| name | mind-clone-creator |
| description | Use when the user wants to create their own digital twin, mind clone, personal AI advisor, or a persona-plus-workflow work clone from their own experience, judgment style, and work habits. Trigger on requests like "创建分身", "工作分身", "数字分身", "克隆自己", "create my clone", "build my work clone", or "把我的经验做成 AI 顾问". |
Mind Clone Creator
Overview
Create the user's own digital twin through structured self-interview.
The foundation is always a persona-layer clone skill for OpenClaw: it restores how the user thinks, judges, responds, and collaborates.
This skill also supports a first-class persona-plus-workflow path: keep the workflow track alive from the top-level bundle, then compile blueprint/runtime artifacts after one concrete work unit is defined.
It still does not, by default, claim a full work-replacing agent with autonomous workflow execution, task routing, or stage decisions before that workflow layer is explicitly modeled.
Use This Skill When
- The user wants to create their own digital twin, mind clone, personal AI advisor, or reusable consultation persona.
- The user wants to package their own experience, judgment style, boundaries, or work habits into an OpenClaw skill.
- The user already has self-authored materials and wants them turned into
clone_config.yaml, mind_profile.md, system_prompt.md, and a runnable clone package.
- The user wants to extend that persona into one explicit workflow blueprint for a specific kind of recurring work.
- The user wants both: a clone that sounds like them and a workflow-backed clone for one recurring category of work.
Do not use this skill when:
- The user wants to clone another person. Redirect to
mind-clone-advisor.
- The user expects a fully autonomous work-replacing agent without workflow-layer design.
User Flow
Keep the user-visible flow at 5 steps:
- Build the outline
- Run the interview
- Confirm the profile
- Generate and evaluate automatically
- Deliver
final or draft
The user only answers questions and confirms the profile.
Environment checks, helper checks, profession research, packaging, and evaluation stay in the background unless they affect a user decision.
Local Files And Privacy
To support multi-turn progress, this workflow may write local artifacts such as personal_interview.md, interview state files, bundle manifests, and generated outputs in the current workspace or chosen output directory.
Treat these as user-controlled local working files, not hidden external storage.
If the user wants a low-retention run, minimize intermediate files when feasible and explain which local files can be deleted after delivery.
Core Guards
Always enforce these gates:
- The user is creating their own clone, not cloning another person.
- The user accepts this is an honest approximation, not perfect replication.
- The user's own answers, examples, and optional self-provided materials are the source of truth.
- A
final clone requires coverage for identity, skills, knowledge, work_process, thinking, and expression.
If the user wants a clone that can replace real work, state clearly that persona is still the foundation.
Start the persona-plus-workflow path immediately, keep workflow_interview.md active even before the first target_work_unit is final, and do not claim workflow artifacts are ready until that target is explicit.
Helper Skills
mind-clone-creator stays the controller skill for the whole workflow.
If helpers are needed, invoke them inside this workflow instead of switching the user onto another primary skill.
Allowed helper patterns:
- bundled
references/multi-search-engine/SKILL.md: default path for all networked search
deep-research: profession patterns, evaluation scenarios, knowledge-base suggestions
content-harvester: normalize the user's own content into reusable materials
docx / pdf / xlsx: extract user-provided files
tikhub-api-helper: fetch the user's own platform content
find-skills / skill-installer: discover or install missing skills when the real workflow exceeds current capability
Rules:
- Check actual helper availability first; do not assume the user's environment matches the creator environment.
- If a missing helper or external skill is likely needed, surface that right after the outline step, before the deep interview starts.
- Never install or rely on an external skill without user confirmation.
- Helper skills never replace the user's interview answers or the final
draft/final decision.
Execution Rules
- Follow SKILL-OPERATING-STANDARD.md for language, privacy, quality, and output rules.
- If
python3 is available, prefer the scripts in scripts/; otherwise follow the manual process in steps/.
- Ask one interview block at a time. If the user's answer is vague or lacks an example, follow up before moving on.
- If the user asks for the quick version, follow the quick interview path in steps/02_self_interview.md.
- If the user asks for persona + work clone together, prefer
scripts/bootstrap_working_clone_bundle.py in persona-plus-workflow mode instead of treating workflow as a later side quest.
