I build automation pipelines on top of Claude Code and AI agents, the kind of work most teams still pay people to do by hand. The two pieces I lean on hardest are a recipe-driven browser automation engine that submits a product to directory after directory without me sitting there, and a parallel subagent pattern that runs multiple AI auditors over long documents at once.I built both of these for my own products first, because I needed them. That is the point of showing them here. They are real systems that run, not slides. The same engine that submits my tools to directories can submit a client's tool, and the same review pattern that audits my manuscripts can audit any large body of work for consistency.
What I Built
The centerpiece is a self-contained directory submission engine. The job it solves is boring and slow when you do it by hand: take one product, and get it listed across a long list of software directories, each with its own form, its own dropdowns, its own captcha, and its own idea of what a successful submission looks like.The engine works off reusable JSON recipes. A recipe targets a form pattern, not a single site, so one recipe covers a whole family of directories that share the same template. I have 15 recipes that together cover most of the directory ecosystem, from WordPress Contact Form 7 sites to modern React and Webflow forms to the older PHP directory templates that still gate submissions behind image and math captchas.Alongside that, I run a parallel subagent pattern for quality work. Instead of one agent reading a long document top to bottom, I dispatch several AI auditors at once, each checking a different dimension, then loop the whole pass until everything comes back clean. I use it to audit long manuscripts chapter by chapter against an established canon, which is the kind of cross-referencing that is easy to get wrong when a human does it tired at the end of a project.
How It Works
A few real tools do the heavy lifting:
Claude Code orchestrates the work. It writes and adjusts recipes, runs the engine, reads the result logs, and iterates when a site changes.
Playwright drives a real Chromium browser through each form. The recipe action types include navigate, fill (with optional sequential typing to look natural), select, click, upload, and a four-strategy click fallback for the awkward sr-only inputs and role-based checkbox buttons that modern forms hide behind.
Tesseract OCR reads image captchas. A solve_captcha action runs multi-PSM OCR to crack the digit captchas on the older directory templates, and a solve_math action regex-parses the "X plus Y equals" style questions straight out of the form row.
Reusable JSON recipes are the unit of reuse. Each one is a small script of actions plus a success check, so adding a new directory family usually means copying a similar recipe and adjusting selectors rather than writing anything from scratch.
Success detection is explicit. A check_success step verifies the real confirmation, by text marker, URL change, URL pattern, or in one case a JavaScript alert dialog registered before the click. This matters because an earlier headless version of the engine claimed wins that were actually silent rejections. The recipe-driven version only counts a submission when the site itself confirms it.
The CLAUDE.md instruction hierarchy keeps the whole thing honest. Conventions, structure, and hard rules live in committed instruction files that Claude Code reads first, so the agent stays inside the lines instead of improvising.
The parallel subagent QA runs on the same Claude Code foundation. For a review pass I spin up four auditors at once, each with a single focus, and keep looping passes until none of them flag anything. It is the difference between one tired reader and a small focused team that never gets bored.
Results
110+ directories targeted from a single committed source list, spanning a wide range of domain authority.
15 reusable recipes that generalize across form patterns, so the engine scales to new directories without new code most of the time.
Confirmed real submissions, not guesses. In one live session on one product, 73 sites were attempted and 20 came back as confirmed real submissions, including a live public listing that anyone can load in a browser today.
Proven across more than one product. Running the same 15 recipes unattended against a second product confirmed 11 of the 15 working across repeated runs, again including a live public listing. That second run is also where new capabilities got added: the math-captcha solver, the alert-dialog success check, and the multi-strategy click fallback.
60+ chapters audited by parallel agents. On a single long manuscript, parallel AI auditors checked the whole book against an established canon and surfaced consistency fixes a single pass would have missed.
Why This Is the Hire Proof
This is the lane my services actually sell, and these are the receipts. The directory engine and the parallel auditor pattern both started as tools I needed for my own work, which is why I trust them: I run them on things I care about before I would ever point them at a client's. The work is portable. The recipes, the OCR captcha handling, the explicit success detection, and the parallel review loop all transfer directly to whatever repetitive, high-volume, or large-document problem a client is currently paying a human to grind through.
