agentleFS
iwe-orgiwe-org/dev-workspace/AGENTS.md

dev-workspace

AGENTS.md11 starsChanged 2 days agoMITView on GitHub

You are operating a dev workspace: a markdown knowledge graph that is a software project's memory and system of record. - Code lives in the project's own repository. Writing it is your normal work — this workspace doesn't change how you code. - Project state lives here, in data/ — what the product is, how it must behave, how it's designed, what's planned, shipped, broken, and released. Every working session must leave a record in the graph; this is your…

  • Installs packages
  • 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 31,439 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.

  • In this repository8 other files it gives its agents.

The file

We split this AGENTS.md into the rules it gives an agent

8 rules. About 0 are common, 1 is distinctive, and 7 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.

  • 7

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Every working session must leave a record in the graph.

    line 10
  2. Very specific:

    Read data/product.md. If ## What is it is still an unfilled ✏️, the workspace has not been set up: run the setup skill before anything else.

    line 20
  3. Very specific:

    Check the state of work: the plans under ## Active in data/plans.md, and the high-priority tasks — iwe find --filter '{stage: planned, …

    line 25
  4. Very specific:

    Consult: before acting: the data/spec/ and data/architecture/ docs for the area, the feature or bug doc the work belongs to, and the …

    line 34
  5. Very specific:

    Stamp: every document you create or meaningfully change: generated: { by: <your agent>/<your model>, at: <ISO 8601 now> }, naming the agent …

    line 59
  6. Very specific:

    data/ is an OKF v0.2 bundle: every document carries a type, and data/index.md and data/log.md keep their reserved shapes (SCHEMA.md, …

    line 86
  7. Very specific:

    Run iwe normalize after any manual edit instead of wrapping text or fixing link targets by hand.

    line 108
  8. Very specific:

    If iwe is not installed, the workspace is still plain markdown you can read and edit, but queries, renames and validation need it.

    line 112

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

Agent operating manual

You are operating a dev workspace: a markdown knowledge graph that is a software project's memory and system of record.

  • Code lives in the project's own repository. Writing it is your normal work — this workspace doesn't change how you code.
  • Project state lives here, in data/ — what the product is, how it must behave, how it's designed, what's planned, shipped, broken, and released. Every working session must leave a record in the graph; this is your memory across sessions. Never keep project state only in conversation.

This manual holds what every session needs. The fields, stage values and body shape of each document type are in SCHEMA.md, the workflows are the skills listed at the end, and the reasons for the layout are in STRUCTURE.md. Read those when the work calls for them.

Start of every session

  1. Read data/product.md. If ## What is it is still an unfilled ✏️, the workspace has not been set up: run the setup skill before anything else. If only other sections still hold a ✏️, setup ran and the developer skipped them: name them once, then carry on. ## Constraints and ## Authoring rules bind everything you write.
  2. Check the state of work: the plans under ## Active in data/plans.md, and the high-priority tasks — iwe find --filter '{stage: planned, priority: high}' --included-by data/backlog -f keys.

The operating loop

  1. Pick the next piece of work: the user's request, an active plan, or the backlog head.
  2. Consult before acting: the data/spec/ and data/architecture/ docs for the area, the feature or bug doc the work belongs to, and the learnings on it (iwe find --lexical "<topic>" --included-by data/learnings -f keys).
  3. Execute in the codebase. Work that deserves a plan gets one first.
  4. Record the state change, in the same session:

    • Work starts → plan skill. Being built → implement skill.
    • Work ships, a bug is fixed, a plan is dropped, a release is cut → verify skill, then ship skill. Changed behavior goes into its spec there.
    • A bug is found or reported → look for its ticket first, open or archived (iwe find --lexical "<symptom>" --filter '{type: bug}' -f keys); a bug that came back reopens its old ticket. Otherwise file data/bugs/<slug>.md before writing any fix (SCHEMA.md, Bugs).
    • Something learned the hard way — a pitfall, a standing rule the developer gave, a belief they corrected → data/learnings/<slug>.md. Only what the next session needs and the code, specs and tickets don't already say.
    • A design decision → data/architecture/<slug>.md, with the alternatives that were rejected. A vision insight → data/concept/<slug>.md.
    • An actionable item → data/backlog/<slug>.md. An idea that is not a commitment → data/someday/<slug>.md.

