Skip to content

CONSTRAINTS.md

Latest commit

 

History

History
44 lines (26 loc) · 2.48 KB

CONSTRAINTS.md

File metadata and controls

44 lines (26 loc) · 2.48 KB
Raw

Constraints

Constraints for this marketplace — what DVstack is allowed to ship. Not a product definition of done.

What exists

Only document plugins and skills that exist in the tree. Do not advertise empty plugins, future topics, or README placeholders.

Shipped plugins:

  • dvstack-frontend — testable-ui, font-switcher
  • dvstack-testing — twenty Playwright skills
  • dvstack-agents — using-dvstack, repo-recon, agent-ops-bar, verify-loop, agent-session, agent-rigor, anti-rationalization, doubt-check, source-check
  • dvstack-harness — harness-engineering, agent-execution-surfaces, agent-tool-design, safeguard-parity, agent-boundary-tests, agent-risk-chains

Skills

  • One narrow subject per skill.
  • SKILL.md frontmatter: name (same as the folder) and description (what it does and when to use it).
  • No attribution or source lists in public skills.
  • New skills should end with a ## Verification checklist. The Playwright skills use ## Verify plus ## Rules for the same purpose — follow the plugin's local convention rather than mixing both.
  • Claims about a framework API are checked against the installed version's official docs before they ship, per source-check. A wrong API in a skill is worse than no skill.

Agents vs harness

Agents stay light (habits). Harness owns heavy system concerns (contracts, tools, verify/recover, durable state, feature maps, hard CI, plus execution-surface / tool-design / safeguard-parity / boundary-test / risk-chain QA). Do not duplicate across that boundary — point instead.

Scope

Only subjects covered by real, first-hand experience: Playwright and test automation, the front-end side of making an app testable, and the system around coding agents. No back-end, database, SQL, or infrastructure skills.

Commands

Plugin commands/*.md are force-loaders for installed skills. They tell the agent which SKILL.md to Read and follow. They must not invent parallel rules.

Token-light

AGENTS.md is the single always-on file, and it stays short. Detail lives in skills that load when relevant.

Claude Code reads AGENTS.md as project instructions when a repository has no CLAUDE.md (v2.1.277+, on the default setting), and Cursor and other agents read it too — so one file covers every tool. Do not add a CLAUDE.md here: its presence makes Claude Code skip AGENTS.md entirely, and the two would drift.

Public-safe

This repo is public. No private themes, secrets, personal tokens, or unpublished product internals.