Fix release workflow: Clarify MCP tool call vs bash file operations

[Copilot]

Problem

In run #21382926511, the agent used bash to write /tmp/gh-aw/safeoutputs/update_release_001.json instead of calling the update_release MCP tool. Result: artifact not captured, release notes not updated.

Root cause: With both bash: ["*"] and MCP tools available, the agent simulated the tool call via file operations rather than invoking the MCP protocol.

Changes

.github/workflows/release.md (lines 471-500)

• Add explicit MCP tool clarification with implementation details
• Warn against manual file writes and bash simulation
• Update example with concrete version number

**HOW TO CALL THE TOOL:**

The `update_release` tool is an **MCP (Model Context Protocol) tool**, 
not a bash command or file operation. To call it:

1. **Use the MCP tool call syntax** - The tool is registered as 
   `update_release` (or `safeoutputs___update_release`)
2. **DO NOT write JSON files manually** - Writing files like 
   `/tmp/gh-aw/safeoutputs/update_release_001.json` will NOT work
3. **DO NOT use bash to simulate tool calls** - The tool must be 
   called through the MCP protocol

**IMPORTANT**: This is a real tool call through MCP, NOT a bash command 
or file write operation. The tool will automatically write to the correct 
output file (`/opt/gh-aw/safeoutputs/outputs.jsonl`).

This prevents multi-tool confusion when workflows provide both bash and MCP capabilities.

Original prompt

Reference: https://github.com/githubnext/gh-aw/actions/runs/21382926511/job/61553811780#step:25:1

Investigate why the safe outputs tool update-release was not called in this agentic run.

Custom agent used: agentic-workflows
GitHub Agentic Workflows (gh-aw) - Create, debug, and upgrade AI-powered workflows with intelligent prompt routing

---

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