Add noop safe output for transparent workflow completion

[Copilot]

Agents need to emit human-visible artifacts even when no actions are required (e.g., "No issues found"). The noop safe output provides a fallback mechanism ensuring workflows never complete silently.

Changes

Schema & Types

• Added NoOpOutput schema definition with required message field
• Added TypeScript interfaces (NoOpItem, NoOpConfig) for type safety
• Updated workflow schema to accept noop configuration

JavaScript Implementation

• collect_ndjson_output.cjs: Validation case with message sanitization (neutralizes @mentions, etc.)
• notify_comment_error.cjs: Integrated noop message display into conclusion job with dual-mode handling:
  • When activation comment exists: appends noop messages to the comment
  • When no activation comment: writes noop messages to step summary
• Bundling Fix: Moved notify_comment_error.cjs from js.go to scripts.go to enable proper bundling of load_agent_output.cjs dependency. The bundler now inlines all required modules directly into compiled workflows, eliminating runtime require() errors.
• Default max: 1 message (consistent with other safe outputs)

Go Integration

• noop.go: Configuration parser with default max of 1
• safe_outputs.go: Auto-enable logic - noop enabled by default when safe-outputs exists; added noop to tool filtering logic for MCP integration
• notify_comment.go: Updated to always generate conclusion job when safe-outputs exist (not just when activation comments are present)
• scripts.go: Added getNotifyCommentErrorScript() bundling function with lazy bundling and caching, following the same pattern as other bundled scripts

MCP Integration

• safe_outputs_tools.json: Added noop tool definition with proper schema for MCP server integration
• Tool includes description and input schema (message field required)
• Enables AI agents to discover and use the noop tool through the safe outputs MCP server
• Updated test expectations to include noop in the tool list

Architecture

Noop messages are processed in the conclusion job (not as a separate job). The conclusion job is now always generated when safe-outputs exist, ensuring noop messages can be handled regardless of whether activation comments are configured.

Behavior:

• With activation comment (reactions configured): Noop messages are appended to the completion status comment
• Without activation comment: Noop messages are written to the step summary for transparency

This guarantees that noop messages are always visible somewhere, preventing silent workflow completion.

Bundling Implementation

The notify_comment_error.cjs script uses require("./load_agent_output.cjs") to load agent output artifacts. To ensure this works in GitHub Actions runtime:

1. Script moved to scripts.go (bundled scripts) from js.go (standalone scripts)
2. Bundler automatically inlines load_agent_output.cjs content directly into the compiled workflow
3. No external require() calls remain in generated YAML files
4. All 80 workflows compile successfully with dependencies properly inlined

Usage

---
safe-outputs:
  noop:  # Optional, enabled by default with max: 1
    max: 1
---

# Output format
{"type": "noop", "message": "Analysis complete - no issues found"}

Noop is always available as a fallback and integrates with the conclusion job to ensure human-visible artifacts are produced in all scenarios. The tool is automatically exposed through the safe outputs MCP server for AI agent discovery.

Custom agent used: copilot-add-safe-output-type
Adding a New Safe Output Type to GitHub Agentic Workflows

Custom agent used: copilot-add-safe-output-type
Adding a New Safe Output Type to GitHub Agentic Workflows

Original prompt

Add a "no-op" safe output (always enabled).

Fallback and No-Op Outputs: Finally, to improve safety and clarity, the instructions could define a “no-operation” or fallback safe output. This would be used when the agent decides to take no significant action but still needs to produce a human-visible artifact. For example, a log-only or noop safe output might simply post a comment or discussion stating the agent’s analysis results or that no changes were needed. This addresses scenarios where the agent’s task is informational or where it cannot proceed – instead of failing silently or producing nothing (which, as observed, can confuse the workflow)  , the agent would still output something for transparency. Given the project’s emphasis that agent execution “always leave human‑visible artifacts (comments, PRs, logs) instead of silent mutation” , formalizing a fallback safe output ensures that even if no issue/PR is created, the run isn’t silent. For instance, if an agent finds no problems in code, it could use this output to comment “No issues found in this review” as a record of completion

Takes a message string argument.

