OpenProse Workflows

Automated multi-agent workflows for Netclaw, written in OpenProse.

Available Workflows

security-audit.prose -- Security Audit

Runs a comprehensive security audit against the Netclaw codebase using 4 specialized agents and 8 parallel scan categories.

Frameworks: OWASP Top 10 for LLM Applications (2025), Netclaw SEC-001 through SEC-009 (PRD-002)

Usage:

prose run .prose/security-audit.prose

To focus on a specific area:

prose run .prose/security-audit.prose --input focus_area=prompt-injection

Focus areas: all (default), prompt-injection, tool-grants, acl, shell-path, credentials, self-modification, exposure, webhooks

What it does:

1. Phase 1 -- Scan: 8 parallel scans covering prompt injection resistance, tool grant permissiveness, ACL policy completeness, shell/path policy coverage, credential exposure, self-modification boundaries, exposure mode safety, and webhook/persistence safety.
2. Phase 2 -- Critique: Validates findings against Netclaw's 13 built-in defense layers to filter false positives and adjust severities.
3. Phase 3 -- Fix: Produces before/after code proposals for each confirmed vulnerability.
4. Phase 4 -- Report: Generates a structured report with OWASP LLM Top 10 coverage matrix, SEC control compliance, defense-in-depth assessment, and maturity model evaluation.

Directory Structure

• *.prose -- Workflow source files (checked in)
• runs/ -- Runtime output from workflow executions (git-ignored)