marketing-workspace
AGENTS.md6 starsChanged 3 days agoMITView on GitHub
You are operating a marketing workspace: a markdown knowledge graph that is the campaign's memory and system of record. - Campaign state lives here, in data/ — positioning, venues, people, leads, posts, replies, mentions, readings, plan, backlog. Every action you take must leave a record in the graph; this is your memory across sessions. Never keep campaign state only in conversation. - Marketing technique is your own judgment plus the venue notes. If the user has installed a marketing skill…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
The file
We split this AGENTS.md into the rules it gives an agent
7 rules. About 0 are common, 1 is distinctive, and 6 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 9
Never keep campaign state only in conversation.
- Very specific: line 18
Read data/product.md and data/strategy.md.
- Very specific: line 26
Check the backlog and plan: iwe find --filter '{type: task, stage: planned, priority: high}' and iwe find --filter '{type: plan, stage: …
- Very specific: line 29
Read the active decision notes (iwe find --filter '{type: note, kind: decision, stage: active}') — they are choices the user already made — …
- Very specific: line 38
Consult: before acting: the venue entity (data/community-<slug>.md for standalone sites, data/communities/reddit/<name>.md for subreddits — …
- Very specific: line 162
Links carry
.md: Run iwe normalize after bulk edits rather than hand-writing link targets. - Very specific: line 178
If iwe is not installed, reading and editing still work, but renames, queries and validation do not.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Agent operating manual
You are operating a marketing workspace: a markdown knowledge graph that is the campaign's memory and system of record.
- Campaign state lives here, in
data/— positioning, venues, people, leads, posts, replies, mentions, readings, plan, backlog. Every action you take must leave a record in the graph; this is your memory across sessions. Never keep campaign state only in conversation. - Marketing technique is your own judgment plus the venue notes. If the user has installed a marketing skill pack (e.g. marketingskills), use its skills when the task is how to market. A pack is optional and the user's to install: never install one yourself, and never block on one.
Start of every session
- Read
data/product.mdanddata/strategy.md. The workspace is set up when four sections ofdata/product.mdhave content: What is it, Ideal customer profile, Positioning and Messaging pillars. A section marked✏️ skipped <date>is the user's decision to leave it for later: it counts as dealt with, does not trigger setup and does not block drafting. If one of the four still shows a bare ✏️, offer the setup flow (.claude/skills/setup/SKILL.md) before other marketing work — work without positioning is guessing. - Check the backlog and plan:
iwe find --filter '{type: task, stage: planned, priority: high}'andiwe find --filter '{type: plan, stage: in-progress}'. - Read the active decision notes
(
iwe find --filter '{type: note, kind: decision, stage: active}') — they are choices the user already made — and list the replies due their review:iwe find --filter '{type: reply, outcome: {$exists: false}}', any published three or more days ago.
The operating loop
- Pick the next action from the backlog (or the user's request).
- Consult before acting: the venue entity (
data/community-<slug>.mdfor standalone sites,data/communities/reddit/<name>.mdfor subreddits —notescarry posting rules and flair requirements) anddata/product.mdfor positioning. - Execute — draft the post, prepare the outreach, file the research.
- Record — write the state back, using the list below.
- Stamp — every document you create or meaningfully change gets
generated: { by: <your agent>/<your model>, at: <ISO 8601 now> }, naming the agent and model you are actually running as (e.g.claude-code/opus-5), a one-sentencedescriptionif it has none, and — when you researched it from an external page — asourcesentry with that page'sresource. Whenever you setstage, derive OKFstatusfrom the table inSCHEMA.mdand set or clear it in the same edit. Append a line todata/log.mdunder today's## YYYY-MM-DDgroup (create the group if it isn't there, newest first); if the top-level set of documents changed, updatedata/index.mdtoo. - Validate —
iwe schema validatemust pass. In a git repository, then commit with a short message describing the state change (see Git is optional below).
What goes where
One line each; SCHEMA.md has the fields and the skills have the steps.
- Post →
data/drafts/<platform>/YYYY-MM-DD-<slug>.md, moved todata/posts/<platform>/when published (the draft skill). On Reddit the platform folder nests one level deeper, per subreddit:data/posts/reddit/<subreddit>/. This is the only nesting rule. - Comment in someone else's thread → thread page, lead, then reply, in the
order the draft skill gives under Replying in someone else's thread; its
outcomeat the three-day review (the update skill). - Lead converts → create their
data/people/record, link it from the lead, set the lead'sstage: converted, move its link indata/leads.md. - Contact → the person's doc in
data/people/, inclusion-linked from its role hub. - Outreach email →
data/emails/YYYY-MM-DD-<slug>.md, recipient linked inline; when sent,stage: sentwithsent, and update the person'sstage. - Mention →
data/mentions/YYYY-MM-DD-<slug>.md, linked fromdata/mentions.mdand the venue's## Mentions. - Interview or survey →
data/interviews/YYYY-MM-DD-<slug>.md, interviewee linked inline; distill the best language intodata/product.md. - Launch plan step finished → append a
## Retro(numbers, what worked) before closing it. - Venue worked or ruled out → flip its
stage(active, orcancelledwith the reason innotes). - Task finished →
stage: done,completed, and its link moved to## Doneindata/backlog.md. - Something tried →
data/experiments/<slug>.md; on conclusionverdictandresult. Before proposing one, check what was tried:iwe find --filter '{type: experiment}' --project '$title,stage,verdict'. - Competitor fact →
data/competitors/<slug>.md: dated Changelog entry, bumpupdated;threatonly with the reason in the body. - What the record means → a reading at
data/readings/<slug>.md, itsquestionagreed with the user first; linked fromdata/readings.mdbyimpact, moved todata/readings-archive.mdwhenstageleavesopen. - Reason to buy →
data/selling-points/<slug>.mdwith itsevidence. Ahypothesispoint never goes into copy as a claim. - Outside material →
data/references/<slug>.mdwith the ideas written down;stage: usedand a link where an idea went. - Audience segment →
data/audiences/<slug>.md; aninactiveone keeps its reason innotes. - Decision, learning or gotcha →
data/notes/<slug>.md.kind: decisiononly for what the user stated or confirmed; a replaced note is setsuperseded, never deleted. - Number → baselines in
data/metrics.md(value and dated snapshot); weekly numbers as one appended row indata/metrics-weekly.csv, never an edit to a past row. - Positioning or messaging changed →
data/product.mdand its Changelog, then regenerate.agents/product-marketing.md.
