copilot: Route ug copilot through a refreshing loopback auth proxy

.github/workflows/integration.yml

@@ -30,6 +30,10 @@ on:
         description: Exact OpenCode version (non-blocking lane)
         default: 1.18.31
         required: true
+      copilot_version:
+        description: Exact GitHub Copilot CLI version (non-blocking lane)
+        default: 1.0.91
+        required: true
       dependency:
         description: Optional exact dependency for reproduction (e.g. tomlkit==0.14.0)
         default: ''
@@ -53,6 +57,7 @@ env:
   CLAUDE_VERSION: ${{ inputs.claude_version || '2.1.290' }}
   CODEX_VERSION: ${{ inputs.codex_version || '0.154.0' }}
   OPENCODE_VERSION: ${{ inputs.opencode_version || '1.18.31' }}
+  COPILOT_VERSION: ${{ inputs.copilot_version || '1.0.91' }}
   PACKAGE_INDEX: ${{ inputs.default_index || 'https://pypi.org/simple' }}
 
 jobs:
@@ -211,6 +216,7 @@ jobs:
             claude) args=(--claude-version "$CLAUDE_VERSION") ;;
             codex) args=(--codex-version "$CODEX_VERSION") ;;
             opencode) args=(--opencode-version "$OPENCODE_VERSION") ;;
+            copilot) args=(--copilot-version "$COPILOT_VERSION") ;;
           esac
           if [[ -n "$DEPENDENCY" ]]; then
             args+=(--dependency "$DEPENDENCY")
@@ -351,6 +357,28 @@ jobs:
       ARTIFACT_NAME: integration-full-opencode-${{ matrix.os }}
     steps: *live-steps
 
+  # Non-blocking pending stabilization, like OpenCode: continue-on-error and absent from the
+  # cujs `needs`.
+  copilot:
+    name: Full journeys · Copilot · ${{ matrix.os == 'windows-latest' && 'Windows' || 'Linux' }} (non-blocking)
+    if: ${{ !cancelled() && needs.workspace.result == 'success' }}
+    needs: workspace
+    continue-on-error: true
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-22.04, windows-latest]
+    timeout-minutes: 30
+    env:
+      UCODE_TEST_WORKSPACE: ${{ secrets.UCODE_TEST_WORKSPACE }}
+      DATABRICKS_BEARER: ${{ secrets.DATABRICKS_BEARER }}
+      DEPENDENCY: ${{ inputs.dependency }}
+      AGENT: copilot
+      TEST_MARKER: live and copilot
+      ARTIFACT_NAME: integration-full-copilot-${{ matrix.os }}
+    steps: *live-steps
+
   # Non-blocking pending stabilization: continue-on-error and absent from the cujs `needs`.
   # Runs on the managed e2e workspace with an injected config that enables Claude/Codex but not
   # OpenCode (a service principal authenticates it, like the `managed` lane).

README.md

@@ -75,6 +75,13 @@ because it overrides the model sent to the gateway. Restart Copilot through `ug`
 to change the wire model or API; in-session model selection does not rebuild its
 provider configuration.
 
+`ug copilot` runs Copilot against a loopback proxy for the session. Copilot holds only
+a random session secret; the proxy adds a fresh Databricks token to each gateway
+request, so OAuth sessions keep working past token expiry without a restart, and no
+token is written to `~/.copilot/ucode.env`. If the Databricks sign-in itself expires,
+Copilot shows a "Databricks sign-in expired" error; run `databricks auth login` and
+restart `ug copilot`. Proxy errors are logged to `~/.ucode/logs/copilot-proxy.log`.
+
 Without a managed workspace config, `ug claude` automatically discovers gateway
 models for Claude Code's `/model` picker. Discovery defaults to `system.ai` when
 no provider or model location is selected. Use `--provider` or `--model-location`

scripts/run_integration.py

@@ -30,6 +30,7 @@
     "claude": "@anthropic-ai/claude-code",
     "codex": "@openai/codex",
     "opencode": "opencode-ai",
+    "copilot": "@github/copilot",
 }
 WINDOWS_PATHEXT = ".COM;.EXE;.BAT;.CMD"
 # Claude exports no spans on Windows, where ug writes no machine-wide Claude settings.
@@ -44,9 +45,18 @@
     "claude": "test_ug_claude_headless.py::test_ug_claude_headless_prompt_argument",
     "codex": "test_ug_codex_headless.py::test_ug_codex_headless_prompt_argument",
     "opencode": "test_ug_opencode_headless.py::test_ug_opencode_headless_prompt_argument",
+    "copilot": "test_ug_copilot_headless.py::test_ug_copilot_headless_completes_file_task",
 }
 
