Activity reference

An activity (a "skill" on the command line) is one security engagement with its own brief and report shape. Pick one or more with --skill; several run at the same time against the same target.

Activity--skillIn one line
Security Architecture Reviewsecurity-architecture-reviewIs the design sound? Controls, boundaries, CCM v4.
Threat Modelingthreat-modelingWhat could go wrong? STRIDE, attack chains, residual risk.
SASTsastCode-level vulnerabilities with taint paths and fixes.
SCAsca-reachabilityRisky dependencies, whether they're reachable, and an upgrade plan.
Secret Scanningsecret-scanningExposed credentials in code, config, history and docs.
Prompt Reviewprompt-reviewLLM prompt, tool and agent-behaviour security.
Deployment Readinessdeployment-readinessProduction security and operational readiness.
Responsible AI Complianceresponsible-ai-complianceAI governance, transparency, fairness and oversight.
Third-Party Risk Assessmentthird-party-risk-assessmentA vendor's security documents against CCM v4.

Which ones should I run?

You want to…Run
Catch vulnerabilities before mergingsast, secret-scanning, sca-reachability
Sign off a design at a milestonesecurity-architecture-review, threat-modeling
Prepare for a production launchdeployment-readiness, plus the code checks above
Review a feature built on an LLMprompt-review, responsible-ai-compliance
Onboard a vendorthird-party-risk-assessment, by uploading their documents

Quick or standard depth

quick (default)Triage. The agent reports the 3 to 8 most important findings, favouring critical and high severity and the riskiest entry points. It may describe a category of problem once with one example instead of listing every instance.
standardA thorough review against the activity's whole checklist, examining the codebase in depth rather than sampling. Slower and more expensive.

Set a default with depth = "standard" under [defaults] in ~/.secfoo/config.toml.

What every report has in common

  • An overall risk rating line near the top.
  • A register table of every finding, most severe first.
  • A detail section per finding, headed ### [SEVERITY] ID: title, with a location and a concrete recommendation.
  • Coverage notes: what was reviewed, what couldn't be reached, and the agent's confidence.

secfoo counts findings and drives the dashboards from those fixed parts, not from prose. Reading a report walks through them.

Security Architecture Review

security-architecture-review

Answers: Is this design sound? A control-centric review of components, trust boundaries and topology against secure design principles.

What the agent checks:

  • Trust boundaries and data flow, with a Mermaid data-flow diagram.
  • Whether security controls are present, correctly placed and layered.
  • Design principles: least privilege enforced by topology, fail-safe defaults, blast-radius containment.
  • Operational and regulatory fit: recoverability, key lifecycle, maintainability.
  • Conformance with each of the 17 CSA Cloud Controls Matrix (CCM v4) domains.
  • Whether the documentation matches what the code actually does.

Report sections: Executive Summary, Scope, Data Flow Diagram, Trust Boundaries, Security Controls Matrix, Design Principles Assessment, Operational & Regulatory Fit, CCM Domain Conformance, Findings Register, Detailed Findings, Documentation vs. Implementation, and Design Verdict & Remediation Roadmap. The design verdict is Sound, Sound with conditions (conditions named) or Not sound.

Shows up in: the Security Architecture program dashboard (coverage, CCM heatmap, gate-decision mix, recurring root causes) and exceptions.

Give it your design docs with --confluence where you can: the review compares what the docs say against what the code does. That needs a Confluence MCP server; see Connecting agent CLIs (MCP).

secfoo run --skill security-architecture-review --agent claude \
  --confluence https://yourteam.atlassian.net/wiki/spaces/ENG/pages/123/Design

Threat Modeling

threat-modeling

Answers: What could go wrong, and who would make it go wrong? An adversary-centric enumeration of threats from assets and trust boundaries.

What the agent checks:

  • Assets, trust boundaries and data flow, with a data-flow diagram.
  • The assumptions the design relies on, each stated so it can be checked.
  • Adversary classes, and STRIDE threats against each asset class.
  • Privacy threats (LINDDUN) when personal data is in scope.
  • Attack chains that combine several weaknesses, and where an attack could be detected.
  • A disposition for every threat: Mitigated, Gap, Accepted or Transferred, with the test that proves a mitigation where one exists.

Report sections: Executive Summary, Scope & Assets, Trust Boundaries & Data Flow, Assumptions, Adversary Classes, STRIDE Analysis, Privacy Threats (LINDDUN), Threat Register, Detailed Threats, Attack Chains, Detection Opportunities, Residual Risk Statement, Mitigation Roadmap.

Shows up in: the Threat Modeling program dashboard (STRIDE coverage matrix, dispositions, recurring threat archetypes, models needing rework), threat acceptances and post-build misses.

secfoo run --skill threat-modeling --agent claude --depth standard

SAST — Static Code Analysis

sast

Answers: Which code-level vulnerabilities exist, and how does untrusted input reach them?

What the agent checks:

  • Injection, authentication and authorization flaws, unsafe deserialization, cryptographic misuse and similar code-level issues.
  • The taint path from where untrusted input enters to where it is used dangerously.
  • A CWE and OWASP mapping for each finding, and a CVSS v3.1 vector from which secfoo computes the base score.
  • A concrete fix for each finding, with an illustrative diff.

Report sections: Executive Summary, Scope, Findings Register, Detailed Findings, Taint Summary, Remediation Roadmap, Code-Fix Appendix.

