Reading this with an AI? Start at /for-agents/ (orientation) or /llms.txt (index).

Skip to content
TODD BROWN
« All builds

Job Search Pipeline

parked

An agent-operated job-search pipeline for a family member. Orchestration platform built, piloted, then deleted.

Claude Code · Python

What it is

An agent-operated job-search pipeline for a family member: it discovers postings, scores each across 13 dimensions, routes it into one of five pursuit strategies, generates tailored materials, runs an ATS review, and tracks every job through an approval gate. Fifteen agent roles do the work; a conversational agent is the interface, not a dashboard.

It's parked. Last recorded activity: 11 May 2026. It did its job — applications went out, interviews happened.

Why it's on this site: parked, not abandoned — the person it was built for isn't job-searching right now, and the tool sits ready if that changes.

Why it exists

Early in the build I piloted a third-party orchestration platform: a server, an embedded database, an org chart, issue-based task routing. One job went through it successfully. Then I deleted every trace of it. The same fifteen roles, the same approval gates and scoring logic, run today on two Python packages and plain files on disk. It wasn't removed for failing; it didn't cut any decision I wasn't already writing myself.

How it's built

Status is derived from artifacts on disk, never an agent's self-report: a job is tracked if its file exists, a step is complete if its output exists and its upstream gate has cleared. 25,605 lines of Python across 56 files, two dependencies, no database, no server.

What broke

That governance rule caught its own violation: an initialization function marked a step "done" because its output file existed, without checking whether its approval gate had cleared. 26 of 31 jobs were marked complete before approval happened, fixed with a dependency-enforcement pass and 6 new tests. A related gap: the audit log held 1 entry for 31 jobs, since initialization never logged its own actions; it grew to 208 once fixed.

Separately, a values-conflict near miss added a governance stage: an application cleared scoring, materials and approval before I caught the employer's business conflicted with the candidate's values. No stage had asked; the fix added a review stage gating approval.

Receipts

Figure What Source
25,605 lines of Python source across 56 files file count
445 test functions today (147 passed, prior snapshot) grep / sprint validation report
26 of 31 jobs incorrectly marked complete before the dependency fix sprint validation report
1 → 208 audit log entries, before and after the fix sprint validation report
2 Python dependencies; no database, no server project dependency file

Counts are recorded in the project's own files, not re-run for this page. More on the receipts page.