Doppel
Build your digital-twin writing voice
Antreas Antoniou
@antreasantoniou
Install
$ openclaw skills install @antreasantoniou/doppelDoppel — Digital Twin Writing Voice
Build a bounded digital twin of how a consenting person moves through writing. A Doppel models writing voice from evidence; it does not simulate the whole person. The subject remains the author and final ratifier, and every output remains a draft.
Boundaries
- Use only material authored by the subject and supplied with explicit consent for this task.
- Refuse third-party impersonation, non-consensual cloning, deceptive attribution, or requests to speak as a public figure.
- Keep corpora, manifests, profiles, context packs, and drafts local and out of Git.
- Never send, submit, post, publish, or attribute a draft without the subject's approval of that exact artefact.
- Voice evidence is not fact authority. Verify current claims independently and never invent beliefs, feelings, results, or relationships.
- Avoid reproducing distinctive source phrases. Preserve movement, emphasis, cadence, and decision patterns—not memorised sentences.
Read references/safety-and-consent.md before handling source material.
Workflow
-
Confirm the subject owns the source material, consents to its use, and will ratify the final text.
-
Copy
examples/voice-manifest.example.jsonto an ignored local filename such asvoice-manifest.local.json. Add subject-authored sources and their SHA-256 hashes. -
Validate and build a deterministic, local conditioning pack:
python3 scripts/build_voice_context.py \ --manifest voice-manifest.local.json \ --task "Explain the project launch" \ --audience "technical peers" \ --register public-essay \ --output voice-context.local.md -
Read the context pack completely. Extract recurring reasoning moves, compression/expansion patterns, boundaries, and register shifts. Treat that evidence-backed model as the writing-voice Doppel; do not reduce the person to punctuation or catchphrases.
-
Draft from the subject's intent and current facts. Label the output as a draft requiring their ratification.
-
Review twice: first for factual/semantic correctness, then for whether one mind appears to move through the piece. Remove copied phrases, invented claims, and generic polish.
-
Return the draft to the subject. Publication requires explicit approval of that exact version.
For a read-only invocation, validate sources and build the context in memory or in a temporary directory; do not leave a manifest, context pack, or draft in the source project.
For narrow edits, preserve the original structure and strongest authored lines. For a new piece, ask only for missing intent or facts that would materially change the result.
Top skills in this category
self-improving agent
@pskoettCaptures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Claude ('No, that's wrong...', 'Actually...'), (3) User requests a capability that doesn't exist, (4) An external API or tool fails, (5) Claude rea
Skill Vetter
@spclaudehomeSecurity-first skill vetting for AI agents. Use before installing any skill from ClawdHub, GitHub, or other sources. Checks for red flags, permission scope, and suspicious patterns.
Self-Improving + Proactive Agent
@ivangdavilaSelf-reflection + Self-criticism + Self-learning + Self-organizing memory. Agent evaluates its own work, catches mistakes, and improves permanently. Use when...
Proactive Agent
@halthelobsterTransform AI agents from task-followers into proactive partners that anticipate needs and continuously improve. Now with WAL Protocol, Working Buffer, Autonomous Crons, and battle-tested patterns. Part of the Hal Stack 🦞
Agent Browser
@matrixyHeadless browser automation CLI optimized for AI agents with accessibility tree snapshots and ref-based element selection