 
+def agent_version_output(agent: str, output: str) -> str:
+    """The agent's ``--version`` text, normalized for exact-version matching."""
+    if agent == "copilot":
+        # "GitHub Copilot CLI 1.0.91." plus an update hint on later lines.
+        return output.splitlines()[0].removesuffix(".") if output else output
+    return output
+
+
 def redact_secrets(value: str, secrets: Iterable[str]) -> str:
     for secret in sorted({secret for secret in secrets if secret}, key=len, reverse=True):
         value = value.replace(secret, "<redacted>")
@@ -253,11 +263,13 @@ def arguments(
     parser.add_argument("--claude-version", type=exact_npm_version)
     parser.add_argument("--codex-version", type=exact_npm_version)
     parser.add_argument("--opencode-version", type=exact_npm_version)
+    parser.add_argument("--copilot-version", type=exact_npm_version)
     parser.add_argument("--claude-model", default=environment.get("UG_INTEGRATION_CLAUDE_MODEL"))
     parser.add_argument("--codex-model", default=environment.get("UG_INTEGRATION_CODEX_MODEL"))
     parser.add_argument(
         "--opencode-model", default=environment.get("UG_INTEGRATION_OPENCODE_MODEL")
     )
+    parser.add_argument("--copilot-model", default=environment.get("UG_INTEGRATION_COPILOT_MODEL"))
     parser.add_argument(
         "--claude-provider",
         default="main.ucode.ci_e2e_anthropic_nonrelay_mps",
@@ -356,9 +368,12 @@ def arguments(
             args.pytest_args.extend([flag, str(value)])
     if selected.x:
         args.pytest_args.append("-x")
-    if not (args.claude_version or args.codex_version or args.opencode_version):
+    if not (
+        args.claude_version or args.codex_version or args.opencode_version or args.copilot_version
+    ):
         parser.error(
-            "Select --claude-version, --codex-version and/or --opencode-version explicitly."
+            "Select --claude-version, --codex-version, --opencode-version and/or "
+            "--copilot-version explicitly."
         )
     if args.ug_version != "checkout" and not re.fullmatch(
         r"[0-9][0-9A-Za-z.!+_-]*", args.ug_version
@@ -516,9 +531,11 @@ def run(command, *, cwd=output, env=base_env, timeout=600) -> str:
             "claude": args.claude_version,
             "codex": args.codex_version,
             "opencode": args.opencode_version,
+            "copilot": args.copilot_version,
             "claude_model": args.claude_model,
             "codex_model": args.codex_model,
             "opencode_model": args.opencode_model,
+            "copilot_model": args.copilot_model,
             "claude_provider": args.claude_provider,
             "claude_relayed_provider": args.claude_relayed_provider,
             "claude_provider_model": args.claude_provider_model,
@@ -731,7 +748,9 @@ def run(command, *, cwd=output, env=base_env, timeout=600) -> str:
         report["agents"] = {}
         for agent in agents:
             agent_command = npm_executable(agent_bin, agent)
-            version = run([agent_command, "--version"], env=runtime_env, timeout=30)
+            version = agent_version_output(
+                agent, run([agent_command, "--version"], env=runtime_env, timeout=30)
+            )
             expected = getattr(args, f"{agent}_version")
             if not re.search(rf"(?<![\w.]){re.escape(expected)}(?![\w.])", version):
                 raise RuntimeError(f"Expected {agent} {expected}, got {version!r}")
@@ -877,10 +896,13 @@ def run(command, *, cwd=output, env=base_env, timeout=600) -> str:
             raise RuntimeError("Pytest returned success without a test report.")
         # A bootstrap/update path must not silently alter the selected agent version.
         for agent in agents:
-            after = run(
-                [npm_executable(agent_bin, agent), "--version"],
-                env=runtime_env,
-                timeout=30,
+            after = agent_version_output(
+                agent,
+                run(
+                    [npm_executable(agent_bin, agent), "--version"],
+                    env=runtime_env,
+                    timeout=30,
+                ),
             )
             if after != report["agents"][agent]:
                 raise RuntimeError(f"{agent} changed version during the suite: {after}")

src/ucode/agents/__init__.py

@@ -465,7 +465,7 @@ def configure_tool(
             result = kilo.write_tool_config(state, model)
         else:
             result = opencode.write_tool_config(state, model)
-    # gemini/opencode/copilot/pi return (state, token); codex/claude return state
+    # gemini/opencode/kilo/pi return (state, token); codex/claude/copilot return state
     if isinstance(result, tuple):
         return result[0]
     return result

src/ucode/agents/codex.py

@@ -50,15 +50,14 @@
     CUSTOM_OAUTH_TIMEOUT_MS,
     CustomOAuthConfig,
     build_custom_auth_token_argv,
-    get_custom_client_token,
+    launch_token,
 )
 from ucode.databricks import (
     CodexCatalogSource,
     CodexMpsModelCatalogUnavailable,
     _fetch_codex_model_catalog,
     build_auth_token_argv,
     build_tool_base_url,
-    get_databricks_token,
     ug_binary,
 )
 from ucode.launcher import exec_or_spawn
@@ -1181,17 +1180,7 @@ def _launch_token(state: dict, workspace: str, *, force_refresh: bool = False) -
     """The token Codex authenticates with: a custom-OAuth client token when configured,
     else the CLI profile. The single auth-selection point — the OTLP proxy's token
     provider (_otel_token_provider) delegates here so both mint the same principal."""
-    custom_oauth = state.get("custom_oauth")
-    if isinstance(custom_oauth, dict):
-        return get_custom_client_token(
-            workspace,
-            custom_oauth["client_id"],
-            custom_oauth["redirect_url"],
-            scopes=custom_oauth["scopes"],
-            profile=custom_oauth.get("profile"),
-            force_refresh=force_refresh,
-        )
-    return get_databricks_token(workspace, state.get("profile"), force_refresh=force_refresh)
+    return launch_token(state, workspace, force_refresh=force_refresh)
 
 
 def _tool_args_select_model(tool_args: list[str]) -> bool: