a single skill/agent directory or a collection directory.	no	""
type	What to scan. One of "skill" (looks for SKILL.md) or "agent" (looks for AGENT.md).	no	skill
version	Version of oxidized-agentic-audit to download (e.g. "v0.5.0"). Use "latest" to always fetch the newest release.	no	latest
strict	Treat warnings as errors. Exit code 1 on any warning.	no	false
fail-on-warnings	Fail the action step when warnings are present, even without errors.	no	false
format	Output format for the scan report. One of pretty, json, sarif.	no	sarif
sarif-output	File path where the SARIF report will be written.	no	oxidized-agentic-audit-report.sarif
config	Path to a custom oxidized-agentic-audit.toml configuration file.	no	""
min-score	Minimum security score (0–100) required for the scan to pass. If any scanned skill scores below this threshold the action exits with code 1. Leave empty to disable the threshold gate.	no	""
skip-download	Skip binary download and use an oxidized-agentic-audit binary already present in PATH. Intended for CI environments that build the binary locally before invoking this action.	no	false

name	description
sarif-file	Absolute path to the generated SARIF report file.
errors-count	Number of error-severity findings.
warnings-count	Number of warning-severity findings.