Support common AI patterns

[sdirix]

Use Cases

Feature: Common AI ecosystem conventions work out of the box

  Scenario: AGENTS.md is respected
    Given a workspace contains an AGENTS.md file
    When an agent session runs in this workspace
    Then the AGENTS.md content is part of the agent's context
    And the user can inspect that it was applied

  Scenario: Common configuration files are honored
    Given a workspace contains common AI configuration files (e.g. mcp.json, rules)
    Then they are discovered and applied where supported
    And unsupported files are at least surfaced to the user

  Scenario: Workspace trust is respected
    Given an untrusted workspace contains AI convention files
    Then they are not applied until the workspace is trusted

Description

The AI ecosystem has converged on convention files: AGENTS.md, rules directories (e.g. .cursor/rules), mcp.json, prompt/skill folders. Theia references some of these inside system prompts but does not auto-load them as first-class project context. Users coming from other tools expect their existing files to just work.

This is a generic ticket: inventory the common patterns, decide which Theia supports natively, and build the shared mechanics. Further patterns become follow-up issues on top of the same mechanics.

Suggested architecture / Implementation Details

• Inventory: AGENTS.md, CLAUDE.md, .cursor/rules/.cursorrules, mcp.json, prompt/skill folders; document semantics and precedence per tool.
• Discovery: a workspace-scanning service (with file watching) that reports found conventions.
• Application: integration points per pattern, system prompt injection via the prompt/variable services for context files, import into Theia's MCP configuration for mcp.json.
• Inspection: UI to see which conventions were found and applied to a session (trust and transparency).
• Gate everything behind workspace trust and keep it configurable/removable for adopters.

Expected results

• The pattern inventory with a support decision per pattern.
• Discovery mechanics plus AGENTS.md applied end-to-end, inspectable in the UI, trust-gated.
• Follow-up issues for the remaining patterns.