feat: extract agent startup into reusable NewAgent for test framework

[dejanzele]

Pull request description

Extracts runner/service/runnerClient creation from main.go into runner.NewAgent() so the control plane test framework can spin up real agents against a test gRPC server without duplicating startup logic.

Checklist (choose whats happened)

• breaking change! (describe)
• tested locally
• tested on cluster
• added new dependencies
• updated the docs
• added a test

Breaking changes

Changes

Fixes

[olensmar]

@greptile

[greptile-apps]

Greptile Summary

This PR extracts agent startup logic from cmd/api-server/main.go into a reusable runner.NewAgent() factory, enabling the control-plane test framework to spin up real agents against a test gRPC server without code duplication. The refactoring is well-structured and functionally safe for the primary use case (main.go).

Key changes:

• pkg/runner/start.go (new): Introduces NewAgent(), AgentConfig, AgentCredentials, AgentIdentity, and helper factories
• pkg/runner/agent.go: Exports NewAgentLoop (was newAgentLoop) for external testability
• Call sites updated: service.go, agent_conn_test.go updated to use exported name
• cmd/api-server/main.go: Replaces inlined construction with runner2.NewAgent(AgentConfig{...})

Design note: The AgentCredentials struct used by the test framework path lacks OrgSlug and EnvSlug fields that are present in ProContext and used in execution records. Test agents will record executions with empty slug fields (falling back to IDs).

Confidence Score: 4/5

• Safe to merge. The primary use case (main.go path) is functionally equivalent to the original code with no new risks. The test framework path works but has missing slug fields in execution records.
• The refactoring is sound: main.go behavior is unchanged, call sites are properly updated, and the new external API works for its intended purpose. The score is 4/5 rather than 5/5 because the AgentCredentials path used by test frameworks omits OrgSlug and EnvSlug fields that affect execution record metadata, though the system includes fallbacks to use IDs instead. This is a design choice worth documenting but not a blocker.
• pkg/runner/start.go — Consider whether AgentCredentials should expose slug fields or be clearly documented as intentionally omitting them.

_{Last reviewed commit: 03170c5}

[greptile-apps]

_{5 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

AgentCredentials struct lacks OrgSlug and EnvSlug fields. When buildProContext() constructs a ProContext from AgentCredentials, these fields remain empty strings. However, they are used downstream in execution records — see agent.go lines 355-356:

OrganizationSlug: a.proContext.OrgSlug,
EnvironmentSlug:  a.proContext.GetEnvSlug(environmentId),

When test frameworks use the AgentCredentials path, execution records will have empty slug fields (falling back to IDs via runner.go's fallback logic).

Consider:

1. Adding OrgSlug and EnvSlug to AgentCredentials so test agents can populate them, or
2. Documenting explicitly that these fields are intentionally omitted for the test scenario

This affects observability and any slug-based tracking in execution records.