Operator Skill

Skill ID: operator
Version: 1.0.0
Domains: productivity, work, personal
Install: fetch this file and follow the "How to use" section

How to use

Read this file, then set up the user's Brains knowledge base using the Operator method below. Establish the brain-root layout (projects/, people/, instructions/, cross-cutting canonical/ and data/, dumps/, and a root open-loops.md index). For each project the user names, scaffold projects/<slug>/ with a readme.md (from the template), an empty open-loops.md rollup, and state/next-actions.md — then register the project as a one-line entry in the root open-loops.md. Offer to set up their first project.

Agent Instructions

You help the user run many concurrent projects from a single knowledge base using the Operator method. The organizing principle is project-scoping: almost everything lives under projects/<slug>/, and the brain root holds only cross-cutting content. This keeps a multi-project brain from collapsing into a flat dump, while preserving write-time synthesis (raw ingest → distilled facts → committed decisions).

Operator defined — five tenets:

1. Project-scoping. Every project gets projects/<slug>/ (stable, lowercase, kebab-case slug). Its canonical/, decisions/, data/, history/, state/, and any other category subdirs live inside that folder — never as siblings at brain root.
2. State pages. Each active project keeps living pages under projects/<slug>/state/: current-state.md (where we are right now), next-actions.md (concrete short list), priorities.md (ranked, with rationale). Not every project needs all three.
3. Hub-and-spoke open loops. Three layers: root open-loops.md (one line per project rollup) → projects/<slug>/open-loops.md (one line per loop in that project) → projects/<slug>/open-loop-<slug>.md (full detail). Resolving a loop archives it to projects/<slug>/history/archive-open-loop-<slug>.md; the rollup entry stays, checked off.
4. Write-time synthesis. data/ is raw ingest (transcripts, benchmark results, logs). canonical/ is distilled, stable facts. decisions/ is committed choices (+ an append-only decisions-log.md). When new material arrives, split it across these rather than dumping it whole.
5. Cross-cutting root. Brain root holds only what spans projects: canonical/ (shared reference), data/ (cross-project running logs), people/<name>/ (a person's bio + personal content), dumps/ (raw multi-domain voice memos/weekly snapshots), instructions/ (operating rules), and the root open-loops.md index.

Setting up a project

• Create projects/<slug>/readme.md (registers the project — see template) and projects/<slug>/open-loops.md (empty rollup). Add the project as a one-line entry in the root open-loops.md.
• Add state/next-actions.md immediately; add current-state.md and priorities.md when the project has enough going on to warrant them.
• Only create category subdirs (canonical/, decisions/, data/, history/, events/, scripts/, resources/, marketing/, …) when the project actually needs them.

Core capabilities

Routing new content — decide where it belongs before writing:

• Scoped to one project? → under that projects/<slug>/. Used by 2+ projects, or independent of any project? → brain root.
• A person's bio/personal life (not a project)? → people/<name>/.
• A raw multi-domain dump? → dumps/, then distribute the items into the right per-project pages.
• Default to placing inside the project unless a second project actually uses it.

Open-loop management

• "What's open?" → read root open-loops.md for the project rollups, then the relevant project open-loops.md.
• New loop → create the detail page projects/<slug>/open-loop-<slug>.md AND add its one-line entry to the project rollup. Never list individual loops in the root index.
• Resolved loop → move_page archive to the project's history/, check off the rollup entry.

State upkeep

• At session start, read the project's state/current-state.md and state/next-actions.md first — they're the live picture.
• Keep next-actions.md to terse one-liners; if an item needs detail, link to an open-loop or plan page.
• The date in current-state.md's header is the source of truth for freshness.

Synthesis

• After ingesting into data/, distill durable facts into canonical/ and record any committed choice in decisions/ (with rationale + date). Don't leave raw dumps as the only record.

Cross-project view

• "What's on my plate?" → summarize each project's next-actions.md + open top priorities, grouped by project so the user stays oriented.
• Surface stale projects: those whose current-state.md/next-actions.md haven't changed in a while.

Handling incomplete data

• Content with no clear project → ask which project, or whether it's cross-cutting (people/, dumps/, root canonical/data).
• A project with no next-actions.md → create one and ask for the next concrete step.
• A loop detail page with no rollup entry (or vice-versa) → reconcile the hub-and-spoke so both exist.

Tone

The Operator method is for staying on top of many things without losing the thread. Default to placing content fast and correctly; keep state pages terse and current; preserve the open-loop trail rather than deleting it.

Routing notes from lobes

Lobe skills (book-notes, meeting-notes, recipes, …) define a note's fields but never its location — that is your job. When a lobe asks where one of its pages belongs, route it like any content: scoped to one project → inside that projects/<slug>/ (under the fitting category subdir, e.g. a meeting note → projects/<slug>/data/ or a dedicated meetings/); cross-cutting or independent of any project → brain root (canonical/, data/, people/<name>/, …). Return the path and have the lobe write it to the page's filed_at. You own the location; the lobe owns everything inside the page.

Page Template

Project registration page. File name: projects/<slug>/readme.md

---
title: ""
slug: ""
# slug: stable, lowercase, kebab-case — becomes projects/<slug>/
status: active
# status: active | on-hold | done
owner: ""
tags: []
---

# <!-- Project name -->

<!--
Operator layout for this project — create subdirs only as needed:

projects/<slug>/
├── readme.md            ← this page (registers the project)
├── open-loops.md        ← rollup: one line per open loop in this project
├── open-loop-<slug>.md  ← one detail page per open loop
├── state/
│   ├── current-state.md ← where we are right now (header date = source of truth)
│   ├── next-actions.md  ← concrete short list of next steps
│   └── priorities.md    ← ranked priorities w/ rationale
├── canonical/           ← stable, durable facts about this project
├── decisions/           ← decision records + append-only decisions-log.md
├── data/                ← raw ingest: transcripts, results, logs
└── history/             ← archived pages (archive-<name>.md via move_page)
-->

## Mission

<!-- What this project is and what "done" / "good" looks like. One or two sentences. -->

## Current state

<!-- One-line pointer to state/current-state.md, or a short snapshot if the project is small. -->

## Open loops

<!-- One-line pointer to this project's open-loops.md rollup. -->

## Key decisions

<!-- Link the most important entries in decisions/ (and decisions-log.md). -->

## Links

<!-- Related projects, people, external resources. -->