A research-first Openclaw Skills workflow that turns public sources into a portable, evidence-grounded style model and runtime skill.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install imitation-author
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install imitation-author using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Imitation Author?
Imitation Author is an Openclaw Skills protocol for researching a person, brand, publication, creator, or other style-bearing target and translating that research into a reusable, self-contained skill. It emphasizes broad discovery, saturation evidence, and a claim-backed style taxonomy so a fresh AI can draft, critique, or revise work without reopening the source corpus.
Unlike simple tone cloning, this skill models what the target notices, how it reasons, how it composes, and how its language behaves. It also enforces authorization boundaries, holds out evidence for evaluation, and produces a portable SKILL.md that supports original generation, critique, and style analysis.
Imitation Author Use Cases
- Building a reusable style guide for a creator, brand, publication, or product voice.
- Turning broad public research into a portable Openclaw Skills runtime skill.
- Testing whether AI-generated work matches a target across topic, channel, and time.
- Generating high-fidelity but evidence-backed samples from an authorized, deceased, or public-domain target.
- Creating high-level inspired-by guidance for public-living targets without impersonation.
- Comparing candidate outputs against held-out evidence and a generic negative control.
- Auditing where the target differs from genre conventions or adjacent creators.
How Imitation Author Works
- Frame the run by defining the target, intended output, authorization boundary, material channels, requested topic, and what counts as a useful portable skill.
- Discover the source universe across official properties, archives, feeds, public accounts, works, interviews, transcripts, videos, product surfaces, press, criticism, and adjacent genre examples.
- Build a deliberate corpus by labeling each source as inference, held-out, context, or counter, then deduplicating mirrors, syndication, excerpts, and transcript copies.
- Research to saturation by continuing multi-pass discovery until coverage is sufficient or documented gaps remain and two consecutive passes add no new high-confidence traits.
- Build the evidence and claim registry, assigning Claim IDs and separating target-specific signals from genre conventions, channel effects, and outdated material.
- Model the style with a taxonomy that covers selection, worldview, reasoning, composition, linguistic fingerprint, variation, genre baseline, and anti-caricature rules.
- Write the portable runtime skill in
SKILL.mdso a fresh AI can generate, rewrite, or critique work without reopening the research files. - Evaluate on unseen evidence using held-out artifacts, a requested-topic sample, a cross-topic sample, a negative control, and separate scores for fidelity, naturalness, originality, safety, and portability.
Imitation Author Setup
- Make sure the required protocol files are available before starting:
ls references/research-protocol.md references/evaluation-protocol.md
- Create the canonical target workspace for Openclaw Skills output:
TARGET_SLUG=<target-slug>
mkdir -p "generated-skills/${TARGET_SLUG}"/{research,outputs,evaluations}
touch "generated-skills/${TARGET_SLUG}"/{SKILL.md,README.md}
Run the workflow in the required order: research protocol first, then evaluation protocol, then generate the runtime skill.
Fill the templates in sequence: evidence, style taxonomy, generated skill, target README, and evaluation.
Verify the target folder contains exactly one runtime
SKILL.md, while supporting research stays inresearch/, generated examples stay inoutputs/, and scored artifacts stay inevaluations/.If the target is a living person without authorization, configure the run for high-level inspired-by output only; do not plan for impersonation or misleading attribution.
