agentleFS
foversefoverse/seoverse/AGENTS.md

seoverse

AGENTS.md4 starsChanged 10 days agoMITView on GitHub

OpenSEO also runs behind an Every App Gateway (AUTH_MODE=everyapp; how it works is in docs/maintainers/every-app.md).

1 other project uses a version of this.

Adopted by

Steady: 0 new projects took it in the last 4 weeks, none the 4 weeks before.thin data

Week of Sep 13: 0 new projectsWeek of Sep 20: 0 new projectsWeek of Sep 27: 0 new projectsWeek of Oct 4: 0 new projectsWeek of Oct 11: 0 new projectsWeek of Oct 18: 0 new projectsWeek of Oct 25: 0 new projectsWeek of Nov 1: 0 new projects
Sep 2026Oct 2026Nov 2026
New projects taking a version of this file each week, over the 8 weeks our latest run read: only 0 in all, so this is a hint, not a trend.

Most recently taken up by

The file

Show its rules and text

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

21 rules. About 0 are common, 14 are distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈14

    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:

    Prefer simple, readable, flat code with minimal indirection.

    line 5
  2. Distinctive:

    Search for existing implementations and installed libraries before creating new helpers or abstractions.

    line 7
  3. Distinctive:

    Keep product data normalized and relationships explicit.

    line 9
  4. Distinctive:

    Keep schema changes, queries, and mutations compatible with both SQLite and Postgres.

    line 11
  5. Distinctive:

    Use idiomatic TypeScript. Use Zod to validate untrusted data and narrow runtime values at trust boundaries.

    line 12
  6. Distinctive:

    Prefer established project helpers and libraries over hand-rolled implementations.

    line 13
  7. Distinctive:

    Prefer idiomatic TanStack Query, Router, and Form patterns for server state, routing, and submitted forms.

    line 14
  8. Very specific:

    Before running everyapp, read the everyapp skill.

    line 22
  9. Very specific:

    Run the app with pnpm dev:everyapp (everyapp dev): both Workers behind a local gateway with a seeded dev user.

    line 23
  10. Distinctive:

    Don't add tests just for the sake of it.

    line 30
  11. Distinctive:

    Keep tests as simple as possible, and always review them looking for simplifications.

    line 31
  12. Distinctive:

    Test behavior at the public entry point.

    line 32
  13. Very specific:

    Never re-declare a production class in a test.

    line 34
  14. Distinctive:

    Don't mock ORM builder chains.

    line 38
  15. Distinctive:

    Don't export a function only so a test can reach it.

    line 39
  16. Very specific:

    When a bump is warranted, keep plugins/openseo/.codex-plugin/plugin.json, plugins/openseo/.claude-plugin/plugin.json, and …

    line 48
  17. Very specific:

    Run pnpm sync-plugin-skills after changing a shipped skill and include the updated plugin copy, even when no version bump is needed.

    line 50
  18. Distinctive:

    Do not put internal working notes, content briefs, drafts, asset-sourcing notes, implementation logs, or agent handoffs in those …

    line 55
  19. Very specific:

    Use docs/maintainers/ for engineering decisions, development workflows, and maintenance notes that should be shared and versioned.

    line 56
  20. Very specific:

    Write PR descriptions in the simple-issue-description format, with no validation logs or per-file changelog.

    line 57
  21. Distinctive:

    Do not mine an entire session for papercuts or start a broad cleanup unless the user explicitly asks.

    line 63

≈ “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 guidance

Engineering principles

  • Prefer simple, readable, flat code with minimal indirection.
  • Work on bugs a real user hit or would plausibly hit. Don't open a ticket or PR for an edge case found only by reading the code; flag it instead. Security holes and significant billing abuse are the exception.
  • Search for existing implementations and installed libraries before creating new helpers or abstractions.
  • Abstract when it prevents meaningful drift and makes the result simpler to maintain. Avoid speculative or one-use abstraction layers.
  • Keep product data normalized and relationships explicit. Do not encode relational data in JSON or text merely to avoid joins.
  • For new application-backed backend functionality, default to: TanStack server function → service → repository.
  • Keep schema changes, queries, and mutations compatible with both SQLite and Postgres.
  • Use idiomatic TypeScript. Use Zod to validate untrusted data and narrow runtime values at trust boundaries.
  • Prefer established project helpers and libraries over hand-rolled implementations.
  • Prefer idiomatic TanStack Query, Router, and Form patterns for server state, routing, and submitted forms.
  • shadcn components in src/client/components/ui/ are built on Base UI, not Radix. Compose them with the render prop, not asChild.
  • Specs under docs/maintainers/specs/ are public design records: what a feature does, how it works, the alternatives considered and why they lost. No line numbers, migration mechanics, test plans, incidents, costs, or internal infrastructure details.

Every App

OpenSEO also runs behind an Every App Gateway (AUTH_MODE=everyapp; how it works is in docs/maintainers/every-app.md).

