msaad00/agent-bom Scan

Run agent-bom in CI/CD to scan AI agents, MCP servers, images, and IaC. Emit SARIF and map blast radius from packages to credentials and tools.

View on GitHub

Trust Signals

Scorecard Score: not yet scored
Maintenance Recency: Activelast commit Jul 11, 2026
License: Apache 2.0

Pinned Snippet

workflow.ymlSHA-pinned

uses: msaad00/agent-bom@93cdcaedf4478c642f3a30bc6753627e3300b2cb # v0.94.2

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

Inputs

name	description	required	default
scan-type	Scan mode. Default `scan` runs the full agent/MCP discovery pipeline. Focused modes: image, fs, iac, sbom, secrets, code, agents, and skills.	no	scan
scan-ref	Target reference for focused modes — image reference, filesystem path, SBOM file path, or IaC paths (comma-separated). Not required for the default `scan` mode.	no	""
severity-threshold	Fail if vulnerabilities at or above this severity (critical, high, medium, low). Default high; set to empty to disable the gate.	no	high
warn-on-severity	Warn (non-blocking) if vulnerabilities at or above this severity. Use with severity-threshold for two-tier gates.	no	""
policy	Path to policy file (JSON/YAML) for policy-as-code gates.	no	""
enrich	Enable NVD CVSS, EPSS, and CISA KEV enrichment.	no	false
format	Output format. Core: sarif (default), json, html, cyclonedx, spdx. Plain/no-color: plain (alias: text). Other: console, mermaid, svg, graph, graph-html, badge, prometheus.	no	sarif
output	Output file path for scan results. Default: agent-bom-results.sarif (for SARIF format).	no	""
upload-sarif	Upload SARIF results to GitHub Security tab.	no	false
auto-update-db	Auto-update vulnerability database if stale (>7 days). Disable for air-gapped or fully cached runs.	no	true
fail-on-kev	Fail if any CISA Known Exploited Vulnerability is found.	no	false
ai-enrich	Enable LLM-powered risk analysis (requires Ollama or API key).	no	false
ai-model	LLM model for AI enrichment (e.g. ollama/llama3.2, openai/gpt-4o-mini).	no	""
python-version	Python version to use.	no	3.11
agent-bom-version	agent-bom package version to install (for example 0.75.x). Empty = latest from PyPI.	no	""
install-from-source	Install agent-bom from local source (pyproject.toml) instead of PyPI. Use in CI dogfood to test unreleased code.	no	false
image	Docker image to scan (e.g. nginx:1.25).	no	""
config-dir	Path to MCP config directory to scan.	no	""
sbom	Path to existing SBOM file (CycloneDX or SPDX JSON) to ingest.	no	""
remediate	Generate remediation plan at this path (e.g. remediation.md).	no	""
badge	Generate shields.io badge JSON at this path (e.g. agent-bom-badge.json).	no	""
pr-comment	Post a findings summary comment on the pull request (requires GITHUB_TOKEN with pull-requests: write).	no	false
annotations	Emit GitHub Actions line-level annotations for SARIF and skills findings.	no	true
ignore-file	Path to ignore/allowlist file (default: .agent-bom-ignore.yaml). Suppress known false positives by CVE ID, package, or finding type.	no	""
iac	Path(s) to scan for IaC misconfigurations (Dockerfile, Kubernetes, Terraform, CloudFormation). Comma-separated.	no	""
ai-inventory	Path(s) to scan for AI component inventory (SDK imports, model references, deprecated models). Comma-separated.	no	""
os-packages	Enable OS-level package scanning (dpkg/rpm/apk).	no	false
baseline	Path to a baseline report JSON for delta scanning (only report new findings).	no	""
skill-only	Scan ONLY skill/instruction files (CLAUDE.md, .cursorrules, AGENTS.md); skip agent/package/CVE scanning.	no	false
fail-on-verdict	For scan-type=skills, fail if any scanned file reaches this trust verdict or worse (suspicious or malicious).	no	""
warn-on-verdict	For scan-type=skills, warn without failing if any scanned file reaches this content verdict or worse (suspicious or malicious).	no	""
fail-on-review-verdict	For scan-type=skills, fail if any scanned file reaches this handling verdict or worse (review, high_risk, blocked).	no	""
warn-on-review-verdict	For scan-type=skills, warn without failing if any scanned file reaches this handling verdict or worse (review, high_risk, blocked).	no	""
skills-ci	For scan-type=skills, enable the CI gate: exit non-zero on suspicious or malicious skills without authoring a policy file (equivalent to fail-on-verdict=suspicious). An explicit fail-on-verdict input takes precedence. Ignored for other scan types.	no	false
gpu-scan	Enable GPU infrastructure scanning (NVIDIA NIM/CUDA containers, DCGM).	no	false
exclude-unfixable	Exclude vulnerabilities without available fixes from SARIF output	no	false

Outputs

name	description
sarif-file	Path to generated SARIF file (if format=sarif).
exit-code	Scan process exit code.
scan-status	High-level scan status (clean, violations, or error).
vulnerability-count	Number of vulnerabilities found, or skill findings when scan-type=skills.
badge-file	Path to generated badge JSON (if badge input is set).
graph-export-path	Path to generated graph export when format is mermaid, svg, graph, or graph-html.
exposure-paths-count	Number of exposure, lateral, or attack paths parsed from JSON graph-capable outputs.
blast-radius-critical-count	Number of critical blast-radius or SARIF findings parsed from the result artifact.
posture-grade	Coarse CI posture grade derived from parsed critical/high/finding counts: A, C, D, F, or unknown.