← All systems
06 / 06Agent workflows · Human accountability

Content operating system

A content pipeline where deterministic code owns state, models own judgment, and people retain approval.

OpenClawObsidianPythonHuman in the loop
Role
Builder & operator · Personal stack
Setting
Personal infrastructure
Period
2026
System architecture / simplified
  1. 01CaptureObsidian + Telegram
  2. 02Controllerstate + retries
  3. 03Review gateversion-bound
  4. 04Learning2h / 24h / 7d
Material draft change → approval resets

Good content needs operational consistency and actual judgment. Manual workflows fail on consistency; unbounded automation turns into a feed of plausible words nobody should have published.

The obvious automation was not the whole problem.

The system needed to capture ideas from anywhere, preserve human edits, recover cleanly from failed scheduled jobs, and make it structurally impossible for a changed draft to pass on the strength of an old approval.

The reliable architecture splits work by its nature: deterministic code for state, language models for judgment, and a human for accountability—enforced by the system, not a reminder.

Build the operating loop, not only the intelligent step.

  1. 01

    Made Obsidian the human source of truth and used a deterministic controller for state, reconciliation, logs, and retries.

  2. 02

    Used scheduled agent turns for research, drafting, and critique rather than for irreversible actions.

  3. 03

    Bound approval to the exact content version so material edits reset review automatically.

  4. 04

    Recorded unavailable analytics as unavailable, never zero, so the learning loop does not optimize on fiction.

04 / Outcome

A daily operating system that captures, researches, drafts, critiques, and measures content—while asking for attention only where judgment and accountability are genuinely required.

2parallel pipelines
8+staged workflow states
100%publishes human-approved

What stayed after shipping.

The goal is not maximum automation. It is precise allocation: determinism for state, models for judgment, and human approval where the consequences should remain owned by a person.