fix(controller): honor agent URL for local workers

[Yusef Syed (YusefSyed)]

Summary

• resolve the local worker gateway from agl_server.agent_url, falling back to agl_server.url
• normalize the trailing slash consistently with the Kubernetes reconciler
• verify both injected worker endpoints for the override and fallback cases

Problem

The documented controller/agent URL split was only honored by build_job_spec(). Local workers always received endpoint variables rooted at the controller-facing URL, so a configured agent-reachable gateway was silently ignored.

This change applies the existing Kubernetes resolution rule to _spawn_for() while leaving the controller API client on agl_server.url.

Closes #581

Verification

• the override regression fails on current main by receiving http://controller:8080 instead of http://agent-gateway:8080
• uv run --extra dev pytest tests/controller/test_local_reconciler_env.py -q — 2 passed
• full test suite — 94 passed, 3 pre-existing dependency warnings
• pyright — 0 errors
• ruff check .
• ruff format --check .
• python scripts/check_headers.py
• uv build --no-sources

This is independent of #580: that PR changes shutdown admission, while this one changes only worker endpoint construction.

[Copilot]

🟢 Approval recommended

The focused implementation matches the established Kubernetes resolution rule and has regression coverage.

Pull request overview

Updates local workers to honor the agent-facing gateway URL, matching Kubernetes behavior.

Changes:

• Uses agl_server.agent_url with fallback to agl_server.url.
• Normalizes trailing slashes and tests both endpoint variables.

File summaries

File	Description
agentlightning/controller/local_reconciler.py	Resolves agent-facing local worker endpoints.
tests/controller/test_local_reconciler_env.py	Covers override and fallback endpoint generation.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 0
• Review effort level: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.