  • Before running everyapp, read the everyapp skill. It lives in the platform checkout beside this repository, ../private-every-app/packages/cli/skills/everyapp/SKILL.md; everyapp create copies it into new apps, and this port reads it from there.
  • Run the app with pnpm dev:everyapp (everyapp dev): both Workers behind a local gateway with a seeded dev user. The keys it needs locally go in .dev.vars, written by hand.
  • Deploy only through the Every App CLI: pnpm everyapp install https://github.com/every-app/open-seo from an Every App checkout (the npm CLI installed here cannot deploy). Never run cf deploy or wrangler deploy for this mode. deploy:selfhost is the other self-hosting path (pnpm deploy is the plain wrangler deploy of wrangler.jsonc) and deploy:postgres is app.openseo.so's production deploy; none of them is this one.
  • Declare resources (D1, KV, Durable Objects, Workflows, the audit Worker) and settings (providers, capabilities, secrets, each with its need and why) in everyapp.config.ts; cloudflare.config.ts builds the Workers from it.
  • The Gateway signs people in: read identity only through the SDK (src/middleware/ensure-user/everyApp.ts), never from a client-supplied header or cookie, and treat every route listed under public in everyapp.config.ts as open to the internet.

Testing

  • Don't add tests just for the sake of it. A test exists to enforce core behavior or a hard-to-spot edge case that could actually occur.
  • Keep tests as simple as possible, and always review them looking for simplifications.
  • Test behavior at the public entry point. Assert argument forwarding to a mocked collaborator only when that mapping is the contract (billing params, telemetry events).
  • Statically import the module under test. vi.mock is hoisted, so per-test await import() and vi.resetModules() are banned unless module-level state must reset — comment why.
  • Never re-declare a production class in a test. Import the real one; if the module is too heavy to import, move the class to a leaf module first (see ga4Errors.ts, gscErrors.ts).
  • beforeEach sets default mock return values only. Vitest's clearMocks already resets call state — no mockReset/mockClear ceremonies.
  • Fixtures contain only the fields the test asserts on or the types require. Shared shapes get a factory with overrides (see ga4-test-fixtures.ts, tool-test-support.ts); a fixture longer than its test's assertions is a smell.
  • One test per invariant. Don't re-test Zod or a library, and don't repeat an output-schema round-trip in every happy path.
  • Don't mock ORM builder chains. Test repositories through services or real SQL evaluation; chain mocks break on refactors that change no behavior.
  • Don't export a function only so a test can reach it. Test through the public entry point (the service object, tool handler, or workflow); Knip fails CI on the orphaned export.
  • Vitest's clearMocks clears call history only. A mockResolvedValue or mockImplementation set inside a test survives into the next one, so every persistent default belongs in beforeEach.
  • A negative test must fail for the reason its name gives. If a different guard rejects first (hosted mode off, missing auth, an earlier validation), the test proves nothing about the behavior it names.

Plugin versions

  • Bump the OpenSEO plugin version when a PR adds a new MCP tool, adds a new shipped skill, or substantially changes a shipped skill's workflow or instructions. Substantial skill changes include changing its required steps, tool choices, or expected output.
  • Routine fixes, refactors, and changes to existing tools' implementation, behavior, schemas, or descriptions do not require a bump. Neither do minor skill edits, internal-only skills, tests, website or app changes, or routine plugin configuration edits. Use judgment for other major plugin changes; do not bump merely because a plugin-related file changed.
  • The plugin points at the hosted MCP server, so server changes reach users on deploy without a plugin version bump. When a fix applies to both the app and an MCP tool, ship it to both.
  • When a bump is warranted, keep plugins/openseo/.codex-plugin/plugin.json, plugins/openseo/.claude-plugin/plugin.json, and plugins/openseo/.cursor-plugin/plugin.json on the same version. Default to a patch bump unless the maintainer requests a minor or major bump.
  • Bump once per PR relative to origin/main, not once per edit or tool. For a PR that requires a bump, if main advances the plugin version before merge, update the branch to a version newer than main. Otherwise, keep main's version. The app's package.json version follows its separate release process.
  • Run pnpm sync-plugin-skills after changing a shipped skill and include the updated plugin copy, even when no version bump is needed. Call out plugin version and instruction changes for maintainer review.

Documentation audience

  • docs/ (except docs/maintainers/) and web/content/docs/ are public, user-facing documentation. Write only material that helps users understand, use, or self-host the product.
  • Do not put internal working notes, content briefs, drafts, asset-sourcing notes, implementation logs, or agent handoffs in those directories.
  • Use docs/maintainers/ for engineering decisions, development workflows, and maintenance notes that should be shared and versioned. This directory is tracked in Git but is not part of the user-facing documentation site. Create notes only when requested or useful for future maintenance; never include secrets.
  • Write PR descriptions in the simple-issue-description format, with no validation logs or per-file changelog. Continue using the dedicated papercut log for repository friction.

Log papercuts

When small, non-blocking repository friction occurs—a retried tool call, confusing setup step, flaky command, stale cache, misleading error, or non-obvious gotcha—use the papercuts skill and append it to .agents/PAPERCUTS.md in the moment. Continue the current task. Real bugs and tracked work are not papercuts, and sensitive data must never be logged.

Do not mine an entire session for papercuts or start a broad cleanup unless the user explicitly asks.

Preserve review learnings

Reviewer context lives in docs/maintainers/review-guidelines.md: the hard invariants (tenant scoping, dual-dialect persistence, billing seams, SSRF handling) and the false-positive controls that keep reviewers from flagging intentional behavior. After a merge-ready or other code review verifies a finding, add to that file only when the finding exposes a recurring or high-risk repository invariant that existing context and automated checks do not capture. Do not promote one-off bugs or preferences into permanent review rules.

Changes to AGENTS.md, .claude/CLAUDE.md, .agents/skills/**, .github/**, and docs/maintainers/review-guidelines.md alter the review control plane and must receive explicit maintainer review. CODEOWNERS requests that review; where repository settings allow, enable GitHub's requirement for code-owner approval.

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.