Built

Things we built, mostly in the open. They’re here as proof, not products. Every one is free to use.

Harnessie

Verify agent-written pull requests claim by claim. Sandboxed checks, an independent verifier, and a fail-closed verdict when a claim can't be proven.

Siteline

Free scanner that grades whether AI agents can find, read, and act on a website. Built under PAICE.work, born as a Snap Synapse audit.

skill-a11y-audit

Portable accessibility-audit skill for Claude Code and Codex. WCAG checks an agent can run and report on.

AI Tool Watch

Compare AI platforms feature by feature across categories and pricing tiers. The reference we wanted when clients asked "which one?"

HardGuard25

Human-friendly unique IDs from a 25-character alphabet with no look-alike characters.

PrompterKit

Manage Elgato Prompter scripts from the CLI or a local web GUI. Built for the show, shared for anyone who reads from a prompter.

Methods

More under github.com/snapsynapse. The PAICE.work PBC standards and measurement tools live at paice.foundation.

Project Write-ups

Canonical articles about these tools. Also published on dev.to and Coderlegion.

• Verify Agent-Written Pull Requests Claim by Claim with Harnessie Jul 2026
  Agent-written PRs claim their own verification. harnessie verify re-tests the claims: sandboxed checks, then a fresh-context model, fail-closed exit code.
• SNAP Applicability Scale Jul 2026
  A four-axis signature for labeling how much decision weight a study deserves in your context. Score any study in 60 seconds or less.
• I lost all my Prompter scripts on a Friday. By Sunday, I'd built PrompterKit. Apr 2026
  Camera Hub stores your Elgato Prompter scripts in opaque JSON with no export, no backup, and no way to get your words back out. I fixed that.
• One AGENTS.md, every tool: how I stopped copy-pasting CLAUDE.md everywhere Apr 2026
  Every AI coding assistant wants its own instruction file at a different path. I forked a tiny Go CLI that fixes it with symlinks — and added detect, scan, auto-hooks, and global-config support.
• I built an open spec because every bad 429 was costing me twice Mar 2026
  Your API's rate limit errors are generating more traffic than they prevent. Graceful Boundaries is an open spec that makes every refusal self-explanatory.
• EveryAILaw: A Structured Database of AI Compliance Obligations Across Jurisdictions Mar 2026
  Open-source reference tracking AI regulation obligations across 12 regulations and 8 authorities, with a JSON API and MCP server for programmatic access.
• Knowledge as Code: A Pattern for Knowledge Bases That Verify Themselves Mar 2026
  Apply software engineering practices to knowledge management: plain text, Git-native, zero-dependency, self-healing through AI verification cascades, multi-output from a single source.
• Your site works fine in a browser. AI agents can't use it. Mar 2026
  Most websites silently break when AI agents try to use them. I built a free scanner that shows you exactly where and why.
• HardGuard25: A 25-character alphabet for human-readable unique IDs Mar 2026
  An open ID alphabet that removes 11 confusable characters instead of Crockford Base32's 4. JS, Python, and Go. MIT licensed.
• Teach Your AI Coding Agent to Run Accessibility Audits Mar 2026
  A reusable agent skill for WCAG 2.1 AA audits using axe-core and Lighthouse. Works with Claude Code, Codex, and other AI agents.
• I Wrote a Script to Fix Audible's Unreadable PDF Filenames Feb 2026
  A Python CLI that renames Audible PDF companions from cryptic codes like bk_adbl_022796.pdf to their actual book titles. Three-tier extraction: metadata, text, OCR.
• Your AI Agent Skills Have a Version Control Problem Feb 2026
  A meta-skill for tracking agent skill versions across stateless sessions, surfaces, and platforms. Works with Claude, Codex, Gemini CLI, and any agentskills.io-compatible system.
• I built an AI tool where nobody — not me, not the cloud provider — can read the conversation Feb 2026
  Most 'privacy-preserving' AI tools ask you to trust a privacy policy. I wanted proof. Here's a TEE-protected model cascade with on-chain attestation, built on NEAR.
• I spent years trying to get AI agents to collaborate. Then Opus 4.6 and Codex 5.3 wrote the rules themselves. Feb 2026
  For years I couldn't get AI agents to collaborate without a boss. Then two frontier models wrote their own protocol — using a proto-version of it — across eleven real sessions.
• AI Tool Watch: Stop Googling 'Can I Use That AI Feature on the Free Plan' Jan 2026
  A hypeless open-source reference for AI capabilities, pricing tiers, platform support, and availability across ChatGPT, Claude, Gemini, Copilot, Grok, and more.
• SNAP Method Jan 2026
  A four-part structure for turning messy situations into clear, testable next actions.
• SNAP Panic Scale Dec 2025
  Quick check for change pressure and decision drag.
• How Do You Measure Whether Someone Is Actually Good at Working With AI? Oct 2025
  The engineering problem behind measuring human-AI collaboration quality: why quizzes don't work, what behavioral assessment looks like, and how we built an adaptive simulator to measure it.