Merge assign-to-agent safe output handler into the safe output main handler loop

[Copilot]

Integrates the assign-to-agent safe output handler into the main process_safe_outputs handler manager loop, removing the dedicated standalone step.

Changes

JavaScript

• assign_to_agent.cjs: Complete rewrite as a handler factory pattern. main(config) returns a per-message handler function. Creates a dedicated Octokit instance from GH_AW_ASSIGN_TO_AGENT_TOKEN env var (agent token chain) or per-handler github-token. Resolves temporary IDs via the handler manager's live temporaryIdMap. Exports getAssignToAgentAssigned/Errors/ErrorCount/Summary for the handler manager.
• assign_agent_helpers.cjs: Added optional githubClient parameter (defaulting to global github) to all GraphQL-calling functions. Fully backward compatible.
• safe_output_handler_manager.cjs: Added assign_to_agent to HANDLER_MAP, removed from STANDALONE_STEP_TYPES, exports assign_to_agent outputs (assign_to_agent_assigned, assign_to_agent_assignment_errors, assign_to_agent_assignment_error_count) and calls summary writer after processing.

Go compiler

• compiler_safe_outputs_job.go: Added assign_to_agent to hasHandlerManagerTypes. Removed dedicated standalone step. Updated outputs to come from steps.process_safe_outputs.outputs.*.
• compiler_safe_outputs_steps.go: Adds GH_AW_ASSIGN_TO_AGENT_TOKEN env var to the process_safe_outputs step when assign-to-agent is configured (uses agent token preference chain: GH_AW_AGENT_TOKEN || GH_AW_GITHUB_TOKEN || GITHUB_TOKEN).
• compiler_safe_outputs_core.go: hasCustomTokenSafeOutputs returns true for assign_to_agent, triggering @actions/github installation via safe-output-custom-tokens: true.
• compiler_safe_outputs_specialized.go: Removed buildAssignToAgentStepConfig (no longer needed).
• js.go: Removed now-unused getAssignToAgentScript function.

Workflows

• smoke-agent-*.md: All five smoke-agent-* workflows (smoke-agent-all-merged, smoke-agent-all-none, smoke-agent-public-approved, smoke-agent-public-none, smoke-agent-scoped-approved) updated from engine: codex to engine: claude.

Key behaviors preserved

• Token: dedicated Octokit uses GH_AW_ASSIGN_TO_AGENT_TOKEN (not the step-level App token) — Copilot assignment API rejects App tokens
• Temporary IDs: resolved from handler manager's live temporaryIdMap, with deferred processing when not yet resolved
• 10-second delay between consecutive assignments
• Failure comments posted on issue/PR via the dedicated Octokit
• All existing outputs (assign_to_agent_assignment_error_count etc.) preserved for conclusion job

[pelikhan]

@copilot convert this smoke-agent-* agentic workflows to use claude

[Copilot]

@copilot convert this smoke-agent-* agentic workflows to use claude

Done — all 5 smoke-agent-* workflows updated from engine: codex to engine: claude and recompiled (commit d598748).

[Copilot]

Pull request overview

This PR integrates the assign_to_agent safe-output handler into the main process_safe_outputs handler-manager loop, removing the dedicated standalone “Assign to Agent” step and updating compiler outputs/env wiring accordingly. It also updates smoke-agent workflows to use the claude engine instead of codex.

Changes:

• Refactors assign_to_agent.cjs into a handler-factory invoked by the handler manager, with dedicated Octokit auth via GH_AW_ASSIGN_TO_AGENT_TOKEN.
• Updates Go workflow compiler to route assign-to-agent through process_safe_outputs, export outputs from that step, and inject the new env var / custom-token setup.
• Regenerates multiple workflow lockfiles (including smoke-agent workflows switching engine: claude).

