Skills & Tools Overview

A pyhall skill is a governed AI assistant capability that installs as a native command in every major AI platform.

Write one canonical skill manifest. Run pyhall skill publish. Now any user can run:

pyhall skill install yourorg/your-skill

…and their Claude Code, ChatGPT, Gemini, or LangChain agent gains the capability natively — attested, certified, and governed by the WCP chain on every invocation.

pyhall-wcp skill

The pyhall-wcp skill is published and installable now. It teaches any AI assistant how to govern, register, and attest AI workers using the pyhall CLI and WCP SDK.

Install in Claude Code:

claude mcp add pyhall-wcp

Or install the CLI and use the skill directly:

npm install -g @pyhall/cli    # or: pip install pyhall-wcp
pyhall auth login
pyhall skill install pyhall/pyhall-wcp

What the pyhall-wcp skill enables:

• Register AI workers with verified identities and capability declarations
• Check WCP namespaces (x.* for community, org.<name>.* for organizations)
• Request routing decisions — ask pyhall whether a worker is authorized for a capability
• Inspect worker attestation records and verify signed capability cards
• Manage the community ban list — check, query, or report worker hashes
• Install governed skills from the registry
• Audit the history of routing decisions and denials
• Authenticate via GitHub OAuth (pyhall auth login)

The skill ships with safe execution defaults: read-only commands first, explicit confirmation required for mutations, no raw secrets in output, deny-by-default on ambiguous context.

Think slash commands, with governance

If you’ve used Claude Code’s /skills, you understand the concept: a named capability loads into your AI assistant and it knows how to do something new.

pyhall skills work the same way — except:

• They install across every AI platform, not just one
• Every invocation is attested and audited through the WCP governance gate
• Skills are certified by the registry before anyone can install them
• Publishers and organizations control exactly who can install what

What a skill compiles to

Write once. The pyhall compiler generates native adapters for every platform in your stack:

Platform	Becomes
Claude Code	A /skill slash command
Anthropic MCP	An MCP tool definition
OpenAI / ChatGPT	A function tool / GPT Action
Google Gemini	A function declaration
LangChain	A BaseTool subclass
CrewAI	A Tool definition
AutoGen	A function tool schema
YAML / JSON	Portable schema export

All adapters share the same canonical manifest. None bypass the governance gate.

The meta-demo

The first published pyhall skill is pyhall/pyhall-wcp — a skill that teaches any AI assistant how to use the pyhall CLI and WCP SDK.

When a user installs it:

pyhall skill install pyhall/pyhall-wcp

Their AI assistant can now run pyhall commands natively — register workers, check the ban list, query decisions, pull attestation records — all governed by the pyhall registry itself.

Governance governing its own tooling. The registry certifies the skill that enables the registry.

How it works

skill-manifest.yaml
       │
       ▼
  pyhall compiler
       │
  ┌────┼────────────────────────────┐
  ▼    ▼                            ▼
Claude Code      OpenAI         Gemini
/skill           function tool  function decl
       │
  WCP governance gate
  ├─ verify manifest hash
  ├─ check worker attestation
  ├─ evaluate policy
  ├─ check ban list
  └─ ALLOW or DENY → audit record

Every adapter routes through the governance gate. There is no way to invoke a pyhall skill without going through the WCP chain.

Security model

Deny-by-default. A skill only executes when:

1. Manifest is valid and the adapter signature matches the registry
2. Worker is registered and attested
3. WCP policy controls are satisfied
4. Namespace is authorized for the calling org
5. Skill is not revoked or banned

Execution is denied and audited if any condition fails — including at install time, publish time, and on every individual tool call.

See Security & Governance for the full control model.

CLI lifecycle

pyhall skill init        # scaffold a new skill manifest
pyhall skill build       # compile adapters for all target platforms
pyhall skill certify     # submit for registry certification
pyhall skill publish     # publish certified skill to registry
pyhall skill install     # install a skill into your AI assistant
pyhall skill verify      # check integrity and drift since last certify

See the Platform Matrix for platform compatibility and Security & Governance for the full control model.