mecatl / user-docs
AGENTS.md308 starsChanged 37 days agoApache-2.0View on GitHub
Public documentation for Mecatl lives here. Keep site infrastructure in website/; see website/AGENTS.md for Docusaurus mechanics. Before editing a page, read the repository's style guide and this authoring contract. Coding agents must use the tech-writer skill when editing user documentation. The style guide owns prose, voice, and terminology; this file owns information architecture, content ownership, links, and verification. Document what readers need but cannot reasonably infer from the interface. Omit routine navigation, self-evident actions, and instructions already shown in UI…
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 31,774 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 repository12 other files it gives its agents.
- AGENTS.mdAGENTS.md
- engine/AGENTS.mdAGENTS.md
- internal/AGENTS.mdAGENTS.md
- website/AGENTS.mdAGENTS.md
- cut-releaseSkill
- mecak8s-kind-manual-verifySkill
- mecatl-learning-configSkill
- mecatl-model-router-configSkill
- panel-reviewSkill
- perf-mcp-interpretationSkill
- perf-optimizationSkill
- tech-writerSkill
The file
We split this AGENTS.md into the rules it gives an agent
12 rules. About 0 are common, 6 are distinctive, and 6 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
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
- Very specific: line 7
Before editing a page, read the repository's style guide and this authoring contract.
- Distinctive: line 13
Document what readers need but cannot reasonably infer from the interface.
- Distinctive: line 21
Start with the reader's task.
- Very specific: line 47
Keep shared behavior in features/.
- Distinctive: line 84
When adding a page that appears to overlap an existing page, identify its canonical page in the pull request description.
- Very specific: line 100
Use front matter on every documentation page.
- Very specific: line 107
Use sidebar_label only when a shorter sidebar label materially improves navigation.
- Very specific: line 121
Write about current behavior in clear technical prose.
- Distinctive: line 136
Use lowercase, hyphenated cloud-native harness except at the start of a sentence, heading, or navigation label, where Cloud-native harness …
- Very specific: line 143
Use content-root-relative Markdown paths for links across sections, including the .md extension:
- Distinctive: line 156
Start with the applicable implementation source:
- Distinctive: line 186
Run these commands from the repository root:
≈ “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
Writing Mecatl user documentation
Public documentation for Mecatl lives here. Keep site
infrastructure in website/; see website/AGENTS.md
for Docusaurus mechanics.
Before editing a page, read the repository's style guide and this
authoring contract. Coding agents must use the
tech-writer skill when editing
user documentation. The style guide owns prose, voice, and terminology; this
file owns information architecture, content ownership, links, and verification.
Document what readers need but cannot reasonably infer from the interface. Omit routine navigation, self-evident actions, and instructions already shown in UI hints or help. Keep non-obvious behavior, prerequisites, consequences, and limitations that affect the reader's task. A UI change does not require another walkthrough of its controls.
Find the right place
Start with the reader's task. The autogenerated sidebar follows these directories; do not create a manual sidebar to express this structure.
| Location | Audience and content | Exclusions |
intro.md | New readers choosing among the terminal-client, operator, and builder paths. | Detailed instructions and reference material. Link to the owning section. |
cloud-native-harness.md | Readers understanding Mecatl's defining architecture and how it differs from desktop harnesses. | Task-oriented setup, deployment, and configuration. Link to the owning guide. |
mecatui/ | Terminal users who need to install, start, connect, configure, or troubleshoot the mecatui client. | Server deployment, server policy, and shared feature semantics. Link to the owning server or feature page. |
operating/ | Operators deploying, securing, and maintaining mecated, mecak8s, Studio, and remote services. | Client workflows, embedding APIs, and shared capability behavior. Link to the owning client, builder, or capability page. |
building/go/ | Go builders following tutorial, concepts, embedding, extension points, and compatibility. | Operator runbooks and duplicate capability explanations. |
building/ci/ | Builders integrating one-shot agent jobs into CI. | Long-running service deployment. |
building/typescript-sdk/ | TypeScript developers connecting applications, controlling sessions and runs, or owning a local daemon. | Wire-protocol reference and shared feature semantics. Link to the owning reference or feature page. |
building/go/embed-engine.md | Go developers embedding the engine and supplying its runtime dependencies. | Shared capability behavior and operator procedures. Link to the owning feature or operations page. |
building/go/extension-points/ | Developers integrating with Mecatl's public seams, ports, and APIs. | How to operate a supplied deployment or use an existing feature. |
features/ | Readers understanding a capability of the shared agent and server core. The sidebar labels this section Capabilities. Each page owns shared behavior and availability. | Terminal-client controls, embedding APIs, and deployment runbooks. |
reference/ | Readers looking up generated configuration or SDK fields and exact gRPC and HTTP/SSE contracts. | Tutorials, deployment workflows, and conceptual background. Link to the owning guide. |
The top-level navigation follows three journeys: Use Mecatl, deploy and
operate Mecatl, or build with Mecatl. features/ is the canonical capability
spine rather than a fourth persona journey. Persona-specific pages briefly
introduce shared behavior, then link to the capability page that owns it.
