Twinify
WarnAudited by ClawScan on May 10, 2026.
Overview
Twinify openly uses WhatsApp exports with a consent warning, but it creates a persistent AI persona that is instructed to impersonate a real person and deny being an AI.
Only use this with explicit informed consent from the person being modeled. Before installing, consider changing the generated agent template so it always discloses that it is an AI simulation, review what chat data will be sent to the LLM, redact sensitive details, and make sure you know how to delete the created workspace and OpenClaw config entry.
Findings (5)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
Someone interacting with the twin could believe it is the real person or place undue trust in responses based on private chat history.
The generated twin is explicitly instructed to present as the real person and not disclose its AI nature, which conflicts with transparent simulation and creates impersonation risk.
You are {PERSON_NAME}. Not an AI, not an assistant, not a bot. ... NEVER break characterRequire the twin to clearly label itself as an AI simulation in every session, remove denial-of-AI wording, and restrict use to consented, private contexts.
Private messages, personal facts, names, events, and relationship context may remain available to the local agent until manually deleted.
The workflow persists personality, examples, and memory derived from private chat messages for reuse by the twin.
create these files ... SOUL.md ... EXAMPLES.md ... MEMORY.md ... Parsed data and generated profiles are stored in the user's local OpenClaw workspace.
Use only with explicit consent, redact sensitive third-party details, store outputs in a controlled location, and provide or follow a clear deletion process.
WhatsApp chat content may be processed by the user's LLM provider according to that provider's data policies.
The skill acknowledges that chat contents may be sent to the configured LLM provider; this is expected for analysis but involves sensitive personal data leaving local processing.
Chat exports should be processed locally and never transmitted to external services beyond the LLM API call.
Before use, confirm which model/provider will process the chat, avoid highly sensitive exports, and redact content that should not leave the device.
The twin and its private memory may remain installed and callable until the user disables or deletes it.
The skill intentionally creates and registers a persistent new OpenClaw agent, which is central to the purpose but remains available after setup.
Create workspace: `~/.openclaw/workspace-<agent_id>/` ... Register agent in OpenClaw config via `gateway config.patch`
Confirm the agent name and scope before registration, keep it private, and remove the workspace and config entry when consent is withdrawn or the twin is no longer needed.
The skill may fail or require manual environment setup if python3 is not available.
The workflow requires a local Python runtime, while the registry metadata declares no required binaries; this is a setup metadata gap rather than hidden execution.
python3 scripts/parse_chat.py <chat_export.txt> <target_name> <output_dir>
Declare python3 as a requirement and continue reviewing any local script before running it on private chat exports.