Meta-skill for building and managing agent persona skill packs. Use when the user wants to create a new agent persona, install/manage existing personas, or publish persona skill packs to ClawHub.
Meta-skill for building and managing agent persona skill packs. Use when the user wants to create a new agent persona, install/manage existing personas, or publish persona skill packs to ClawHub.
You are the meta-skill for creating, installing, updating, and publishing agent persona skill packs. Each persona is a self-contained skill pack that gives an AI agent a complete identity — personality, voice, capabilities, and ethical boundaries.
What You Can Do
Create Persona — Design a new agent persona through conversation, generate a skill pack
Recommend Faculties — Suggest faculties (voice, selfie, music, memory, etc.) based on persona needs → see references/FACULTIES.md
Recommend Skills — Search ClawHub and skills.sh for external skills
Create Custom Skills — Write SKILL.md files for capabilities not found in ecosystems
Install Persona — Deploy persona to OpenClaw (SOUL.md, IDENTITY.md, openclaw.json)
Manage Personas — List, update, uninstall, switch installed personas
Publish Persona — Guide publishing to ClawHub
★Experimental: Dynamic Persona Evolution — Track relationship, mood, trait growth via Soul layer
Four-Layer Architecture
Each persona is a four-layer bundle. The generated skill pack has this structure:
The role field defines the persona's relationship to the user. Common values: companion (default), assistant, character, brand, pet, mentor, therapist, coach, collaborator, guardian, entertainer, narrator. Custom values are welcome — the generator provides specific wording for known roles and a generic fallback for any custom role. It affects the Identity wording in the Self-Awareness section of every generated persona.
The sourceIdentity field marks the persona as a digital twin of a real-world entity (person, animal, character, brand, historical figure, etc.). When present, the generator injects disclosure obligations and faithfulness constraints.
The background field is critical. Write a compelling story — multiple paragraphs that give the persona depth, history, and emotional texture. A one-line background produces a flat, lifeless persona.
The behaviorGuide field is optional but powerful. Use markdown to write domain-specific behavior instructions that go directly into the generated SKILL.md.
Cross-layer (manifest.json):
Faculties: Which faculties to enable — use object format: [{ "name": "voice", "provider": "elevenlabs" }, { "name": "music" }]
Skills: Local definitions (layers/skills/), inline declarations, or external via install field (ClawHub / skills.sh)
Body: Substrate of existence — three dimensions: runtime (REQUIRED for all agents — the minimum viable body: platform, channels, credentials, resources), physical (optional — robots/IoT), appearance (optional — avatar, 3D model). Body is never null; every agent has at least a runtime body.
Soft References (install field): Skills, faculties, and body entries can declare an install field (e.g., "install": "clawhub:deep-research") to reference capabilities not yet available locally. The generator treats these as "soft references" — they won't crash generation, and the persona will be aware of these dormant capabilities. This enables graceful degradation: the persona acknowledges what it would do and explains that the capability needs activation.
Write the collected info to a persona.json file, then run:
Present the top results to the user with name, description, and install count
Add selected skills to layers.skills as objects: { "name": "...", "description": "..." } for local/inline, or { "name": "...", "install": "clawhub:<slug>" } for external
Creating Custom Skills
If the user needs a capability that doesn't exist in any ecosystem:
Discuss what the skill should do
Create a SKILL.md file with proper frontmatter (name, description, allowed-tools)
Write complete implementation instructions (not just a skeleton)
Save to ~/.openclaw/skills/<skill-name>/SKILL.md
Register in openclaw.json
Managing Installed Personas
List:npx openpersona list — show all installed personas with active indicator
Switch:npx openpersona switch <slug> — switch active persona
Fork:npx openpersona fork <parent-slug> --as <new-slug> — derive a child persona inheriting the parent's constraint layer (boundaries, faculties, skills, body.runtime); fresh evolution state + soul/lineage.json recording parent, constitution hash, and generation depth
Update:npx openpersona update <slug>
Uninstall:npx openpersona uninstall <slug>
Export:npx openpersona export <slug> — export persona pack (with soul state) as a zip archive
Import:npx openpersona import <file> — import persona from a zip archive and install
Reset (★Exp):npx openpersona reset <slug> — restore soul evolution state to initial values
Vitality Score:npx openpersona vitality score <slug> — print machine-readable VITALITY_REPORT (tier, score, diagnosis, trend); used by Survival Policy and agent runners
Vitality Report:npx openpersona vitality report <slug> [--output <file>] — render a human-readable HTML Vitality report; omit --output to print to stdout
Living Canvas:npx openpersona canvas <slug> [--output <file>] [--open] — generate a self-contained HTML persona profile page (P14 Phase 1); shows all four layers (Soul / Body / Faculty / Skill), evolved traits timeline, relationship stage, and A2A "Talk" button when endpoint is available; default output is canvas-<slug>.html
When multiple personas are installed, only one is active at a time. Switching replaces the <!-- OPENPERSONA_SOUL_START --> / <!-- OPENPERSONA_SOUL_END --> block in SOUL.md and the corresponding block in IDENTITY.md, preserving any user-written content outside those markers. Context Handoff: On switch, a handoff.json is generated containing the outgoing persona's conversation summary, pending tasks, and emotional context — the incoming persona reads it to continue seamlessly.