Show a summary per file

File	Description
pkg/workflow/safe_outputs_integration_test.go	Updates integration expectations to use process_safe_outputs instead of a standalone assign_to_agent step.
pkg/workflow/js.go	Removes now-unused getAssignToAgentScript() getter.
pkg/workflow/compiler_safe_outputs_steps.go	Injects GH_AW_ASSIGN_TO_AGENT_TOKEN into the handler-manager step env when assign-to-agent is configured.
pkg/workflow/compiler_safe_outputs_specialized.go	Deletes the specialized standalone-step builder for assign-to-agent.
pkg/workflow/compiler_safe_outputs_job.go	Routes assign-to-agent through handler manager, exports outputs from steps.process_safe_outputs.outputs.*, removes standalone step emission.
pkg/workflow/compiler_safe_outputs_job_test.go	Updates tests to validate process_safe_outputs contains GH_AW_ASSIGN_TO_AGENT_TOKEN and no dedicated step exists.
pkg/workflow/compiler_safe_outputs_core.go	Ensures assign-to-agent triggers safe-output-custom-tokens setup (for @actions/github).
pkg/workflow/assign_to_agent_test.go	Updates workflow compile test to assert handler-manager integration and exported outputs.
actions/setup/js/safe_output_handler_manager.cjs	Adds assign_to_agent to HANDLER_MAP, removes it from standalone types, exports assign-to-agent outputs and writes its summary.
actions/setup/js/assign_to_agent.cjs	Rewrites assign-to-agent as a handler factory with dedicated Octokit auth, temp-ID deferral, per-message processing, and exported getters/summary writer.
actions/setup/js/assign_agent_helpers.cjs	Adds optional githubClient parameter to GraphQL helpers for dedicated Octokit usage.
.github/workflows/workflow-generator.lock.yml	Switches assign-to-agent outputs to come from process_safe_outputs, enables safe-output-custom-tokens, adds GH_AW_ASSIGN_TO_AGENT_TOKEN, removes standalone step.
.github/workflows/smoke-agent-scoped-approved.md	Changes workflow frontmatter engine from codex to claude.
.github/workflows/smoke-agent-scoped-approved.lock.yml	Regenerates lockfile for claude engine and related runtime/tooling changes.
.github/workflows/smoke-agent-public-none.md	Changes workflow frontmatter engine from codex to claude.
.github/workflows/smoke-agent-public-none.lock.yml	Regenerates lockfile for claude engine and related runtime/tooling changes.
.github/workflows/smoke-agent-public-approved.md	Changes workflow frontmatter engine from codex to claude.
.github/workflows/smoke-agent-public-approved.lock.yml	Regenerates lockfile; routes assign-to-agent through handler manager and enables custom-token install.
.github/workflows/smoke-agent-all-none.md	Changes workflow frontmatter engine from codex to claude.
.github/workflows/smoke-agent-all-none.lock.yml	Regenerates lockfile for claude engine and related runtime/tooling changes.
.github/workflows/smoke-agent-all-merged.md	Changes workflow frontmatter engine from codex to claude.
.github/workflows/smoke-agent-all-merged.lock.yml	Regenerates lockfile for claude engine and related runtime/tooling changes.
.github/workflows/issue-monster.lock.yml	Switches assign-to-agent outputs to process_safe_outputs, enables safe-output-custom-tokens, adds GH_AW_ASSIGN_TO_AGENT_TOKEN, removes standalone step.

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Comments suppressed due to low confidence (1)

actions/setup/js/assign_to_agent.cjs:487

• The step summary labels all skipped results as "(ignore-if-error enabled)" and "assignment failed due to error", but skips can now also occur for reasons like maxCount being reached (and potentially other future skip reasons). This produces misleading summaries. Track a skip reason (e.g. skipped_reason) and render different messaging for ignore-if-error vs max-count vs other skips.

  if (skippedResults.length > 0) {
    summaryContent += `⏭️ Skipped ${skippedResults.length} agent assignment(s) (ignore-if-error enabled):\n\n`;
    summaryContent += skippedResults
      .map(r => {
        const itemType = r.issue_number ? `Issue #${r.issue_number}` : `Pull Request #${r.pull_number}`;
        return `- ${itemType} → Agent: ${r.agent} (assignment failed due to error)`;
      })

• Files reviewed: 23/23 changed files
• Comments generated: 2

[Copilot]

main() is now a handler factory that returns a per-message function, but the repo still has JS unit tests (e.g. assign_to_agent.test.cjs) that call main() expecting it to read GH_AW_AGENT_OUTPUT and execute immediately. As written, those tests will fail (and any remaining standalone callers would break). Update the tests to invoke the returned handler with messages/temporaryIdMap, or add a small backwards-compatible wrapper entrypoint for standalone execution if that mode is still supported.

[Copilot]

When maxCount is exceeded, the handler returns { success: false, skipped: true } without an error/reason. The handler manager will log a generic message ("Handler returned success: false with skipped: true"), which makes it hard to understand why the assignment was skipped. Consider returning a specific error/reason string (or using success: true, skipped: true) for this path.

Suggested change

	core.info(`⏭ Max count (${maxCount}) reached, skipping agent assignment`);
	const agentNameForSkip = message.agent ?? defaultAgent;
	_allResults.push({ issue_number: message.issue_number || null, pull_number: message.pull_number || null, agent: agentNameForSkip, owner: null, repo: null, success: false, skipped: true });
	return { success: false, skipped: true };
	const error = `Max count (${maxCount}) reached, skipping agent assignment`;
	core.info(`⏭ ${error}`);
	const agentNameForSkip = message.agent ?? defaultAgent;
	_allResults.push({ issue_number: message.issue_number || null, pull_number: message.pull_number || null, agent: agentNameForSkip, owner: null, repo: null, success: false, skipped: true, error });
	return { success: false, skipped: true, error };