See what content is possible
Give your AI a brief about your product and readers, and see the formats, channels and routes that fit, each with what it takes.
How Content Atlas is built
AGENTS.md — content-atlas
content-atlas › AGENTS.md
1# AGENTS: content-atlas23Operating procedure for this worker. [PROFILE.md](PROFILE.md) owns its declaration, interfaces and authority. [INVARIANTS.md](INVARIANTS.md) owns required properties and verification. This file owns routing, execution, recovery and delivery.45## Routing67| Request / condition | Execution path / stop point |8| --- | --- |9| "What can we do for …", "what content works at …", a requirement without existing content | `.claude/skills/resolve-possibilities` → artifact with Steps, Links, checked Routes |10| Existing content supplied (list, sitemap, inventory), "what do we have", "what is missing", "where can this go" | `.claude/skills/map-existing-content` → artifact with Placement, Coverage, Steps, Links, checked Routes |11| A focus on a finished result ("owned and earned channels", "written formats", "B2B", "Secondary and Primary fits") | `.claude/skills/filter-possibilities` on that artifact → new artifact |12| A requirement with a focus and no artifact yet | resolve-possibilities or map-existing-content first, then filter-possibilities; the focus goes into filter.json |13| Which pieces to make, when, in what order | out of scope: planner; return the possibilities artifact as its input |14| Messages, claims, copy, production | out of scope: message owner, content producer |15| Adding or changing atlas values, fits, transformations, recipes | out of scope for a request; record the proposal in `handoff.md` |1617Use the simplest supported executor. Do not create additional agents or workflows for a request that a skill can complete.1# PROFILE: content-atlas23## Identity45| Field | Value |6| --- | --- |7| Id | `content-atlas` |8| Name | Content Atlas |9| Type | Declarative |1011## Responsibility, purpose1213**Content requirement or existing content × the Content Atlas → possibilities: the constructions, links and checked routes the atlas holds for it.**1415Given a requirement and context, the worker uses AI judgment within the declared atlas to resolve applicable content constructions and the routes between them, with programs executing the atlas's enumeration, checks and filters. The result is used by whoever plans or produces content: it shows everything that is possible for the people and states in question, how one piece leads to the next, what existing content already covers and where nothing exists. Its fitness rests on three properties: every value is an atlas value, every construction and link is checked against the atlas rules, and nothing is cut before the requester asks for a filter.161---2name: resolve-possibilities3description: Given a content requirement in ordinary words, resolve everything the Content Atlas holds for it — the steps reachable from the people and states it is about, the links between them, and checked routes — as a possibilities artifact. Use when a requester asks what content they can do, for whom, where.4---56# Resolve possibilities78Requirement × atlas → possibilities: Steps, Links, checked Routes and the atlas rows they use, in `artifacts/<name>/`.910## Inputs1112The requirement as stated. Optional: states and audiences the requester names. Shape: contracts/inputs.md › Resolve request.1# Outputs23Every result is possibilities: what the atlas holds for the request, with the routes composed over it. Nothing is ranked. Resolve and map never cut by channel, ownership, format, track or grade; filter does, as a separate step on a finished result.45## Files in `artifacts/<name>/`67| File | Reader | Holds |8|---|---|---|9| request.json (or filter.json) | anyone | the request as run |10| possibilities.json | AI, further processing | the request and every sheet below under `sheets`: name, description, columns, rows |11| possibilities.html | people | self-contained, offline interactive result: content pieces, audience, reader state, format, channel, placement, supporting details, connections, checked routes and summary; all tables remain inspectable |12| possibilities.xlsx | people | one tab per sheet, styled like the atlas workbooks |13| summary.md | anyone, first | the request, rows per sheet, route findings, placement and request check; no data |14| routes.json | the worker | the composed routes, before they are checked; after a filter, the routes whose every step stays |15| work-result.md | requester | state, deliverables, acceptance with evidence, decisions, unresolved work (AGENTS.md › Result) |1617All existing commands retain workbook output by default. For intermediate runs, `--no-xlsx` writes the same canonical JSON and findings plus its summary, and removes any earlier workbook. Finish with `python tools/possibilities.py export-xlsx <result-folder>` to render the final workbook and refresh the summary without rerunning enumeration or route checks. The normal skill delivery includes the final workbook and HTML. Every result write refreshes the HTML, including `--no-xlsx` operations. At final delivery run `python tools/possibilities.py export-html <result-folder> --open`; this uses the saved JSON without changing its structure or bytes, routes or an existing workbook. The HTML embeds its data and needs no server or network. Search and filters change only what the reader sees, never the artifact. An unavailable browser is reported with the usable file path.18mainUTF-8 · Markdown
| Skill | Transformation |
|---|---|
filter-possibilities | Possibilities artifact × params → a narrower possibilities artifact |
map-existing-content | Existing pieces × atlas → placement, coverage and possibilities anchored on the pieces |
resolve-possibilities | Requirement × atlas → possibilities |
Example
The LeanOS venture's content possibilities, produced with Content Atlas.
Give me content possibilities for LeanOS as a whole venture. We build systems that operate with AI and engineer with AI, and we run our own venture that way. The aim is adoption of this way of working and my recognition as a practitioner, through practice in public on constructions that run. Our readers are practitioners developing AI workers and engineers building with coding agents. Show the reader states, formats, channels and routes that could reach them, so I can choose what to make.Possibilities
artifacts/leanos-venture/possibilities.json
- Customer States7Unaware · Problem-aware · Solution-aware · Product-aware
- Audiences6Prospect user · Prospect champion · Prospect buyer · Prospect blocker
- Formats78Blog post / article · Pillar page / topic hub · Case study · Whitepaper
- Channels74Homepage · Blog · Resource / content hub · Product pages
- Surfaces73Audio ad slot · Booth · Definition page · Email slot
- Steps175Problem-led short video · Paid reach to a defined segment
- Routes10Practitioners developing AI workers: explanation of a real construction → live workshop comparing approaches → the construction open to examination
| What you get | You’ll need |
|---|---|
| The atlas: 148 formats, 175 channels and 11 reader stages, for your AI to work from. | An agent that reads skills, such as Claude Code or Codex |
| 3 skills: explore a content question, map the content you have, narrow a result. | Python 3.10 or later, with openpyxl |
| Format transformations, to see what one piece can become. | Git, to copy the repository |
| Each result as a workbook to read, JSON to reuse and a summary. |
Get it
- Licence
- MIT, for your own ventures and client work. Copy, change and share it.
- Access
- The public GitHub repository, with no purchase or sign-up.
- Updates
- Released to the public repository.
- Help
- Raise an issue on the repository.