agent_docs/README.md

Rama main · réplica de lectura del repositorio original.

# Agent Documentation Directory

This directory contains detailed documentation for AI coding agents working on the ClickStack codebase. These files use **progressive disclosure** - they're referenced from `AGENTS.md` but only read when needed.

## Purpose

Instead of stuffing all instructions into `AGENTS.md` (which goes into every conversation), we keep detailed, task-specific information here. This ensures:

1. **Better focus**: Only relevant context gets loaded per task
2. **Improved performance**: Smaller context window = better instruction following
3. **Easier maintenance**: Update specific docs without bloating the main file

## Files

- **`architecture.md`** - System architecture, data models, service relationships, security patterns
- **`tech_stack.md`** - Technology choices, UI component patterns, library usage
- **`development.md`** - Development workflows, testing strategy, common tasks, debugging
- **`code_style.md`** - Code patterns and best practices. Read before adding a type, Zod schema, helper, or component in any package (grep-first rule, plus a map of what already lives in `common-utils`), and before any `packages/app` UI change.
- **`page_layout.md`** - PageHeader, PageLayout, and consistent page chrome (titles, actions, tool pages)
- **`naming_objects.md`** - How users name and rename objects (dashboards, saved searches, tiles) with `InlineNameInput`. Read before adding or changing any place where something gets a name.
- **`data_viz_colors.md`** - Chart, heatmap, and semantic status colors. Read before adding or changing any color in a chart, sparkline, heatmap, legend, or status pill.
- **`themes.md`** - ClickStack appearance and color mode (light/dark/system). Read before changing anything in `packages/app/src/theme/`, adding semantic CSS variables, or hard-coding a color, logo, or favicon.
- **`observability.md`** - Instrumentation standards (tracing, metrics, context) and the shared helpers. Read when adding or changing a feature.
- **`evals.md`** - MCP eval framework: dual-slot setup, running A/B comparisons between branches, interpreting results. Read before running evals or benchmarking MCP changes.

## Usage Pattern

When starting a task:
1. Agent reads `AGENTS.md` first (always included)
2. Agent determines which (if any) docs from this directory are relevant
3. Agent reads only the needed documentation
4. Agent proceeds with focused, relevant context

## Maintenance

- Keep files focused on their specific domain
- Use file/line references instead of code snippets when possible
- Update when patterns or architecture change
- Keep documentation current with the codebase