The work

Constrain what an agent can do, and prove what it did.

That is one offer with four proofs rather than four services. Each project below is the same idea at a different layer. All four ship, all four are open source, and each has its own page here with its own documentation, so you can decide whether an hour on a call is worth it before you spend the hour.

The four

umbra

You gave an agent a shell. Now name every command it can run. umbra validates argv before execve, checks a scope token per verb, routes egress through a per-invocation proxy, appends every call to an audit log, and publishes an exit-code taxonomy that separates a policy refusal from a tool failure. MIT, on brew and scoop, Linux, macOS, and Windows.

mcp-beaver

You need an MCP server for one API. Now write a Go package, a handler and a schema per tool, and a Dockerfile. mcp-beaver renders one guardfile into a running MCP server, an HTTP tool API, and a helm release. One grant becomes one tool, and an operation nobody declared has no tool and no endpoint at all rather than a guarded one. MIT, preview.

housecast

You changed what a role is allowed to do. The evaluation that checked it did not notice. housecast declares roles, personalities, and boundaries in one YAML roster, validates it, emits an immutable bundle, and runs the behavior evaluations against exactly that bundle. The graded artifact and the shipped artifact are the same thing. MIT, preview, Python.

agent-compose

You gave your agents different jobs. They all load the same prompt. agent-compose materializes a named specialist with its own charter, its own set of things it hands off instead of attempts, and its own tool inventory, as a directory of plain files you can diff before anything runs. The same pack runs unchanged on Claude Code, Codex, Goose, and OpenCode. MIT, on brew and scoop.

What you can check before we talk

All four are MIT licensed and the source is public. Read umbra's guardfile page or housecast's roster page and you will know what I build and how I write about it before we have spoken. That is the point of publishing them.

None of the four ships or calls a model of its own. They bound and record what an agent does, whatever is running underneath, so they carry no model-provenance question into a procurement review.

How an engagement is shaped

Contract shape

Every shape is open. Direct contracting through my own business, consultancy, embedded delivery, staffing-agency W-2, contract-to-hire, and agency-mediated C2C all work. Use whichever one your procurement finds easiest.

Rate

$150 an hour. One rate covers every shape above, agency arrangements included, so the paperwork you prefer does not change the price.

Availability

I plan at about 70 percent utilisation rather than trying to fill every hour, so there is room for the work to overrun without the next thing suffering for it.

Where I am

East Bay, California. Remote works, and I can be in a Bay Area room when being in the room is the thing that moves it.