marketing-workspace

Memory for your marketing agent. A plain-markdown campaign graph — your positioning, launch venues, people, posts, mentions, plan, and backlog — that your AI agent reads to decide what to do next and writes back what happened. Part CRM, part editorial calendar, part system of record.

It is also a conformant Open Knowledge Format v0.2 bundle — an open standard for agent-maintained knowledge — so your campaign graph is portable to any OKF consumer, not locked to one tool.

Skill packs like marketingskills teach your agent how to do marketing — copywriting, CRO, SEO, launches. But skills are stateless: every session starts from zero. No record of what was published where, which venues you've exhausted, who replied to your outreach, or why a post exists.

This workspace is that record. Install the skills for the knowledge; clone this for the memory:

• A campaign graph (data/) — product positioning and messaging pillars, ~65 pre-seeded launch venues with posting rules, a people CRM (customers, prospects, amplifiers), the post pipeline (drafts → published, with engagement tracking), third-party mentions, a plan, and a backlog.
• Typed, validated, portable — every document carries frontmatter checked by iwe schema validate, so agent writes stay consistent across sessions. Provenance (generated, sources) and lifecycle (status, stale_after) live in frontmatter as OKF families, and CI checks conformance on every commit.
• Workspace skills (.claude/skills/) — the state workflows the packs don't have: guided setup, internet backfill, venue-aware drafting, engagement refresh, a weekly digest.

Everything is plain markdown in your repo, managed as a knowledge graph by IWE — queryable, schema-validated, editor-native, and fully yours.

Writing code instead of copy? The sibling dev-workspace template is the same idea for a software project.

Quickstart

1. Use this template (or clone) and open the folder in your editor.

2. Install the iwe CLI (manages the graph, validates the schemas):

   brew tap iwe-org/iwe && brew install iwe   # Homebrew
   cargo install iwe                          # or via Rust
   

   Editor extensions (VS Code, Neovim, Zed) bundle the binaries too — see the quick-start guide for all options.

3. Optionally, install a marketing skill pack — recommended: marketingskills:

   npx skills add coreyhaines31/marketingskills -a claude-code
   

   (or via Claude Code: /plugin marketplace add coreyhaines31/marketingskills, then /plugin install marketing-skills.) The workspace is fully functional without a pack — your agent still plans, drafts, and records against the graph. A pack upgrades its marketing technique; skip it now and add it anytime.

4. Point your agent at the workspace — open Claude Code (or any agent CLI that reads AGENTS.md) in the repo root and say: "run the setup". The agent interviews you, fills in data/product.md and data/strategy.md, and prunes the seeded venue list to your ideal customer.

What happens next — the setup interview, the internet backfill, the daily loop — is described below. No agent? The workspace is still a well-organized markdown KB you can work by hand.

How you use it

The workspace is designed to be operated by the agent, with you: most of the work is automated — the agent does the reading, searching, and filing — and you supply the judgment. A typical start:

Day one — the interview (~15 minutes). Say "run the setup". The agent interviews you about the product, the ideal customer, pricing, and your honest stage, then writes the foundation docs (product.md, strategy.md, brand.md, offer.md), creates your messaging-pillar hubs, and prunes the ~65 seeded venues down to the ones that fit your customer. You answer questions and confirm; it files.

Day one — the backfill. Your marketing didn't start today, so have the agent pull in what already exists. Say "backfill the workspace" — the backfill skill guides it end-to-end — or give any web-capable agent the prompt:

"Research online and backfill the workspace: read our website and fill the gaps in the product doc, find mentions of us on HN, Reddit, and blogs and file them, file the posts we've already published with their engagement numbers, and stub competitor profiles from their public pages."

It files each find where the graph expects it — mentions with verbatim quotes and links to their authors, published posts with engagement, competitors with pricing snapshots — and asks you when it can't verify something. The standing rule (baked into the skills and AGENTS.md): never invent — "unknown" beats plausible fiction.

