hooman-method

This is a personal operating method for using LLMs on work that matters, pauses for weeks, and cannot safely live only in chat history.

The Hooman method is built for a single principal whose attention is split across projects, operational responsibilities, family, civic work, and long gaps between sessions. Its main problem is re-entry: when I come back to a project, I need to know what was decided, what is next, what must not be touched, and what can be delegated without losing judgment.

The method separates three roles:

• Human — owns priorities, constraints, and final decisions.
• Chat — frames the work, watches the long arc, drafts briefs, and reviews output with the human.
• Executor — investigates or mutates files under a narrow brief, with persistent artifacts as the source of truth.

It is not meant to turn every task into a process. It is an escalation protocol: below the threshold, work informally; above it, create just enough durable structure that the work survives interruption, delegation, and review.

This repository is the global tier: the versioned, taggable home for the method's operating documents and assistant skill. Downstream consumers project from a pinned tag; they never fork these docs.

Who this is for

This may look unusually formal because the circumstances are unusually shaped: one person, many unrelated obligations, intermittent project attention, high context-switch cost, and increasing delegation to AI tools. Some parts are broadly recognizable to anyone doing serious LLM-assisted work; the full apparatus is intentionally personal and should be scaled down unless the same pressures are present.

The three-tier model

• Global (this repo) — the method itself: the principal's self-contract, the assistant ground-rules skill, design stances, and coding-convention indexes.
• Family commons — a shared layer for a family of related workspaces. Reference implementation: heddle-workspace; second field sample: dev-commons.
• Workspace — an individual project: a references/ projection of the global docs (a downstream copy, never forked) plus its own anchors, briefs, and findings.

Contents

• hooman-contract.md — the principal's contract with themselves (assistants read narrow sections only for contract/escalation watch).
• hooman-notes.md — companion: rationale, provenance, references. Never needed to operate.
• skills/hooman-assistant/ — the assistant ground rules as an Agent Skill: SKILL.md (core) plus references/contracts.md (output schemas) and references/workspace.md (artifact rules).
• design-stance/ — design stances; universal/ applies everywhere, with per-language overlays alongside (swift/).
• coding-conventions/ — coding-guideline indexes, per language (swift/).
• tools/check-docs.py — local doc-discipline check: budgets, unique rule ids, and contract skeletons.

Using it downstream (projection v0)

Consumers copy the docs they need, pinned to a release tag, and record the source in a references/UPSTREAM marker:

• source repo
• pinned tag
• refresh date
• refresh instruction

Refresh means re-copying at a newer tag. No submodules, no symlinks; the copy rides the consuming repo's own git. The assistant skill installs by copying skills/hooman-assistant/ into the consumer's skill-discovery directory.

A workspace is governed only after a root AGENTS.md or <project>-context.md says so explicitly. references/UPSTREAM confirms the projection, but folder shape alone is not a marker.

Checks

Run python3 tools/check-docs.py before tagging a doc-set release.

Versioning & license

Doc-set releases are tagged; latest is v0.8.1, and v0.8.0 was the first repository release, with provenance running back to v0.4 in history. Docs are licensed CC BY 4.0 (see LICENSE). Graduated from the method's founding gist, which remains in place as a redirect.