Imitation Author Data Schema & Taxonomy
Input Schema
| Field | Type | Purpose |
|---|---|---|
target |
string | Style-bearing person, brand, organization, publication, creator/channel, work, or named style to model. |
target_type |
enum | One of person, brand, organization, publication, creator-channel, product-interface, book-work, series-franchise, campaign-event, movement-style, community-subculture, corpus, format, or unknown. |
style_domains |
enum or array | Writing, voice, editorial, social, visual, product, interface, audio, video, motion, spatial, or mixed. |
authorization_status |
enum | One of owned, authorized, deceased, public-domain, public-living, or unknown. |
intended_output |
string | What a future AI should be able to create or critique. |
requested_topic |
optional string | Triggers topic-specific discovery for the first sample. |
source_hints |
optional array |
URLs, files, archives, works, talks, feeds, accounts, or channels to seed discovery. |
depth |
enum | standard or deep; deep mode requires broader coverage and stricter evaluation. |
Canonical Target Layout
| Path | Contents | Role |
|---|---|---|
generated-skills/<target-slug>/README.md |
Human-readable overview | Explains the target folder and usage. |
generated-skills/<target-slug>/SKILL.md |
Runtime skill | The only executable skill file for the target. |
generated-skills/<target-slug>/research/evidence.md |
Evidence registry | Stores claim-backed source references and support. |
generated-skills/<target-slug>/research/style-model.md |
Style taxonomy | Captures selection, worldview, reasoning, composition, and linguistic fingerprint. |
generated-skills/<target-slug>/research/metrics.md |
Coverage and confidence notes | Tracks saturation, gaps, and evidence floors. |
generated-skills/<target-slug>/outputs/<original-artifact>.md |
Example generation | Sample output created from the runtime skill. |
generated-skills/<target-slug>/evaluations/<artifact-evaluation>.md |
Evaluation record | Holds scored comparisons against held-out evidence and controls. |
Metadata Taxonomy
- Evidence tiers:
Tier A: complete or substantial first-party artifact.Tier B: verified short sample, excerpt, transcript segment, or observable sequence.Tier C: metadata, summary, publisher copy, search snippet, or reception context.
- Corpus roles:
inference: may support style claims.held-out: reserved before synthesis and never used to write the taxonomy.context: biography, metadata, reception, or criticism.counter: same genre or topic by another source, used to identify generic conventions.
- Claim metadata:
Claim IDfor every major rule.- Scope markers for channel, time period, topic, and confidence.
- Counterevidence notes when the claim is limited or contested.
- Quality metadata:
- Discovery Query Log entries.
- Saturation Log stop condition.
- Gap annotations for inaccessible, deleted, paywalled, or missing sources.
Imitation Author Advanced Features
- Saturation-based multi-pass discovery with explicit query logging and stop rules.
- Claim IDs and cross-context support so Openclaw Skills outputs stay auditable and portable.
- Stratified held-out evaluation, negative controls, and phrase-overlap checks to reduce overfitting.
- Channel, time, and topic calibration so one style model can adapt across formats without collapsing into a single voice.
- Anti-Caricature modeling with frequency limits on tics, catchphrases, and copied phrasing.
- Authorization-aware output modes that allow close imitation only when it is owned, authorized, deceased, or public-domain.
- Fresh-context evaluation support for testing whether the skill works without access to the original research files.
- Portable runtime design that packages the complete operating model inside
SKILL.mdfor downstream AI reuse. - Explicit separation of target-specific signals from genre conventions, reducing false positives and style leakage.
- A built-in editing rubric for revising weak outputs instead of inflating them with more surface-level mimicry.
SKILL.md
Loading
Related Openclaw Skills
Yanfu
Yanfu, an Openclaw Skills workflow, rewrites an existing public landing page and 1–5 screenshots into a clear, evidence-backed landing page story without changing product truth.
Cron Job Tester
Cron Job Tester is an Openclaw Skills CLI that simulates cron schedules, previews the next 10 runs, and validates shell scripts before deployment.
Workflow DAG Validator
Openclaw Skills Workflow DAG Validator finds DAG cycles, missing dependencies, and parameter mismatches before bad workflows ship.
Signalis
Signalis provides real-time global intelligence and AI-business narrative tracking for autonomous agents using pay-per-call blockchain payments.
Agentic Framework Auditor
Agentic Framework Auditor in Openclaw Skills scans agentic projects for prompt drift, instruction conflicts, injection exposure, and production-readiness gaps.
Doubao Watermark Removal
Openclaw Skills automates batch Doubao watermark removal for AI images using OpenCV inpainting and configurable rectangular masks.
Featured*
