Software systems and technology consulting, based in Franklin, North Carolina.
We build in three layers, and this org is where the bottom one lives.
- Open tooling at the base. The general-purpose pieces of our production systems get pulled out, cleaned up, and published here as standalone repositories — useful on their own, with no obligation to adopt anything else we make.
- An agent-run engineering practice in the middle. The work is done by AI agents operating under mechanical constraints: gates in the filesystem and in CI that an agent cannot talk its way past, and every change landing as a pull request a human reviews and merges.
- Products on top. Our own products are built on those two layers. They stay private, but the parts of them that are generally useful do not.
- Governance and hygiene — rigscore: a
configuration hygiene checker for AI development environments. Point it at a project and it
reads the files that decide what your agents are allowed to do — governance docs, MCP server
configs, container settings, permissions, skill files — then returns a score out of 100 and a
list of what to fix, ordered by how badly it would hurt. Runs entirely on your machine, no
network calls. MIT.
npx github:Back-Road-Creative/rigscore - Video tooling — keycut: lossless multi-range video cutting that snaps to keyframes instead of lying about it. redactcam: face and licence-plate redaction for video, with a second pass that proves it landed.
- GPS and telemetry — sun-track: solar position, golden-hour windows, and where a GPS track was at a given instant.
- Color and imaging — pycube-lut: reads and applies Adobe .cube 3D LUTs, HALD CLUTs and 1D tone curves to NumPy images.
- Agent tooling — claude-batch-runner: declarative parallel agent campaigns with schema-verified results and a cost table. agent-audit: audits the skills, instruction files and workflows your AI agents actually read.
A couple more areas are still being carved out of our production systems. Not published yet — so these are areas, not links. Each one moves up to the section above as it lands:
- SEO scoring — dependency-free keyword extraction, readability, and per-platform metadata rules.
- Static-site security — a hardened header and content-security-policy baseline, with a build step that fails when the policy did not make it into the output.
Nothing ships until it stands alone: its own tests, its own documentation, and no leftover references to the system it came out of. A short honest list beats a long one full of dead links.
Mechanical enforcement over behavioral rules. If the only thing stopping an agent from doing something destructive is a prompt asking it not to, that is not governance — that is a suggestion. Constraints belong somewhere the agent cannot edit them out of the way.
Write-ups of the practice, and each tool as it ships: headlessmode.com.
📍 Franklin, NC · 🔗 backroadcreative.com · 𝕏 @HeadlessMode