| name | persona-cloner |
| description | Build an OpenClaw-ready public-figure persona package from public materials. Use when the user wants a directly runnable agent that answers as a specific person, with runtime files (`SOUL.md`, `IDENTITY.md`, `AGENTS.md`, `MEMORY.md`) as the default deliverable. Prioritize hard behavioral controls so the shipped runtime stays verdict-first, concise, non-generic, and resistant to default assistant voice drift. |
Persona Cloner
Build the runtime persona package first.
Default user-facing output:
SOUL.mdIDENTITY.mdAGENTS.mdMEMORY.md- optional:
README-agent.md
- optional: install into an OpenClaw workspace
Do not make the internal extraction rail the product unless the user explicitly asks for provenance, evaluation history, or maintenance internals.
Read references only when needed:
references/workflow.mdreferences/output-spec.mdreferences/output-discipline.mdreferences/acceptance-checklist.mdreferences/source-intake.mdreferences/persona-core.mdreferences/worldview-map.mdreferences/decision-rules.mdreferences/contradiction-handling.mdreferences/memory-injection-spec.mdreferences/identity-boundary-statement.mdreferences/decision-fidelity-eval.mdreferences/behavior-eval.md
Use scripts as internal helpers:
scripts/scaffold_persona_clone.pyscripts/extract_persona_pipeline.pyscripts/eval_decision_fidelity.pyscripts/eval_runtime_behavior.pyscripts/manage_persona_memory.pyscripts/validate_persona_package.pyscripts/install_runtime_agent.py
Deliverable rule
Default to the smallest honest package that can run well.
A normal successful delivery is:
- one folder containing
SOUL.md, IDENTITY.md, AGENTS.md, MEMORY.md
- optionally
README-agent.md
- optionally an install step using
scripts/install_runtime_agent.py
Core requirement
Every shipped runtime must contain explicit output-discipline rules:
- response length policy
- verdict-first policy
- expansion only on demand
- anti-lecture policy
- anti-assistant-tone policy
- bluntness / restraint tuning
Put these controls into the runtime artifacts themselves, mainly SOUL.md and AGENTS.md.
Commands
python scripts/scaffold_persona_clone.py "Target Name" --out <output-dir> --mode persona
python scripts/install_runtime_agent.py <built-agent-dir> <target-workspace>
python scripts/validate_persona_package.py <package-dir>
python scripts/eval_runtime_behavior.py run --package-dir <package-dir> --version candidate --model sub2api/gpt-5.4
Final reminder
This skill is for shipping a usable OpenClaw persona package. Internal complexity is allowed, but the default output should stay minimal, runnable, honest, behaviorally disciplined, and hard to wash out with generic assistant tone.
