Support arguments and cleaner transcript handling for saved prompts

Implemented in #3403.

What feature would you like to see?

Problem / Motivation

• Saved prompts today can’t accept parameters — they are static text blocks.
• Prompt bodies leak into the transcript, creating duplication and noise.
• There’s no explicit way to surface user-provided arguments as “high-priority context” for the agent.

Proposed Solution

• Extend /prompt-name <argument> syntax.
• In transcript: show exactly what user typed (/<name> <arg>), without revealing the hidden body.
• To the agent: send the saved prompt body, optionally preceded by a concise preface if an argument is present.
• Handle both single-line and multi-line arguments (multi-line wrapped in fenced code block).
• Introduce hidden message submission (submit_hidden_text_message) and route it through the existing queue so turn ordering is preserved.

Benefits

• Saved prompts become significantly more flexible — they can be reused with different parameters instead of hard-coding variations.
• Cleaner transcripts, no duplication of body text.
• Arguments delivered to the agent as authoritative context.
• Unified queuing semantics: visible and hidden submissions follow the same order.

Are you interested in implementing this feature?

If maintainers are aligned, I’ll prepare a PR with implementation, tests, and updated docs.

Additional information

_No response_