Skip to content

AGENTS.md

Latest commit

 

History

History
52 lines (37 loc) · 3.16 KB

AGENTS.md

File metadata and controls

52 lines (37 loc) · 3.16 KB
Raw

AGENTS

Scope

This file governs only the Workcell product source. It does not modify or replace the business rules of any external project.

Product principles

  • Remain platform-neutral. Platform files for Codex, Claude Code, and similar systems are adapters and must not become the sole source of truth.
  • Remain text-only, reviewable, and distributable or upgradable through Git.
  • The Bootstrap Skill is responsible only for creation, validation, repair, and upgrade. Daily runtime capabilities belong to the Skills in the expanded workspace.
  • Structured manifests, schemas, facts, and policies are sources of truth. Files such as AGENTS are generated AI adapters.
  • Environment, project, and diagnostic runtime facts must be established by initialization and must not be inferred directly from historical documentation.
  • Capability does not imply authorization. Diagnostic and external actions must satisfy environment capability, Provider support, project policy, and task authorization.
  • Default to read-only and fail closed. Installing tools, modifying the environment, accessing remote systems, overwriting user content, and applying upgrades require explicit authorization.

Language convention

  • Canonical product documentation, specifications, operator messages, metadata, code comments, and error messages use English.
  • Skill names, directory names, file names, schema fields, state enums, commands, and stable IDs use English ASCII.
  • Localized content is generated only through an explicit localization workflow. It must remain separate from the canonical English control plane and must not change machine-readable identifiers.

File ownership

  • distribution: maintained by the release and replaceable through a Migration.
  • generated: compiled from structured facts and not maintained manually.
  • workspace: maintained by the workspace, project, or user and preserved by upgrades.

Every new template or specification must declare its ownership.

Skill rules

  • Every Skill must contain a valid SKILL.md; names use lowercase hyphen-case.
  • Keep SKILL.md concise and place detailed specifications in references/.
  • Put repeatable, fragile, or consistency-sensitive behavior in scripts/ whenever practical.
  • Put files intended for expansion in assets/.
  • Do not add README files, installation guides, or duplicated explanations inside a Skill.
  • Every Skill must pass Skill Creator's quick_validate.py.

Changes and validation

  • Limit all modifications to this repository.
  • When modifying scripts, add or update the corresponding tests.
  • When modifying schemas, run JSON parsing and example validation.
  • When modifying templates, run the Bootstrap tests and Workspace Doctor.
  • Do not commit, push, or publish automatically.

Current security boundary

  • The V0.1 Bootstrap does not access the network, install software, or modify the shell. It writes only to an explicitly specified empty directory.
  • When V0.1 diagnostic capabilities are unconfigured, all remote diagnostic actions are denied by default.
  • Self-update defines a protocol only. Do not apply updates automatically until there is a unique repository, formal releases, and signature or hash verification.