kevinslin/agstack/docy

Docy

Overview

Keep policy and reference docs small, named, and injectable. For coding-related tasks, start by loading the baseline rules from ./references/core/, then return to this skill only when the task needs additional targeted guidance. Use the bundled docy CLI to print a doc into the active context with inject or to install a durable managed rule block into AGENTS.md with install.

Doc Layout

All managed docs live under ./references/ and follow this layout:

./references/
# language specific
- lang/
# always be injected
- core/
# frameworks/dependencies
- vendor/
# special topics
- ref/
    - commit-messages.md: Repo-aware commit message drafting; inspect local
      history and mirror conventions before writing a subject line
    - developer-docs.md: Developer documentation writing, editing, and review
      guidance, including page patterns and an editorial checklist
    - concise-instructions.md: Terse operator-facing instructions organized
      around inputs, invocation, completion evidence, and recovery
    - spec.md: Architecture and design specification writing, examples, and
      review guidance
    - openclaw-docs.md: OpenClaw-specific page taxonomy, heading, docs-index,
      editing, and verification conventions layered on developer-docs
    - execution-trace.md: How to write runtime-ordered execution trace docs
    - no-back-compat.md: Hard-cut product policy; no backwards compatibility
    - remove-feature.md: Feature removal hygiene; remove stale docs/tests and
      record spec drift in changelog

Loading Rules

For coding-related tasks, use this loading order:

1. At startup, inject every doc under ./references/core/. Treat these as the

default rule set, not an optional filter.

1. After the core docs are loaded, add non-core docs selectively based on the task.
2. Inject ./references/lang/<language>.md only when the task depends on

language-specific rules or idioms.

1. Inject ./references/vendor/<dependency-or-framework>.md only when the task

depends on framework, library, or platform behavior.

1. Inject ./references/ref/<topic>.md only for focused policies, constraints, or

one-off topics that should shape the solution. For example, inject ./references/ref/remove-feature.md before removing an existing feature.

CLI

Run the bundled CLI directly or put scripts/ on PATH.

docy inject core/main
docy inject ref/commit-messages
docy inject ref/developer-docs
docy inject ref/concise-instructions
docy inject ref/spec
docy inject ref/openclaw-docs
docy inject ref/no-back-compat
docy inject ref/remove-feature
docy inject ref/openclaw-agent-plugins
docy inject ref/execution-trace
docy inject vendor/lerna
 
docy install core/main
docy install ref/commit-messages
docy install ref/developer-docs
docy install ref/concise-instructions
docy install ref/spec
docy install ref/openclaw-docs
docy install ref/no-back-compat
docy install ref/remove-feature
docy install ref/openclaw-agent-plugins
docy install ref/execution-trace
docy install vendor/lerna

Command behavior:

• inject: Discover Markdown files under ./references/ and print the selected

doc to stdout for immediate context injection.

• install: Discover the same reference files and add or update the selected

doc as a managed block in the nearest AGENTS.md so the rule remains durable for later sessions.

Available Docs

• ./references/core/main.md: Core documentation and technical-writing hygiene.

Always load it to keep adjacent durable docs synchronized and apply universal rules for terminology, scope, examples, reviews, and diagrams.

• ./references/vendor/lerna.md: Modern Lerna operating guidance for agents. Use before changing, validating, or releasing code in a Lerna-managed monorepo.
• ./references/ref/commit-messages.md: Repo-aware commit message drafting. Use

before writing a commit message so the subject mirrors recent local history instead of forcing one universal style.

• ./references/ref/developer-docs.md: Developer documentation guidance. Use

before creating, editing, or reviewing READMEs, quickstarts, guides, API, SDK, CLI, testing, troubleshooting, architecture, or internals documentation.

• ./references/ref/concise-instructions.md: Concise instruction-document

guidance. Load it after ref/developer-docs when writing or compressing an operator-facing runbook, setup guide, or command procedure.

• ./references/ref/spec.md: Design specification guidance. Use before creating,

editing, or reviewing architecture and design specifications so scope, ownership, interfaces, examples, failure paths, and implementation detail stay decision-complete without becoming implementation-exhaustive.

• ./references/ref/openclaw-docs.md: OpenClaw developer documentation overlay.

Load it after ref/developer-docs when creating, editing, or reviewing OpenClaw topic pages, guides, references, or troubleshooting documentation.

• ./references/ref/python-preferred-modules.md: Python dependency preferences. Use before building runtime validation or CLI behavior from scratch in Python projects.
• ./references/ref/typescript-preferred-modules.md: TypeScript dependency preferences. Use before building runtime validation or CLI behavior from scratch in TypeScript projects.
• ./references/ref/openclaw-agent-plugins.md: OpenClaw plugin authoring guidance. Use before creating or expanding an OpenClaw plugin so capability ownership, entrypoint shape, SDK imports, and route/setup boundaries stay aligned with the architecture docs.
• ./references/ref/execution-trace.md: Execution trace writing guidance. Use before documenting initialization, startup, request, job, command, or other runtime flows where ordered control flow matters.
• ./references/ref/no-back-compat.md: Hard-cut product policy. Use before changing

codepaths that would otherwise introduce migrations, fallback behavior, adapters, or other backwards-compatibility glue.

• ./references/ref/remove-feature.md: Feature removal hygiene. Use before deleting

or sunsetting an existing feature so docs, tests, and changelog entries stay consistent while historical specs remain unchanged.