docs: complete tool reference, document MCP resources, expand risk model

[summerstateofmind]

Summary

• Complete tool reference: Added detailed sections for the five tools that were listed in the overview but missing from the Tool Reference (set_flag_rollout, get_flag_state, toggle_flag_environment, remove_flag_strategy, cleanup_flag). Each follows the same structure as the existing four (parameters, when to use, agent prompt, tool payload, tool output).
• MCP resources: New section documenting the three registered resource URIs (unleash://projects, .../feature-flags, .../feature-flags/{flagName}), pagination parameters, caching behavior, and an example read.
• Risk model expansion: Extended the risk assessment section with scoring thresholds, the excluded-file category, confidence score semantics, and a source link to riskPatterns.ts.
• Architecture tree, endpoints, contributing: Updated to reflect the current codebase (9 tools, resources directory, dual-mode entry points, all Admin API endpoints used).

Motivation

The README documents the core workflow (evaluate → detect → create → wrap) well. The flag-management tools, MCP resources, and risk-model internals were present in code but not in docs, which limits how effectively agents and integrators can use the full surface.

Test plan

• Verify all parameter descriptions match the Zod schemas in src/tools/*.ts
• Confirm resource URIs and cache TTLs match src/resources/unleashResources.ts
• Confirm risk thresholds match calculateRiskLevel() in src/evaluation/riskPatterns.ts
• Confirm architecture tree matches src/ file listing
• Review rendered markdown for formatting issues

🤖 Generated with Claude Code