Open-source work

ERP Access's public repositories on GitHub, what each one does, and what it demonstrates about the firm's engineering.

Updated 14 November 2026. Read this page as markdown, or get the whole file in one document.

ERP Access publishes its methods so a buyer can inspect how the firm works before engaging it. The public repositories live at github.com/chrbailey, maintained by the firm's principal. Client work is kept in private repositories under the firm's organization and is not published.

Repository What it is License What it demonstrates
SaaS Rebuild A protocol and tooling for auditing a SaaS or ERP tenant you administer: evidence-cited KEEP, SIMPLIFY, DROP or DEFER verdicts, an export plan and a preservation manifest. Ships as a Claude Code plugin with a browser workspace. MIT Evidence discipline: its schemas reject a verdict with no cited evidence, and its validator won't let a quiet 30-day log prove a year-end feature unused. Includes a NetSuite audit prompt for Claude in Chrome.
SAP Transaction Forensics Cross-system ERP transaction forensics for SAP, Salesforce and NetSuite: 18 analysis and data MCP tools plus 9 governance tools, and a discovery layer that proposes new patterns and validates them against evidence. Public synthetic demo. See repository ERP domain depth: segregation-of-duties checks, conformance deviations, journal anomalies, and field cases with actual figures.
PromptSpeak MCP server Pre-execution governance for AI agents: validates tool calls before they run, detects drift, and holds high-risk actions for human approval. Published on npm. MIT Agent safety engineering: 56 tools and 829 passing tests per its README.
Ralph Dispatch A bounded, critic-gated research operator for a single host, Python standard library only. Models get no tools, shell, live web, credentials or publishing authority; work becomes publishable only after a schema-valid critic verdict. MIT Fail-closed design: strict schemas, persistent budgets, append-only transition log, and a published threat model.
deeptrend A structured AI-trend feed built for autonomous agents, published as JSON, RSS and llms.txt. See repository Building for machine readers. Paused since April 2026, as its README states.

Why publish

  • Buyers can verify before they buy. The method behind a SaaS teardown or an ERP forensic review is open to inspection.
  • The same rule shows up in the code. Refusal when evidence is missing, critic gates before anything is published, and human approval for high-risk actions appear in every repository above.
  • Honest status. Paused or experimental projects say so in their own documentation.