    A new document is linked from its hub (data/<folder>.md) in the same edit.

  5. Stamp every document you create or meaningfully change: generated: { by: <your agent>/<your model>, at: <ISO 8601 now> }, naming the agent and model you are actually running as (for example claude-code/opus-5); a one-sentence description if it has none; a sources entry when you derived it from code or an external page. Whenever you set stage, set or clear status in the same edit, from the table in SCHEMA.md. If a hub was added or removed, update data/index.md.
  6. Validate — iwe normalize, then iwe schema validate must pass. If the workspace is a git repository (git rev-parse --is-inside-work-tree), commit with a short message naming the state change. If it is not, there is nothing to commit, and you do not create a repository unasked.

Conventions

  • Inclusion link — a markdown link on its own line makes the target a child in the graph. A hub's inclusion link, not the directory, is what makes a document a plan or a feature. A link inside a sentence or a list item is a soft reference.
  • A stage lives in two places — in frontmatter and as the position of the item's link in its hub. Change both in one edit, as part of the change that finishes the work. Frontmatter is the record: hubs, queries and charts never read the body, so a plan with every box ticked and no stage: done is still active. A closed item's link moves; it is never removed.
  • Links are markdown links, never wiki links, relative to the file that contains them and ending in .md ([Timer](spec/timer.md)).
  • One topic per file, named as SCHEMA.md shows for its type.
  • data/ is an OKF v0.2 bundle: every document carries a type, and data/index.md and data/log.md keep their reserved shapes (SCHEMA.md, Reserved files). Never get past a validation failure by unbinding a schema or loosening one.
  • Git is optional. Where a skill says commit, it means: if the workspace is a git repository. Without one, frontmatter dates and generated.at are the history.
  • Example docs — files ending in -example.md, and the releases named example-<version>.md, show each document shape on a fictional product. Ignore them when reporting real project state. The setup skill deletes them, and this paragraph, at the end of onboarding.

IWE basics

The graph is managed by the iwe CLI (IWE), run from the workspace root.

  • A document's key is its path without the extension (data/product, data/plans/20260801-dark-mode). Its title is its H1.
  • Never mv or hand-delete a document. iwe rename and iwe delete update every link in the graph and say how many documents they changed; those edits are part of the change.
  • Run iwe normalize after any manual edit instead of wrapping text or fixing link targets by hand.
  • The iwe skill has the commands and the query syntax; read it before a query or a restructuring you have not run this session.
  • If iwe is not installed, the workspace is still plain markdown you can read and edit, but queries, renames and validation need it. Ask the user to install it (brew install iwe-org/iwe/iwe or cargo install iwe; see https://iwe.md/quick-start/) before restructuring anything.

Workspace skills

SkillWhat it does
.claude/skills/setup/SKILL.mdOnboarding: scans the codebase, drafts product and architecture docs
.claude/skills/explore/SKILL.mdThinking partner: investigate and compare options; never writes code
.claude/skills/plan/SKILL.mdFiles a plan: discovery, verified anchors, spec impact, Active listing
.claude/skills/implement/SKILL.mdExecutes a plan task-by-task: tests, checkbox ticks, clean boundaries
.claude/skills/verify/SKILL.mdPre-ship gate + drift audit: claims in the graph checked against code
.claude/skills/ship/SKILL.mdCloses the loop: spec sync, stage flips, release recording, release cut
.claude/skills/weekly/SKILL.mdRead-only digest: shipped, in flight, bugs, backlog, graph health
.claude/skills/iwe/SKILL.mdThe iwe CLI reference: reading, querying, restructuring, validating

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.