The no-op text is used when update the add-comment comment in the default reporting.

Custom agent used: copilot-add-safe-output-type
Adding a New Safe Output Type to GitHub Agentic Workflows

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[pelikhan]

@copilot default max is 1

[pelikhan]

@copilot do not generate a separate noop job, do the no-op processing in the conclusion job instead.

Post the no-op messages to the activation comment if any.

[pelikhan]

@copilot continue work

[Copilot]

@pelikhan Unfortunately I hit an unexpected error while processing your comment. I've automatically reported this to GitHub.

You can ask me to try again later by mentioning me in a new comment.

If you want to contact GitHub about this error, please mention the following identifier so they can better serve you: 04bf955a-aef5-4302-b510-a9488eae60d2

Sorry for the inconvenience!

[pelikhan]

@copilot max should be 1

[Copilot]

@copilot continue work

Continued the work and completed the implementation in commit 0330244. All requested changes have been addressed:

1. Default max changed to 1 ✓
2. Separate noop job removed ✓
3. Noop processing integrated into conclusion job ✓
4. Noop messages posted to activation comments ✓

The conclusion job now loads agent output and appends noop messages to the completion status comment when present.

@copilot max should be 1

Fixed in commit 0330244. Default max for noop is now 1 across all files:

• noop.go: parseNoOpConfig default = 1
• safe_outputs.go: auto-enable default = 1
• collect_ndjson_output.cjs: getDefaultMaxForType returns 1
• main_workflow_schema.json: schema default = 1

[github-actions]

Agentic Changeset Generator triggered by this pull request.

[pelikhan]

@copilot the conclusion job should always be generated so that no-ops can be handled

[Copilot]

@copilot the conclusion job should always be generated so that no-ops can be handled

Fixed in commit 34bfcca. The conclusion job is now always generated when safe-outputs exist (because noop is auto-enabled as a fallback).

Key changes:

• notify_comment.go: Removed reaction requirement - conclusion job created whenever safe-outputs exist
• notify_comment_error.cjs: Handles missing activation comments gracefully by writing noop messages to step summary
• Job condition simplified to: (always()) && (needs.agent.result != 'skipped')

Behavior:

• With activation comment: Updates comment with noop messages
• Without activation comment: Writes noop messages to step summary for transparency

All tests pass and 80/80 workflows compile successfully.

[github-actions]

Agentic Changeset Generator triggered by this pull request.

[pelikhan]

@copilot make sure load_agent_output.cjs is inlined by the bundler

See error trace at https://github.com/githubnext/gh-aw/actions/runs/19478013400/job/55742960695#step:5:1

[pelikhan]

@copilot make sure load_agent_output.cjs is inlined by the bundler

See error trace at https://github.com/githubnext/gh-aw/actions/runs/19478013400/job/55742960695#step:5:1

[pelikhan]

@copilot make sure the no-op tool is defined in the safe output tools.JSON

[Copilot]

@copilot make sure load_agent_output.cjs is inlined by the bundler

See error trace at https://github.com/githubnext/gh-aw/actions/runs/19478013400/job/55742960695#step:5:1

Fixed in commit 1934f80.

Changes:

• Moved notify_comment_error.cjs from js.go to scripts.go
• Added getNotifyCommentErrorScript() bundling function
• Updated notify_comment.go to call the bundled version

Verification:

• grep -r "require.*load_agent_output" .github/workflows/*.lock.yml returns 0 results (no require statements)
• grep -r "function loadAgentOutput" .github/workflows/*.lock.yml returns 153 instances (properly inlined)
• All 80 workflows compile successfully

The bundler now automatically inlines load_agent_output.cjs into notify_comment_error.cjs, eliminating the require() error.

[Copilot]

@copilot make sure the no-op tool is defined in the safe output tools.JSON

Fixed in commit 3901953.

