System prompt and Soul Signature

System prompt source

utils.get_system_prompt() returns a single large string that defines:

Identity: OPSIE as an SCI agent for ARPA Corporation, personality blend (Ghost in the Shell, Urusei Yatsura, Sailor Moon tonal notes).
Behavior: Memory-first reasoning, honesty about uncertainty, anti-template conversational style.
Capability catalog: Narrative description of /0x, /read, /imagine, /video, /music, /ask, /markets, /mail, /dna, /room (see also arpahls/rooms for standalone room tooling), voice modes, URL vision, and memory distinctions between OPSIE replies versus third-party command outputs.

At boot, boot_up_sequence():

Optionally applies system_prompt.format(call_name=call_name) when the base string contains {call_name} placeholders (the stock get_system_prompt() body does not require this, but the call is harmless).
Appends user identity, emotional context from DeepFace, and ARPA tier language (R versus non-R).
Replaces convo[0]['content'] with the hydrated prompt.
Appends Soul Signature: and each line from known_user_names[current_user]['soul_sig'].

Soul Signature lines are highest-priority user-specific instructions in the system prompt. Operators manage them with /soulsig:

Command	Effect
`/soulsig`	Print current lines
`/soulsig <text>`	Append a line (persisted through `save_known_user_names()`)
`/soulsig wipe`	Confirmed destructive clear with alert audio
`/soulsig heal`	Restore from `temporary_soul_sig` if a wipe just occurred

The narrative in utils.py instructs OPSIE not to over-expose Soul Signature content unless relevant.

get_agent_intro supplies short spoken-style intros for OPSIE, Nyx, and G1 when a /room session begins.

dreaming_expressions and get_random_expression() support conversational filler before generative jobs (/imagine, /music, and related UX).