8000
Skip to content
View dallascrilley's full-sized avatar

Block or report dallascrilley

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
dallascrilley/README.md

Dallas Crilley

10 years owning lead-to-cash systems; billing automation in production ~19 months; now shipping evaluated agent systems with human gates. Senior applied-AI and business-systems engineer in Python and TypeScript — evals, tests, and CI, not demos.

Site: dallascrilley.com · Hiring? dallascrilley.com/hire · Proof ledger: dallascrilley.com/proof

Production extracts

  • Shipwright: an agent that takes an approved GitHub issue and returns a tested, reviewable pull request — it opens the PR only after the target repository's own test command exits zero in a disposable sandbox
  • Vouch: human review as an API — agents submit work, real reviewers return a consensus verdict; 173 tests plus offline e2e harnesses
  • Holdfast: append-only decision ledger with a human publish gate, enforced inside Postgres by triggers, a hash chain, and unique indexes; ships adversarial attack tests against its own gate

The full catalog — with sources and caveats for every headline metric — lives on the proof ledger.

Catalog provenance: much of this catalog was published to public GitHub in one recent pass while assembling the proof ledger — a repo's publication date is not its development timeline. Where the scaffold-vs-hand-written split matters, the repo carries receipts (e.g. Winnow's docs/receipts.md).

Agent commits

Agents commit under this login; I own the merge gate. The repository's own verification command runs against every change, CI blocks the merge until it passes, and a human makes each irreversible call. The three repos above are that gate as working code: Shipwright won't open a PR until the tests pass, Vouch routes agent output to human reviewers for a consensus verdict, and Holdfast enforces the publish gate inside Postgres. Open any repo's history and you can see the split: agent commits carry the implementation; my commits set direction, resolve review findings, and cut releases.

Writing

Pinned Loading

  1. shipwright shipwright Public

    An agent that turns approved GitHub issues into tested, reviewable pull requests.

    TypeScript

  2. vouch vouch Public

    Human review as an API: agents submit work, real reviewers return a consensus verdict.

    TypeScript

  3. holdfast holdfast Public

    Holdfast — an append-only decision ledger with a human approval gate: AI can propose, only a person can publish

    TypeScript

0