hermes chat is fully interactive, but there is no way to start it with a first user message already in the conversation. The single-query options all break interactivity:

• hermes chat -q QUERY runs once and exits
• hermes -z PROMPT is oneshot, also exits
• hermes chat --resume <id> requires a pre-existing session and adds no new turn
• piping to stdin hangs because chat needs a tty

That blocks launchers and wrappers that want to open hermes in a fresh terminal tab pre-seeded with context: thread-resolution prompts, project briefings, agent handoffs from claude or codex, and so on. claude "PROMPT" and codex "PROMPT" both accept a positional initial prompt and keep the REPL alive afterwards; hermes has no equivalent.

proposal

Add -i / --initial PROMPT to hermes chat:

1. accept a prompt string with the same rules as -q
2. inject it as the first user message in the conversation
3. run that turn through cli.chat() so tools, skills, memory, and AGENTS.md load normally
4. fall through to interactive cli.run() instead of returning

touch points

The diff is small:

• hermes_cli/_parser.py:237: register -i, --initial next to -q, --query
• hermes_cli/main.py:1316-1340: thread args.initial into the kwargs forwarded to cli.main()
• cli.py:11976: add an if initial: branch that runs one turn and falls through to cli.run() at line 12039, instead of returning at line 12036 like the existing if query or image: branch does

why not a workaround

• pty-driven send-keys (kitty send-text, tmux send-keys) is terminal-specific, races against Rich/Textual UI startup, and breaks on multi-line prompts without bracketed-paste tricks
• a plugin that calls cli._pending_input.put(msg) works in-process but every user has to install and enable it
• both reimplement logic the CLI already has for -q

related merged PRs

Small chat-flag PRs have been accepted before:

• #4314 added --max-turns
• #11024 tightened -Q quiet mode