Enable all LLM tools in prompts (agent mode)

[bendrucker]

Enables canBeReferencedInPrompt for all LLM tools.

Set to true if the tool can be used in agent mode or referenced in chat.

Previously, these tools could only be used via the @githubpr chat participant. Only the activePullRequest tool was available directly. Now, they can all be called with # syntax or invoked by agent mode.

Perhaps this was disabled intentionally and there is additional polish to do. But this works and #6792 (comment) suggested that there was some confusion on whether this is live today (it's not).

Before

Using the web fetch tool works for public repositories, albeit with a lot of HTML noise as compared to feeding the model a JSON representation of the issue.

Agent mode is unable to fetch details about any issues in private repositories.

After

Related

• Fetching issues fails in Copilot Agent mode #6792

[bendrucker]

@microsoft-github-policy-service agree

[alexr00]

This was disabled intentionally. canBeReferencedInPrompt was just for referencing a tool in a prompt with # or using the attach context button. I'm confirming that canBeReferencedInPrompt also impacts tool use in agent mode.

[bendrucker]

Some other references:

https://code.visualstudio.com/updates/v1_99#_vs-code-extension-tools-in-agent-mode

Any tool contributed to this API which sets toolReferenceName and canBeReferencedInPrompt in its configuration is automatically available in agent mode.

microsoft/vscode#245129
microsoft/vscode#231927

While that language isn't 100% definitive, my testing has shown that to be required for an agent to call the tool.

The screenshot I included with the checklists is triggered by the Select Tools button:

And it seems reasonable to assume that list is the authoritative set of tools available to the agent.