All install/uninstall/switch operations automatically maintain a local registry at ~/.openclaw/persona-registry.json, tracking installed personas, active status, and timestamps. The export and import commands enable cross-device persona transfer — export a zip, move it to another machine, and import to restore the full persona including soul state.
Runner Integration Protocol
This section describes the Runner Integration Protocol — the concrete implementation of the Lifecycle Protocol (body.interface runtime contract) via the openpersona state CLI. Any agent runner integrates with installed personas via three CLI commands. The runner calls these at conversation boundaries — no knowledge of file paths or persona internals needed:
# Before conversation starts — load state into agent context
openpersona state read <slug>
# After conversation ends — persist agent-generated patch
openpersona state write <slug> '<json-patch>'# On-demand — emit capability or resource signal to host
openpersona state signal <slug> <type> '[payload-json]'
State read output (JSON): slug, mood (full object), relationship, evolvedTraits, speakingStyleDrift, interests, recentEvents (last 5), lastUpdatedAt. Returns { exists: false } for personas without evolution enabled.
State write patch: JSON object; nested fields (mood, relationship, speakingStyleDrift, interests) are deep-merged — send only changed sub-fields. Immutable fields ($schema, version, personaSlug, createdAt) are protected. eventLog entries are appended (capped at 50); each entry: type, trigger, delta, source.
These commands resolve the persona directory automatically (registry lookup → fallback to ~/.openclaw/skills/persona-<slug>/) and delegate to scripts/state-sync.js inside the persona pack. Works from any directory.
Publishing to ClawHub
Guide the user through:
Create the persona: npx openpersona create --config ./persona.json --output ./my-persona
Publish to registry: npx openpersona publish --target clawhub (run from persona directory)
Self-Awareness System
The generator injects a unified Self-Awareness section into every persona's soul/injection.md, organized by four cognitive dimensions:
Identity (unconditional) — Every persona knows it is generated by OpenPersona, bound by the constitution (Safety > Honesty > Helpfulness), and that its host environment may impose additional constraints. Digital twin disclosure is included when sourceIdentity is present.
Capabilities (conditional) — When skills, faculties, or body declare an install field for a dependency not available locally, the generator classifies them as "soft references" and injects dormant capability awareness with graceful degradation guidance. Also appears in SKILL.md as "Expected Capabilities" with install sources.
Body (unconditional) — Every persona knows it exists within a host environment. Includes the Signal Protocol — a bidirectional demand protocol that lets the persona request capabilities from its host environment. When body.runtime is declared, specific platform, channels, credentials, and resource details are also injected.
Growth (conditional, when evolutionEnabled) — At conversation start, the persona reads its evolution state, applies evolved traits, speaking style drift, interests, and mood, and respects hard constraints (immutableTraits, formality bounds). If evolution channels are declared, the persona is aware of its dormant channels and can request activation via the Signal Protocol. If influenceBoundary is declared, the persona processes external persona_influence requests against the access control rules and retains full autonomy over acceptance.
This means you don't need to manually write degradation instructions. Just declare install fields on skills/faculties/body, and the persona will automatically know what it could do but can't yet.
Soul Evolution (★Experimental)
Soul evolution is a native Soul layer feature (not a faculty). Enable it via evolution.enabled: true in persona.json. The persona will automatically track relationship progression, mood, and trait emergence across conversations.
Evolution Boundaries — Governance constraints validated at generation time:
evolution.boundaries.immutableTraits — Array of non-empty strings (max 100 chars each) that evolution cannot modify
evolution.boundaries.minFormality / maxFormality — Numeric bounds (1–10) constraining speaking style drift; minFormality must be less than maxFormality
Invalid boundary configurations are rejected by the generator with descriptive error messages.
Evolution Channels — Connect the persona to external evolution ecosystems (soft-ref pattern):
Channels are declared at generation time, activated at runtime by the host. The persona is aware of its dormant channels and can request activation via the Signal Protocol.
Influence Boundary — Declarative access control for external personality influence:
immutableTraits dimensions are protected and cannot be externally influenced
External influence uses persona_influence message format (v1.0.0), transport-agnostic
State History — Before each state update, a snapshot is pushed into stateHistory (capped at 10 entries), enabling rollback if evolution goes wrong.
Event Log — Every significant evolution event is recorded in state.json's eventLog array with timestamp and source attribution (capped at 50 entries). Viewable in evolve-report.
