57 lines (37 loc) · 1.86 KB

Agent Harness: Veerify

This file is the agent entrypoint and map, not the full manual.

Startup

Read CLAUDE.md first.
Follow the chain to .agents/CLAUDE.md and treat it as the authoritative instruction set.
Run yarn harness:context to collect current repository and git context.

Knowledge Map

Authoritative project conventions: .agents/CLAUDE.md
Context and file map: docs/agent/context-map.md
Deterministic workflows and validation loop: docs/agent/workflows.md
Chrome MCP browser verification workflow: docs/agent/chrome-mcp.md
Active implementation plans and design notes: docs/plans/
CI truth for enforced checks: .github/workflows/ci.yml

Required Validation Gates

Run these after every change:

yarn typecheck
yarn test
yarn lint
yarn test:e2e:if-available

Or run the harness command:

yarn harness:verify

UI Legibility Runtime

Set up Chrome DevTools MCP: yarn codex:mcp:chrome:setup
Start app with deterministic worktree runtime: yarn dev:worktree
Inspect resolved worktree URL/port: yarn worktree:env

Commit Format

Use Conventional Commits: type(scope): subject (or type: subject when no scope is needed).
Preferred types: feat, fix, chore, docs, refactor, test, ci, build, perf, revert.

Playwright Guard

yarn test:e2e:if-available must run Playwright only when environment is cloud/CI or PLAYWRIGHT_FORCE=1, and a database is configured and reachable.
If the guarded Playwright command skips, report the skip reason in updates/final output.

UI Change Rule

Any user-facing UI behavior change requires Playwright coverage updates for the affected workflow.
Re-run Playwright until the updated UI workflow tests pass, not just once.

Maintenance Rule

Keep AGENTS.md short and stable.
Move detailed guidance into docs/agent/* and link from here.