Shows up in: the SAST dashboard and findings list, where each finding stays open until a later scan no longer finds it, and can be reviewed as confirmed, false positive or accepted risk. Exportable as SARIF.

secfoo run --skill sast --agent claude

SCA — Reachability & Upgrade Triage

sca-reachability

Answers: Which of my dependencies carry known risks, and does my code actually reach the vulnerable part?

What the agent checks:

  • An inventory of dependencies from your manifests and lockfiles, and version hygiene (pinning, stale versions).
  • For each risky package: Reachable, Conditionally reachable or Not reachable from your code.
  • A grouped upgrade plan, so one bump can fix several issues.
  • Supply-chain hygiene: lockfiles, sources, install scripts.

Report sections: Executive Summary, Dependency Inventory, Version Hygiene, Risk Register, Detailed Findings, Reachability Summary, Upgrade Plan, Supply-Chain Hygiene.

Shows up in: the SCA dashboard and its cross-project vulnerability register. secfoo checks each package and version against OSV.dev, so advisory IDs come from a real database. The agent is told never to invent a CVE ID.

secfoo run --skill sca-reachability --agent claude

Secret Scanning — Code, Confluence & Docs

secret-scanning

Answers: Are credentials exposed anywhere they shouldn't be?

What the agent checks:

  • API keys, tokens, passwords, private keys and connection strings in source, config, git history and documentation, including linked Confluence pages.
  • Whether each one looks live, is unclear, or is a placeholder.
  • A rotation-first remediation list: revoke and rotate before you clean up history.
  • How to stop it recurring (pre-commit hooks, secret managers).

Report sections: Executive Summary, Scope, Findings Register (with Source: code, config, history or docs, and Validity: Looks live, Unclear or Placeholder), Detailed Findings, Documentation & Confluence Coverage, Remediation Roadmap, Prevention.

Shows up in: the Secret Scanning dashboard. secfoo stores only the first 8 characters of any secret evidence, even if the report contains more.

secfoo run --skill secret-scanning --agent claude \
  --confluence https://yourteam.atlassian.net/wiki/spaces/OPS/pages/456/Runbook

Prompt Review

prompt-review

Answers: Can the LLM features in this codebase be manipulated, or do more than they should?

What the agent checks:

  • Every prompt surface: system prompts, templates, few-shot examples, tool and function definitions.
  • Prompt-injection resistance where untrusted input reaches instruction context.
  • Whether model-callable tools are scoped to the minimum they need.
  • Validation or human confirmation before side effects such as shell commands, SQL or payments.
  • Secrets or personal data leaking into prompts or logs, guardrail robustness, and limits on cost and call volume.

Report sections: Summary (with the overall risk rating), Findings Register, Findings, Coverage Notes.

Shows up in: the project page and the generic activity page under Analysis.

secfoo run --skill prompt-review --agent claude

Deployment Readiness

deployment-readiness

Answers: Is this ready to run in production, securely and reliably?

What the agent checks:

  • Security gates in CI/CD, and whether they block or only warn.
  • Secrets management, health checks and rollback, observability.
  • Infrastructure as code: least-privilege IAM, network segmentation, public exposure.
  • Rate limiting, backup and disaster recovery, dependency pinning and SBOMs, container hardening.

Report sections: Summary (with the overall risk rating), Findings Register, Findings, Coverage Notes.

Shows up in: the project page and the generic activity page under Analysis.

secfoo run --skill deployment-readiness --agent claude

Responsible AI Compliance

responsible-ai-compliance

Answers: Is this AI system governed, transparent and fair enough for what it does?

What the agent checks:

  • Model and data documentation, bias and fairness testing.
  • Human oversight for high-stakes decisions, and transparency to end users.
  • Data provenance and consent, alignment with any framework the project claims (such as the NIST AI RMF or EU AI Act risk tiers).
  • Audit trails, and a way for users to contest an AI-influenced decision.

Report sections: Summary (with the overall risk rating), Findings Register, Findings, Coverage Notes.

Shows up in: the Responsible AI page. The report's overall risk rating becomes the assessment's Responsible AI risk badge, which stays "provisional" until the assessment is marked Completed.

secfoo run --skill responsible-ai-compliance --agent claude --assessment 3

Third-Party Risk Assessment

third-party-risk-assessment

Answers: Does a vendor's own security documentation show they meet the controls we need?

What the agent checks:

  • Each uploaded vendor document: SOC 2 reports, ISO 27001 certificates, pen-test reports, questionnaires, policy excerpts.
  • Evidence for each of the 17 CCM v4 domains, rated Conformant, Partial, Non-conformant or Not Assessed (no evidence). Marketing claims don't count as evidence.
  • Exactly what each certification or audit claims (scope, date, standard version), and nothing broader.

Report sections: Executive Summary, Documents Reviewed, CCM Domain Conformance, Findings, Detailed Findings, and a Recommendation: Approve, Approve with conditions, Needs more documentation or Reject.

Shows up in: the Third-Party page.

This one reviews documents, not code. You normally don't run it by hand: uploading a PDF, DOCX or PPTX to a third-party assessment in the dashboard starts it. See Reviewing a vendor.

# Runs automatically when you upload vendor documents
# to a third-party assessment in the dashboard.
← CLI reference Reading a report →