OpenWriterControle/evals/agents/meta/repo-manager/README.md
Darren Hinde bf02ad0d5c
refactor: Rename repository to OpenAgents Control (#100)
* refactor: Rename repository to OpenAgents Control

Update all project references from "OpenAgents" to "OpenAgents Control":
- Updated core config files (package.json, registry.json)
- Updated documentation (README.md, docs/)
- Updated scripts (install.sh, update.sh, test scripts)
- Updated agent and command files
- Updated integrations and packages
- Updated all GitHub URLs to use OpenAgentsControl

Naming convention:
- Display name: "OpenAgents Control"
- Repository URL: OpenAgentsControl

* feat: Enable subagent-to-subagent delegation (ContextScout integration)

Allow coding-related subagents to call ContextScout for context discovery:
- CoderAgent: Can now discover coding standards on-demand
- TestEngineer: Can now discover test coverage and TDD guidelines
- CodeReviewer: Can now discover code review and security patterns
- BuildAgent: Can now discover build validation standards
- DocWriter: Can now discover documentation formatting guidelines

Changes per subagent:
- Added `task: true` to tools section
- Added permission allowlist for ContextScout only (deny all other delegation)
- Added Context Discovery workflow section with when-to-call guidance

This enables subagents to be self-sufficient for finding relevant standards
instead of relying on orchestrator to provide all context files.

Benefits:
- Subagents can discover what they need when context isn't explicit
- Maintains security (only ContextScout allowed via task tool)
- Follows existing patterns (uses task tool like main agents)

* fix: Set OPENCODE_BRANCH in CI tests to use PR branch for downloads

- Add OPENCODE_BRANCH environment variable to E2E tests
- Add OPENCODE_BRANCH environment variable to profile smoke tests
- Ensures tests download files from the PR branch instead of main
- Fixes E2E test failures after repository rename to OpenAgentsControl
2026-01-18 09:31:35 +00:00

2.2 KiB

Repository Manager Agent - Eval Tests

Agent: meta/repo-manager
Version: 1.0.0
Category: Meta
Type: Meta Agent


Overview

Eval tests for the Repository Manager meta agent. This agent orchestrates OpenAgents Control repository development with context-aware planning, task breakdown, and automatic documentation.

Test Suites

Smoke Test (smoke-test.yaml)

Basic functionality verification:

  • Context loading for informational queries
  • Approval gate enforcement
  • Context awareness for repo-specific questions

Run:

cd evals/framework
npm run eval:sdk -- --agent=meta/repo-manager --pattern="smoke-test.yaml"

Context Loading Test (context-loading-test.yaml)

Verifies appropriate context loading for different task types:

  • Agent creation tasks
  • Eval testing tasks
  • Registry management tasks

Run:

cd evals/framework
npm run eval:sdk -- --agent=meta/repo-manager --pattern="context-loading-test.yaml"

Delegation Test (delegation-test.yaml)

Verifies delegation decisions and context bundle creation:

  • Complex features → delegate to task-manager
  • Simple tasks → execute directly
  • Documentation → delegate to documentation subagent

Run:

cd evals/framework
npm run eval:sdk -- --agent=meta/repo-manager --pattern="delegation-test.yaml"

Running All Tests

cd evals/framework
npm run eval:sdk -- --agent=meta/repo-manager

Test Coverage

Context Loading: ✅ Covered
Approval Gates: ✅ Covered
Delegation Logic: ✅ Covered
Context Bundles: ✅ Covered
Validation: ⚠️ Partial (needs integration tests)
Documentation Updates: ⚠️ Partial (needs integration tests)

Known Limitations

  • Integration tests for full workflow not yet implemented
  • Validation script execution not tested in isolation
  • Documentation update detection needs more test cases

Future Test Cases

  • Full agent creation workflow (end-to-end)
  • Validation failure handling
  • Documentation update detection
  • Session cleanup verification
  • Multi-step task coordination
  • Error recovery scenarios

Last Updated: 2025-01-21
Maintainer: opencode