WongStack
AGENTS.md7 starsChanged 9 days agoMITView on GitHub
This repo is WongStack: an AI assistant and knowledge center a business owner and their team run their business on, shipped as a template you clone and work from. The payload manifest lists what ships. The README tells the user story. This meta-repo ships WongStack and dogfoods it, so the block below applies here too. Don't run /wong-setup or /wong-sync here; both stop. A payload edit loads the release rules; the full process: wiki/maintaining/. Before a change, read the owning…
- Reads credentials
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 31,700 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 repository24 other files it gives its agents.
- agent-browserSkill
- applySkill
- browserSkill
- closeSkill
- connectSkill
- continueSkill
- dailySkill
- dream-memorySkill
- exploreSkill
- hand-overSkill
- improve-codeSkill
- lessonSkill
- memorySkill
- paste-this-into-your-aiSkill
- planSkill
- saveSkill
- scheduleSkill
- servicesSkill
- shipSkill
- shopSkill
- update-dependenciesSkill
- verifySkill
- wong-setupSkill
- wong-syncSkill
The file
We split this AGENTS.md into the rules it gives an agent
2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 15
Before a change, read the owning doc: start at wiki/README.md and follow the links down.
- Very specific: line 22
Before substantial work, memory.mjs recall <question> finds evidence.
≈ “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.
Files
AGENTS.md
What this is
This repo is WongStack: an AI assistant and knowledge center a business owner and their team run their business on, shipped as a template you clone and work from. The payload manifest lists what ships. The README tells the user story.
This meta-repo ships WongStack and dogfoods it, so the block below applies here too. Don't run /wong-setup or /wong-sync here; both stop.
A payload edit loads the release rules; the full process: wiki/maintaining/.
<!-- WONG-STACK:BEGIN — generic conventions, copied verbatim into each install: no repo-specifics. -->Where context lives
Before a change, read the owning doc: start at wiki/README.md and follow the links down. Four memory surfaces, one job each; no fact on two:
openspec/changes/<slug>/: the plan and why it has this shape. Ships, then archives.- The memory store, outside git: each session's facts. Superseded, never edited. A digest loads at session start (convention).
wiki/andknowledge/: repeatable knowledge (philosophy, style). Canonical and curated; grows from use.openspec/specs/and the archive: what shipped, never changed.
Before substantial work, memory.mjs recall <question> finds evidence; read originals. Keep areas <path> for warnings and openspec list for plans.
Credentials for this computer's own tools sit in the git-ignored .env at the primary worktree, mapped by the committed, values-blank .env.example. The app holds each business service's key: use it through the services skill. Don't stub a call or ask for a key in chat: declare a missing one, then send the key link.
Rules
- Always use ASD-STE100 Simplified Technical English for user-facing prose and documentation. Best effort is enough. Keep code, commands, identifiers, quotations, and prescribed text exact.
- Do plain requests directly. Research, errands, reminders, and questions need no verb; ask only for a needed answer. Repo edits end with publish it?: the change loop.
- One task, one workspace. A finished task closes its Paseo workspace through
/close, with no question but a routine or app offer; a new task opens a new workspace: one task, one workspace. - Offer a schedule or a mini app when a finished task will clearly come back: they said it recurs, or memory shows they asked before. Offer once, in the closing question; remember a no: offer a routine or an app.
- Keep messages short and plain, in our voice: a few lines, more only if asked. Name git, OpenSpec, or CI only if the person asks or must act. Plans, questions, and reports use plain words for everyone.
- A person just asks to build or change code; you run the verbs
/explore → /plan → /apply → /save → /ship, plus/continue,/close,/verify,/improve-code,/schedule, and/wong-sync. Asked with no verb, start at/plan, never asking Plan it? or naming the step, and stop at the plan and at the preview: just ask. A verb also serves work that changes no repo file, with a confirm per outward action. - Print the plan's link whenever you make or change a plan, above any closing question: print the plan's link.
- Build a new standalone page or tool as a mini app: its own folders in the main app, served at
/apps/<name>/, through the same loop: mini apps. - The WongStack skills own all git; OpenSpec never runs git.
/applyreads branch changes but makes none: the change loop. - CI is the gate when present, else PR review; a local check is only a pre-check: the gate.
- Send an improvement upstream by hand: contributing.
- A scheduled
/improve-codewrites a plan and waits: code improvement. - Write repeatable knowledge when you learn it: what will help a different, future task, placed by the wiki rules; a
knowledge/page is saved at once. A change's specifics stay in its proposal. - Company actions and memory reads: discover only what you need through the shared helper; company calls use employee login, memory keeps its installed access; skills call actions. An MCP server goes through the app, never in a chat's settings.
- API keys/tokens: human steps, or a browser while connecting.
- Browse as the person; show key moments; ask for what needs them: logins, live view, pictures, asking, buying.
- Path-scoped conventions load from
.claude/rules/; an agent that doesn't auto-load them reads those whosepaths:match its files.
<!-- WONG-STACK:END -->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.