mecatui is a terminal client over an embedded or connected server, not a
separate agent implementation. building/ is for application builders and
embedders. operating/ is for people who own a deployed service.
Keep shared behavior in features/. Put differences among mecated, mecak8s,
and embedded mecatui in operating/ and the
capability matrix.
Keep one canonical page
One page owns the complete reader-facing treatment of a topic. Other pages may briefly introduce it for their audience, then link to the canonical page. A page can separately describe an embedding interface or a deployment-specific operation when that is the reader's task; it must not repeat the canonical feature guide.
| Topic | Canonical page | Supporting page responsibility |
Installing mecatui or mecated | mecatui/installation.md and operating/mecated.md | Each guide owns its package and release-archive instructions. The shared archive-verification fragment keeps verification commands identical. |
| Using the terminal client | mecatui/ | Feature pages link to the relevant client guide rather than repeat terminal-client workflows. |
| Agent loop and built-in tools | features/sessions/agent-loop.md and features/sessions/tools.md | Embedding guides show the minimum wiring and link to these pages for shared behavior. |
| Scheduled tasks | features/sessions/scheduled-tasks.md | Deployment pages own scheduler flags and storage configuration. |
| Permissions and posture | features/security-and-execution/permissions-and-posture.md | building/go/extension-points/permission-policy.md owns custom-policy integration. |
| Install and first run | building/go/first-agent.md and building/go/demo.md | Deployment pages cover their own prerequisites only. |
mecated operation | operating/mecated.md and its workflow guides | Shared semantics stay in Capabilities; complete flags belong in the generated server CLI reference. |
mecak8s operation and Helm values | operating/mecak8s.md and its workflow guides | Feature pages describe shared behavior without restating chart values. |
mecatequi and GitHub Actions | building/ci/mecatequi.md | Workflow maintainers keep implementation notes in .github/workflows/README.md. |
| Mecatl Studio web UI | operating/studio.md | Feature pages describe shared behavior; the Studio page owns the image, browser login, and its environment variables. The workspace's own apps/README.md covers local development modes. Keep external Studio guidance operator-focused; retain unique essential user instructions until product guidance covers them. |
| Provider and model selection | features/sessions/choose-models.md | features/sessions/model-routing.md owns aliases and task routing; operator guides own credential custody and server configuration. |
| Workspace trust | features/security-and-execution/permissions-and-posture.md | features/agent-behavior/project-instructions-and-rules.md describes the project-content consequence. |
| Skills, commands, soul, and user model | features/agent-behavior/skills-commands-and-soul.md and features/agent-behavior/memory.md | Extension-point pages document the importable interfaces. |
| Hooks | features/security-and-execution/hooks.md | building/go/extension-points/hook-runner.md owns host-integration details. |
| Server CLI flags | reference/server-cli.md | Generate from trusted help with task docs:server-cli; verify freshness with task docs:server-cli:check. Guides name only flags needed by their workflows. |
| Runtime observability and resilience | features/runtime/observability-and-resilience.md | operating/observability.md owns collection and exporter setup. |
| Configuration schema | reference/configuration.md | operating/settings.md explains which configuration plane to use. |
| gRPC API | reference/grpc-api.md | building/grpc-http.md explains client integration and transport choice. |
| HTTP and SSE API | reference/http-sse-api.md | building/grpc-http.md explains client integration and transport choice. |
| TypeScript SDK workflows | building/typescript-sdk/first-run.md and building/typescript-sdk/ | Deployment and feature pages link to the SDK guide that owns the application workflow. |
| TypeScript SDK method reference | reference/typescript-sdk-api/ | SDK guides link to the generated entry-point reference instead of copying signatures or option inventories. |
| Troubleshooting | The troubleshooting section nearest the affected workflow | Do not create a second catch-all list when the owning page can provide the remedy. |
When adding a page that appears to overlap an existing page, identify its
canonical page in the pull request description. If the two pages need the same
complete treatment, extend the canonical page instead. URL moves are separate
changes and must follow the published URL policy in
website/AGENTS.md.
