hamzaabduljabbar

Autoconst Claude Change Order Audit — Security skill for Claude Code

Security community

Audit a contractor's change-order proposal against the contract's rate schedule and the drawings — in seconds instead of 1–3 hours per CO.

How to install Autoconst Claude Change Order Audit

This entry records only its repository, not the path inside it, so there is no exact command to give. Open hamzaabduljabbar/autoconst-claude-change-order-audit and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Autoconst Claude Change Order Audit does

**Audit a contractor's change-order proposal against the contract's rate schedule and the drawings — in seconds instead of 1–3 hours per CO.**

Alternatives in Security

  • /web3 Audit — Smart contract security audit using the 10-bug-class methodology 927 ★
  • MCP Security Checklist (SlowMist) — Comprehensive checklist: input validation, rate limiting, RBAC, credential management, container hardening 819 ★
  • ThinkWatch — Self-hosted AI API and MCP gateway for organizations: SSO and RBAC, per-user identity for MCP tool calls, PII 817 ★

README

Change-Order Pricing Audit

**Audit a contractor's change-order proposal against the contract's rate schedule and the drawings — in seconds instead of 1–3 hours per CO.**

Owner / PM-side tool. A contractor sends a change order PDF; a PM normally opens the CO, opens the contract's rate schedule, opens the drawing, and cross-checks line-by-line in Excel — hunting for inflated markups, stacked O&P, padded labor rates, and quantities that don't tie to the drawing. This tool does that pass deterministically and returns dollar figures with page anchors.

It ships as a [Claude Code](https://docs.claude.com/en/docs/claude-code) skill, but the engine is a plain Python CLI you can run on its own.


Why

Every finding is a **$ overcharge tied to a page** — not an opinion. Judgment calls ("is 42 hours reasonable for this scope?") are surfaced for a human, never ruled on. For this persona a wrong "this is padded" costs more trust than staying quiet.

Three problems every honest CO auditor has to handle, baked in by design:

Problem How the tool handles it
Column reflow. pdftotext prints the $ext column out-of-row from the description; naïve tie-out fabricates ~10 false discrepancies on a clean doc. Parsers read only rate and quantity from the line and recompute — the printed $ext column is never trusted.
Silent invalidity. "0 findings" can mean the CO is clean OR that no reference doc was supplied. Coverage-aware summary: 0 findings — but only N of 4 checks had the data to run, with a per-check [checked] / [NOT CHECKED] + reason breakdown.
Scanned pages. A number read from an image must never carry the same certainty as vector text. Scan pages are rendered to PNG, Claude transcribes them into a sidecar .txt, the same checks run — but every finding on a scan-derived page is stamped [SCAN - verify against source page N].

Install

Requires **Python 3.10+**, **Poppler** (`pdftotext`, `pdftoppm`), and one