Pi Enhanced Skills

Enhanced cross-harness skill compatibility for Pi.

Install

pi install npm:@zigai/pi-enhanced-skills

Extension

This package adds compatibility behavior for Anthropic/Claude Code and OpenAI/Codex-style skills.

Invocation Metadata

SKILL.md frontmatter supports:

disable-model-invocation: false
user-invocable: true

• disable-model-invocation: true keeps the skill out of the model-visible skills list.
• user-invocable: false hides the skill from /skill: autocomplete and blocks direct /skill:name use.

OpenAI/Codex sidecars support only this field:

policy:
  allow_implicit_invocation: false

policy.allow_implicit_invocation: false is treated like disable-model-invocation: true for Pi's model prompt. Other OpenAI sidecar fields are ignored.

Dynamic Context Injection

Skill content can include Claude-style shell injection:

Current diff:
!`git diff --stat`

When a known skill is invoked or read in full, the command runs from the skill directory and its output is inserted for that invocation only. The source SKILL.md is not modified. Fenced blocks are also supported:

```!
git status --short
```

Command output is bounded and command execution times out after 10 seconds.

Skill Usage Tracking

Skill usage is appended to <Pi agent dir>/state/pi-enhanced-skills/skill-usage.jsonl as JSONL. Each record includes the session id, current model, project folder, skill name, skill path, file path, and whether the event came from a direct /skill: invocation, a mention-expanded skill block, or an agent read of SKILL.md or a file under the skill directory, with SKILL.md references marked separately. Raw skill contents and user arguments are not stored.

Development

just --justfile ../../Justfile setup
just check
just coverage

License

MIT