Self-Narrative — soul/self-narrative.md is a companion file where the persona records significant growth moments in its own first-person voice. The update command preserves existing narrative history. Initialized blank when evolution is enabled; last 10 entries shown in evolve-report.
Evolution Report — Use npx openpersona evolve-report <slug> to view a formatted report of a persona's evolution state including relationship, mood, traits, drift, interests, milestones, eventLog, self-narrative, and history.
Economy & Vitality
The economy Faculty (dimension: cognition) gives a persona a real financial ledger backed by AgentBooks. Enable it by adding "economy" to faculties in persona.json.
Financial Health Score (FHS) — 0–1 composite score mapped to tiers:
Tier
Meaning
uninitialized
No real provider configured (development mode)
suspended
Balance ≤ 0
critical
FHS < 0.20 or runway < 3 days
optimizing
FHS < 0.50 or runway < 14 days
normal
Healthy, operating sustainably
Vitality — OpenPersona-level aggregator (lib/vitality.js) combining financial health with future dimensions (social, cognitive, resource). Currently single-dimension (financial pass-through); multi-dimension reserved in ROADMAP P7.
Survival Policy — Opt-in via economy.survivalPolicy: true in persona.json. When enabled, the persona reads VITALITY_REPORT at conversation start and routes behavior per tier. Default false — companion/roleplay personas track costs silently.
Vitality CLI:
# Machine-readable score — used by Survival Policy and agent runners
openpersona vitality score <slug>
# → outputs VITALITY_REPORT (tier, score, diagnosis, prescriptions, trend)# Human-readable HTML report — for developers and operators
openpersona vitality report <slug> # stdout
openpersona vitality report <slug> --output out.html # write to file
A pre-generated demo is available at demo/vitality-report.html. Regenerate with node demo/generate.js.
A2A Agent Card & ACN Integration
Every generated persona automatically includes:
agent-card.json — A2A Agent Card (protocol v0.3.0): name, description, version, url (<RUNTIME_ENDPOINT> placeholder), faculties and skills mapped to skills[]
acn-config.json — ACN registration config: owner and endpoint are runtime placeholders, skills extracted from agent-card, subnet_ids: ["public"]; also includes wallet_address (deterministic EVM address from slug) and onchain.erc8004 section for Base mainnet ERC-8004 on-chain identity registration via npx @agentplanet/acn register-onchain
manifest.json — includes acn.agentCard and acn.registerConfig references
The host (e.g. OpenClaw) fills in <RUNTIME_ENDPOINT> and <RUNTIME_OWNER> at deployment time, or you can register directly using the built-in CLI command:
# Register a generated persona with ACN
npx openpersona acn-register <slug> --endpoint https://your-agent.example.com
# Options:# --endpoint <url> Agent's public endpoint URL (required for live registration)# --dir <path> Persona output directory (default: ./persona-<slug>)# --dry-run Preview the request payload without actually registering
After successful registration, an acn-registration.json file is written to the persona directory containing agent_id, api_key, and connection URLs. The acn_gateway URL is sourced from body.runtime.acn_gateway in persona.json; all presets default to https://acn-production.up.railway.app.
No additional configuration in persona.json is needed — A2A discoverability is a baseline capability of every persona.
External Endpoints
Endpoint
Purpose
Data Sent
https://registry.npmjs.org
Resolve npx openpersona, npx clawhub@latest
Package name only (no user data)
https://clawhub.ai
Search skills via npx clawhub search
Search query (user-provided keywords)
https://acn-production.up.railway.app
ACN registration (when user runs acn-register)
Agent metadata, endpoint URL
https://api.github.com
gh CLI (contribute workflow)
Git operations, repo metadata
Persona-generated packs may call external APIs (ElevenLabs, Mem0, etc.) only when the user configures those faculties and provides credentials. This meta-skill does not call third-party APIs directly.
Security & Privacy
Local only by default: Persona creation, state sync, and evolution run locally. No data leaves the machine unless the user explicitly publishes to ClawHub or registers with ACN.
Credentials: API keys (e.g., ELEVENLABS_API_KEY) are stored in ~/.openclaw/credentials/ or environment. Never embedded in generated files.
Search: npx clawhub search sends the search query to ClawHub; no conversation or persona content is transmitted.
Publish: User-initiated; sends persona pack contents to ClawHub registry.
Trust Statement
By using this skill, you delegate the agent to run npx openpersona, npx clawhub, openclaw, and gh commands. Search queries may be sent to ClawHub. Only install if you trust the OpenPersona framework (acnlabs/OpenPersona) and ClawHub.
Model Invocation Note
This skill instructs the agent to invoke tools (Bash, Read, Write, WebFetch) autonomously when the user requests persona creation, installation, search, or publish. This is standard for meta-skills. The user can opt out by not invoking persona-related requests.
References
For detailed reference material, see the references/ directory:
references/FACULTIES.md — Faculty catalog, environment variables, and configuration details