Extend an existing page when it already answers the reader's question. Add a
page only for a distinct user task or concept that has no suitable home. New
section directories need an entry page and _category_.json so the
autogenerated sidebar has an entry point and label. Use index.md for a new
section; when splitting an existing guide into workflow pages, the category can
link to that guide as its entry point. Keep its published URL and avoid adding
a second overview that answers the same question.
Write a page
Use front matter on every documentation page. At minimum, provide a
sentence-case title and a reader-focused description; add sidebar_position
when the page's order matters. Add an explicit Markdown H1 immediately after the
front matter that matches title. Docusaurus uses the explicit H1 in the
rendered page and uses title for page metadata, while the H1 keeps the file
readable on GitHub.
Use sidebar_label only when a shorter sidebar label materially improves
navigation. It changes the autogenerated sidebar label without changing the page
title or H1.
---
title: Page title
description: State the reader outcome or topic in one sentence.
sidebar_position: 1
---
# Page titleWrite about current behavior in clear technical prose. Keep roadmap discussion explicitly separated in cloud-native-harness.md; public task guides and reference must describe delivered surfaces. Contributor qualification commands and test receipts belong in the owning internal documentation. Preserve verified operator limitations when trimming that material.
Use sentence-case
headings, descriptive links, and examples that help readers complete a task.
Style Mecatl as a proper noun, and write the component and binary names
mecatui, mecated, and mecak8s in code font. For deeper guidance, see
Diataxis and the
Google developer documentation style guide.
End a tutorial or task-oriented guide with ## Next steps when readers have a
clear next action. Reference, explanation, index, and narrow troubleshooting
pages do not need that section.
Terminology
Use lowercase, hyphenated cloud-native harness except at the start of a sentence, heading, or navigation label, where Cloud-native harness follows ordinary sentence case. Use cloud-native for ordinary adjectives, such as "cloud-native deployment" or "cloud-native workload." Use cloud-native agent harness as a generic description. Do not use "cloud-native kit" in public documentation.
Use content-root-relative Markdown paths for links across sections, including
the .md extension:
[Run mecated standalone](/operating/mecated.md)For a sibling in the same directory, use a relative .md path. Do not use a
/docs/ prefix. The link rules and their Docusaurus rationale live in
website/AGENTS.md.
Verify behavior
Start with the applicable implementation source:
docs/architecture/*.mdfor subsystem behaviorcontracts/proto/for gRPC messages and services- command flag registration, configuration schemas, handlers, and adapters for deployed behavior
Generated reference pages are exceptions to direct editing:
- For
reference/server-cli.md, change flag registration/help orscripts/server-cli-reference.py, then runtask docs:server-cli. Never edit the generated output by hand. - For
reference/configuration.md, change the configuration schema or generator, then runtask docs:configref. - For
reference/typescript-sdk-api/core.md,node.md, anddeno.md, change the TSDoc insdk/typescript/src/or the SDK documentation generator, then runtask sdk:docs. Usetask sdk:docs:checkto verify that the committed output is current without changing it.
Confirm availability and defaults against the shipped public surface: flags or
configuration, composition, capability advertisement, handlers, and deployed
adapters. A port, proto, or constructor alone does not prove that a feature
ships. Do not write from AGENTS.md or CLAUDE.md directly. They are internal
contracts with implementation detail, not user documentation.
For a feature page, lead with the outcome, then cover prerequisites, a short happy-path example, configuration or permissions, and relevant deployment limits. Report uncertainty when the available evidence does not establish a claim.
Preview and check
Run these commands from the repository root:
task site:dev # preview at http://localhost:3001
task site:build # production build and broken-link check
task site:format # prettier normalization pass (optional, not CI-enforced)task site:build is required for a documentation change. It fails on broken
links. task site:format is optional prose/list/table normalization; nothing in
CI checks formatting. It skips the generated reference pages listed above
(user-docs/.prettierignore) so it never fights their own generators. Do not
put working plans, authoring ledgers, or internal implementation queues in
user-docs/; keep them in issues or pull request descriptions.
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.

