[bot] Document userPromptSubmitted LLM bypass and multi-skill invocation (v1.0.44)

[github-actions]

What's new in v1.0.44

The v1.0.44 release (2026-05-08) shipped two notable user-facing features that weren't yet documented in the Learning Hub:

1. userPromptSubmitted hooks can bypass the LLM

Hooks on the userPromptSubmitted event can now handle a request directly by writing {"response": "..."} to stdout. The CLI delivers that text to the user and skips the model call entirely. Previously the event was documented only as useful for auditing/logging.

Use cases: inline FAQ bots, policy enforcement with clear rejection messages, prompt redirect patterns.

2. Multiple skills per message & mid-input slash commands

Slash commands no longer have to appear at the start of a message — they can appear mid-input. Users can also invoke multiple skills in a single message (e.g., /generate-tests and then /conventional-commit).

---

What was updated

website/src/content/docs/learning-hub/automating-with-hooks.md

• Updated userPromptSubmitted event table row to mention the new LLM-bypass capability
• Added new "Handling Requests Directly with userPromptSubmitted (v1.0.44+)" example section with working script and explanation of the {"response": "..."} pattern
• Updated the FAQ answer ("Can hooks access the user's prompt text?") to describe the bypass capability
• Bumped lastUpdated to 2026-05-08

website/src/content/docs/learning-hub/creating-effective-skills.md

• Expanded the "How do I invoke a skill?" Q&A to document mid-input slash commands and multiple skills per message
• Replaced "Can agents chain multiple skills?" with "Can I invoke multiple skills in one message?" — covering both user-facing multi-skill invocation and agent discovery chaining
• Bumped lastUpdated to 2026-05-08

---

Source announcements

• github/copilot-cli v1.0.44 release notes

Generated by Learning Hub Updater · ● 2.6M · ◷