Conventions
- Git is optional. A workspace created by the iWe app is a plain folder, not
a git repository; check once per session
(
git rev-parse --is-inside-work-tree). Inside a repository, commit each validated state change and readgit diffafter a rename or delete. Outside one, skip every commit and diff step here and in the skills, and re-read the files a rename or delete touched instead. Setup offersgit initonce; do not raise it again. - Inclusion link = a markdown link on its own line; it makes the target a
child in the graph. Hubs inclusion-link their members; links inside sentences
and list items are soft references.
data/index.mdis link bullets only. - Relationships are links, not fields. A post's venue is the link under the
venue file's
## Posts(iwe find --referenced-by data/community-<slug> --filter 'type: post'). A person's role is an inclusion link from adata/people/role-*.mdhub (iwe find --included-by data/people/role-customer). A post's pillar is both: thepillarfield to filter on, and an inclusion link fromdata/pillars/<pillar>.md, whichdata/pillars.mdincludes, so the post is reachable in the tree. - Every publication states
goalandsuccess— the observable signal that means it worked. If you can't name them, the post isn't worth planning. - Dated filenames: posts, mentions, interviews, replies and emails are
YYYY-MM-DD-<slug>.md, kept weekly reportsYYYY-MM-DD.md; people, leads, threads, venues and tasks are undated slugs. - In other people's threads the user writes the words. You find threads,
read them, and give talking points and accuracy checks; the user writes and
posts the comment, and the reply's
originrecords who wrote it. - Each fact is written once, on the record made when it happened. A lead stores no reply count or "answered" flag — that is read from the replies that link to it; a reply stores no thread — the thread page inclusion-links it.
- Facts and readings stay apart. A count, a quote or a date is a fact; what it means is a reading. Before a reading becomes a rule, write down another explanation and the evidence that would tell them apart.
- Closed items move, they are not deleted — to the backlog's
## Done, the leads hub's## Convertedand## Dropped,data/readings-archive.md. - Repurposing trail: a dated post cut from an evergreen template links back to the template inline.
- Example docs:
*-example.mdfiles show each directory's document shape with a fictional product, PlainInvoice. Leave them out of any report of real campaign state; setup deletes them. - Shapes are enforced by
.iwe/schemas/*.yaml(SCHEMA.mdis the readable version). Runiwe schema validate -k <key>after editing a document.data/is also an Open Knowledge Format v0.2 bundle: every document has a non-emptytype,data/index.mdstays sections of link bullets with no frontmatter beyondokf_version, anddata/log.mdstays bullets under## YYYY-MM-DD. Never unbind a schema to get past a failure. - Charts read the records, so never type their numbers into prose. Two
habits keep them true: every post, mention and reply links its venue page once
in the body, and every reply's
## Contextopens with thePerson:line. Measurements that aren't documents are CSV or JSON files beside the page that charts them, one row per period with ISO dates. - Links carry
.md. Runiwe normalizeafter bulk edits rather than hand-writing link targets. .agents/product-marketing.mdis generated by setup fromdata/product.md, the brand voice and the current offer, for skill packs that read it. Never edit it; edit the sources and regenerate.- Web research is part of the job: backfill, update, thread search and a venue refresh fetch public pages. File only what a page you opened shows — never invent a number, a quote or a date.
The iwe CLI
- A document's key is its path from the workspace root without the extension
(
data/product,data/posts/hn/2026-08-01-launch). - Never
mvor hand-delete a document —iwe renameandiwe deleteupdate every link. A deleted document's inclusion links are removed and inline links to it become plain text; remove that leftover bullet yourself. - If
iweis not installed, reading and editing still work, but renames, queries and validation do not. Ask the user to install it (index.mdhas the commands) before restructuring anything. - Commands and query syntax are in the
iweskill (.claude/skills/iwe/SKILL.md); read it before a query or a restructuring you have not run this session.
Workspace skills
| Skill | What it does |
.claude/skills/setup/SKILL.md | Onboarding interview → fills product.md/strategy.md, prunes venues |
.claude/skills/backfill/SKILL.md | Imports existing mentions, posts, competitor facts from the web |
.claude/skills/draft/SKILL.md | Drafts a post, or talking points for a thread reply, then files it |
.claude/skills/update/SKILL.md | Refreshes engagement on published posts and replies |
.claude/skills/weekly/SKILL.md | Prints a weekly digest; keeps it in data/weekly/ when asked |
.claude/skills/iwe/SKILL.md | The iwe CLI reference: reading, querying, renaming, validating |
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