Changes:

• Added noop tool definition to safe_outputs_tools.json with proper schema
• Updated filtering logic in safe_outputs.go to include noop when configured
• Updated test expectations to include noop in the expected tools list

Verification:

• All 80 workflows compile successfully with noop tool included
• Tool definition is properly exposed through the safe outputs MCP server interface
• Unit tests pass

[github-actions]

✅ Agentic Changeset Generator completed successfully.

[Copilot]

Pull Request Overview

This PR adds a new "noop" (no-operation) safe output type that allows AI agents to produce human-visible artifacts when no significant actions are required, ensuring workflows never complete silently. The noop output is automatically enabled by default when safe-outputs are configured, providing a transparent fallback mechanism.

Key Changes

• Added noop schema definition with required message field and comprehensive validation
• Implemented noop processing in conclusion job with dual-mode support (activation comment vs step summary)
• Integrated noop tool into MCP server for AI agent discoverability
• Updated conclusion job logic to always run when safe-outputs exist (not just when activation comments are configured)

Reviewed Changes

Copilot reviewed 93 out of 95 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
schemas/agent-output.json	Added NoOpOutput schema definition with required message field
pkg/workflow/scripts.go	Added bundling support for notify_comment_error script to inline load_agent_output dependency
pkg/workflow/safe_outputs_tools_test.go	Updated test expectations to include noop in tool list
pkg/workflow/safe_outputs.go	Added auto-enable logic for noop (enabled by default when safe-outputs exist) and MCP tool filtering
pkg/workflow/reaction_none_test.go	Updated test to expect conclusion job creation when safe-outputs exist
pkg/workflow/notify_comment_test.go	Updated tests to reflect that conclusion job is always created when safe-outputs exist
pkg/workflow/notify_comment.go	Changed conclusion job creation logic to always run when safe-outputs exist (not requiring activation comment)
pkg/workflow/noop.go	Implemented noop configuration parser with default max of 1
pkg/workflow/js/types/safe-outputs.d.ts	Added NoOpItem TypeScript interface
pkg/workflow/js/types/safe-outputs-config.d.ts	Added NoOpConfig TypeScript interface
pkg/workflow/js/safe_outputs_tools.json	Added noop tool definition for MCP integration
pkg/workflow/js/notify_comment_error.cjs	Enhanced to process noop messages with dual-mode handling (comment update or step summary)
pkg/workflow/js/noop.test.cjs	Comprehensive test coverage for noop processing including staged mode, sanitization, and multiple messages
pkg/workflow/js/noop.cjs	Main noop implementation with message processing and step summary generation
pkg/workflow/js/collect_ndjson_output.test.cjs	Added validation tests for noop output including message requirements and sanitization
pkg/workflow/js/collect_ndjson_output.cjs	Added noop validation case with message sanitization and default max of 1
pkg/workflow/js.go	Added embedded noop script variable
pkg/workflow/compiler.go	Added NoOp field to SafeOutputsConfig struct
pkg/workflow/comment_env_vars_conditional_test.go	Updated tests to reflect that conclusion job always has comment env vars when safe-outputs exist
pkg/parser/schemas/main_workflow_schema.json	Added noop schema definition with configuration options
pkg/cli/workflows/test-copilot-noop.md	Added test workflow for Copilot engine
pkg/cli/workflows/test-codex-noop.md	Added test workflow for Codex engine
pkg/cli/workflows/test-claude-noop.md	Added test workflow for Claude engine
.github/workflows/*.lock.yml	Updated all compiled workflows to include noop tool and conclusion job with bundled load_agent_output
.changeset/patch-add-noop-safe-output.md	Added changeset for patch release

Comments suppressed due to low confidence (2)

pkg/workflow/js/notify_comment_error.cjs:1

• Incorrect property access: loadAgentOutput() returns { success, items } but the code checks for agentOutputResult.data. This should be agentOutputResult.items to match the return structure.
  pkg/workflow/js/notify_comment_error.cjs:1
• Same incorrect property access pattern: should be agentOutputResult.items instead of agentOutputResult.data to match the return structure from loadAgentOutput().

---

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