- Route all web search through
references/multi-search-engine/SKILL.md.
- Use external research only for profession patterns, realistic evaluation scenarios, and suggested materials. It must never override the user's own stated beliefs or boundaries.
- Use the bundled prompts and templates instead of improvising structured artifacts.
- When using persisted interview state, treat
accept_for_now as temporary only; validate final readiness before claiming draft_status: final.
- Use the rubric in eval/scoring_rubric.md before allowing final output.
Preferred Entry Points
Use these entrypoints by stage:
- Persona interview start:
scripts/init_personal_interview.py
- Persona next-question / persisted loop:
scripts/plan_clone_interview_next.py, scripts/build_next_interview_update.py, scripts/init_clone_interview_state.py, scripts/advance_clone_interview_state.py, scripts/run_clone_interview_turn.py
- Bundle bootstrap / refresh / final-readiness:
scripts/bootstrap_working_clone_bundle.py, scripts/refresh_working_clone_bundle.py, scripts/run_working_clone_until_final.py, scripts/validate_working_clone_bundle.py
- Workflow extension:
scripts/init_workflow_interview.py, scripts/validate_workflow_interview.py, scripts/build_workflow_blueprint.py, scripts/bootstrap_workflow_blueprint.py, scripts/build_workflow_clone_skill.py, scripts/bootstrap_workflow_clone_runtime.py
- Final packaging and release gates:
scripts/extract_clone_draft.py, scripts/build_clone_config.py, scripts/validate_clone_interview_state.py, scripts/validate_clone_config.py, scripts/render_delivery_summary.py, scripts/clone_ops.py validate release-readiness
Use references/capability_index.md for the full script map instead of turning this file into the script catalog.
Output Contract
Primary deliverables:
- a personal clone skill folder, for example
my-clone-skill/
- when the user wants persona + workflow together, a working bundle that keeps workflow status explicit until the workflow chain is fully compiled
Inside the personal clone skill:
SKILL.md
clone_config.yaml
mind_profile.md
system_prompt.md
eval_report.md
research_digest.md if profession research was used
Inside the working bundle when workflow is enabled:
workflow_interview.md immediately, even if target_work_unit is still pending
workflow_blueprint.md, workflow clone skill, and workflow runtime bundle after target_work_unit is defined and the workflow chain is compiled
By default this deliverable can:
- answer in a way closer to the user's reasoning and expression
- preserve boundaries, principles, and preferred work style
- act as a consultation or review proxy in bounded scenarios
By default it does not:
- run multi-stage work end to end
- decide current execution stage for arbitrary jobs
- replace the user's full production work loop without extra workflow design
Failure And Fallback
- If the user wants to clone someone else, stop and redirect to
mind-clone-advisor.
- If answers are vague, ask for concrete examples before advancing.
- If the user has no prepared materials, continue by collecting the missing core information through follow-up questions.
- If external research conflicts with the user's own answers, treat the user's answers as primary and mark the research as a tension to review.
- If quality is below 60 or the user's own core answers are still missing, output only a
draft.
- Even in
draft mode, still deliver the runnable personal clone files plus a concrete improvement plan.
- If the user explicitly wants a work-replacing clone, switch to the
persona-plus-workflow path. If the first work category is still unclear, initialize workflow_interview.md, keep the blocker visible, and continue packaging the persona layer without pretending the workflow chain is already done.
Navigation
- Step docs: steps/00_skill_check.md, steps/01_profession_parse.md, steps/01b_domain_research.md, steps/02_self_interview.md, steps/03_mind_profile.md, steps/04_system_prompt.md, steps/05_quality_eval.md, steps/06_output.md, steps/07_workflow_agent_design.md
- Prompt library: prompts/profession_analyzer.md, prompts/domain_research_brief.md, prompts/interview_guide.md, prompts/profile_synthesizer.md, prompts/prompt_generator.md, prompts/workflow_interview_guide.md, prompts/eval_question_gen.md, prompts/eval_scorer.md
- Script map and operator docs: references/capability_index.md, references/operator_playbook.md
- Output formats and examples: references/personal-clone-skill-format.md, references/workflow-clone-skill-format.md, references/profession-adapter-schema.md, eval/human_review_checklist.md,
templates/, examples/ai_engineer/