| name | generate-psych-profile |
| description | Extract the core motivations, decisions under pressure, cognitive biases and mental models of a subject (historical figure / topic subject) to drive psychology-led narrative. Use when the Narrative & Psychology agent must understand WHY a subject acted as they did before drafting a script. |
| metadata | {"agent":"narrative-psychology","source":"Project Atlas Agent Skills Manifest §5","layer":"L3-story"} |
generate-psych-profile
Produce a psychological profile that turns facts into meaning.
When to use
- Before drafting a documentary script for a personality/topic.
- Feeding the
psychological_arcs record (identity, struggle, failure, growth, resilience, reinvention, legacy — spec §4.2).
Function signature (manifest contract)
def generate_psych_profile(subject_entity: str) -> dict:
"""Extract motivations, decisions under pressure, biases, and mental models."""
Inputs / Outputs
- Input:
subject_entity (name or entity id).
- Output:
{focus, motivations[], biases[], mental_models[], arc: {identity, struggle, failure, growth, resilience, reinvention, legacy}}.
Implementation
Call the hermes-bridge MCP tool run_judgment_agent(insight_type="generate-psych-profile", query=<ask>, context={"subject_entity": subject_entity}). Hits Hermes's POST /v1/agents/generate-psych-profile (new hermes/agents/generate-psych-profile/), the same self-correcting LLM loop / mock-fallback pattern as vibe/voice-director. Caveat: this is LLM judgment from training-data knowledge, not retrieval against the actual graph/dossier yet -- treat the output as a first draft until gather-citations + query-semantic-nodes are feeding it real context.
Backend dependency
- Persisting to
psychological_arcs (Railway) is still stubbed until the Postgres/AGE backbone is wired -- the judgment call itself works today, it just isn't durably stored yet.
- Deep biography discovery via Exa (key not present yet).
Model
gemini-direct/gemini-2.5-flash — premium reasoning (psychological judgment, spec §2.4/§11.3).
Skip/limit prompt compression here (nuance matters).