Every session after. Just talk to the agent; the graph is its memory:

• "what should I do next?" — picks from the backlog and the active plan
• "draft a Show HN post" — venue rules + positioning + pack technique, filed as a draft with its goal recorded
• "log this mention: " / "I published it, here's the link"
• "we talked to a churned customer, here are my notes" — filed as an interview, best quotes distilled into the product doc
• "weekly digest" — what shipped, what moved, what to do next week

Every action leaves a record in data/, so the next session — or a different agent entirely — picks up exactly where this one left off. You stay in the loop where judgment matters: confirming positioning, approving venue cuts, and giving the final yes before anything is published.

How the two halves compose

Half	Provides	Lives
Knowledge	marketing technique (the skill pack)	.claude/skills/ (installed)
Memory	your campaign's state (this repo)	data/

The knowledge half is optional and swappable: we deliberately don't bundle a pack (design rationale in STRUCTURE.md) — install whichever one you like, or none. The memory half carries the workspace on its own.

Example: "draft a Show HN post." The agent takes positioning from data/product.md, posting rules from data/community-hn.md, copy technique from the skill pack — then files the draft in data/drafts/hn/, links it from the venue, and records goal and success so future sessions know why it exists.

The integration is native: skill packs like marketingskills read .agents/product-marketing.md for product context before asking you anything. The setup skill generates that file from your workspace (product doc + brand voice + current offer), so every skill in the pack starts pre-briefed — and the workspace's experiment log, metrics baselines, and competitor profiles answer the questions packs otherwise re-ask every session ("what have you tried?", "what's your conversion rate?", "who are your competitors?").

The graph is queryable:

iwe find --filter '{type: community, category: launch-platform, stage: planned}'  # where to launch next
iwe find --filter '{type: post, stage: draft}' -f keys                            # drafts in flight
iwe find --filter '{type: person, stage: contacted}'                              # outreach awaiting reply
iwe find --filter '{type: task, stage: planned, priority: high}'                  # what to do next
iwe schema validate                                                                # is every doc well-formed?

What's inside

data/
  product.md       positioning, ICP, personas, objections, customer language  ← start here
  strategy.md      current stage (first 10 / 100 / 1000), channels, goals
  brand.md         voice, banned words, description variants, asset locations
  offer.md         tiers, value metric, guarantee, pricing change history
  metrics.md       baselines: unit economics, funnel, audience, authority
  website.md       page inventory, keyword map, org facts, redirect ledger
  pillars/         one hub per messaging pillar — links every post carrying it
  community-*.md   ~65 seeded launch venues with posting rules
  communities/     platform-grouped venues (reddit/ holds the 28 subreddits)
  competitors/     one source-of-truth profile per competitor, dated snapshots
  experiments/     everything tried, with hypothesis, verdict, and learning
  people/          your CRM: customers, prospects, amplifiers
  drafts/ posts/   the editorial pipeline, engagement tracked in frontmatter
  evergreen/       reusable platform-independent post templates
  mentions/        third-party mentions of your product
  interviews/      customer research records: quotes, learnings, per interview
  plan/ backlog/   campaign moves and atomic tasks
.claude/skills/    setup · backfill · draft · update · weekly
.iwe/              graph config + frontmatter schemas

SCHEMA.md documents every document type; AGENTS.md is the agent's operating manual.

Built on IWE

IWE is open-source, local-first knowledge-graph tooling for markdown: a CLI (iwe) and an LSP server that plug into VS Code, Neovim, Zed, and Helix. This workspace needs only the CLI — that's what powers the queries, renames, and schema validation above — but with the editor extension installed, the whole campaign graph gets IDE ergonomics: follow links, see backlinks ("which venues link this post?"), rename entities safely, and run extract/inline refactors on your docs.

• Website & docs: iwe.md · quick start
• Source: github.com/iwe-org/iwe

Acknowledgments

Designed to pair with Corey Haines' marketingskills. Venue discovery was guided by Edoardo Stradella's Marketing-for-Founders — a great human-readable collection; go star both. Built on IWE.

License

MIT — see LICENSE.md.