octet

Documentation Development

Examples and compatibility status

Find source examples and their compatibility boundaries.

Start with a prompt or skill and copy it into the matching project .octet/ directory. These resources use normal filesystem discovery and typed contribution boundaries:

  • Prompts: Pi-compatible Markdown and compact octet TOML.
  • Skills: explicit, inspectable skills with bounded text resources.
  • Pi migration skill: low-token cleanup around the zero-token Pi inventory.

Current extension authoring#

Use the Python API 0.4 tool recipe for a bounded local process. The canonical API 0.3 example and generated conformance suite remain live, with their exact versions and host pins. This checkout is a local RC, not a new SDK or bundle publication.

Legacy extension examples#

These are legacy implementation references, not current extension authoring quickstarts. New extensions use API 0.4. The Python Extension runtime handles the feature-negotiated API 0.4 wire and retained API 0.1/0.2; generated API 0.3 types serve the distinct canonical wire. Do not retag these examples.

  • hello-world: API 0.1 initialization, model tool, command, hooks, context, status, renderer, and notification.
  • caffeinate: API 0.2 macOS sleep inhibition during owning/root turns, with bounded cleanup.
  • git-tools: API 0.1 read-only checkpoint, bounded git-status tool, semantic status contribution, and renderer. Generic extension status does not become persistent coding-TUI chrome.
  • local-model-workflow: legacy deterministic prompt/context shaping and semantic status for small contexts.
  • lsp-client: legacy read-only LSP definition, references, hover, and diagnostics through one code_intelligence tool with bounded, typed unavailable states.

For an existing legacy setup, install the dependency-free source SDK from a checkout before copying an extension:

console
python3 -m pip install ./sdk/python

Executable extensions require independent enablement and exact trust. They run under the default full-access policy and stay stopped under --safe-mode. Capability declarations are not an OS sandbox; use a trusted, separately isolated environment. See discovery and trust and the legacy Python runtime.