Primary interface references

Reviewed on September 20, 2026. These are documentation/source-contract checks, not claims that a live installed host accepted this package. This file intentionally separates confirmed public entry points from package limitations.

System	Primary reference	Contract used / qualification
TypeSafe	https://docs.typesafe.ai/introduction/quickstart	POST /v1/systemone; bearer key; state, model, typed questions; answers with Choice distributions/Noul values. Live model availability and paid responses untested.
TypeSafe concepts	https://docs.typesafe.ai/introduction	Typed judgments, not lossless natural-language reconstruction.
Codex hooks	https://learn.chatgpt.com/docs/hooks	hooks.json, command hooks, session/prompt/tool/stop events. Prompt hooks are not a universal active-transcript replacement API.
Codex MCP	https://developers.openai.com/codex/mcp	mcp_servers entries in config.toml; stdio command/arguments.
Claude Code hooks	https://code.claude.com/docs/en/hooks	Settings hook configuration and hookSpecificOutput.additionalContext envelopes. No transcript replacement claimed.
Claude Code MCP	https://code.claude.com/docs/en/mcp	User MCP registry in ~/.claude.json; preserve existing project/server entries.
OpenCode V1 plugins	https://opencode.ai/docs/plugins/	Global/local JS plugins and native V1 hooks. Experimental message/system transforms require live version checks.
OpenCode MCP	https://opencode.ai/docs/mcp-servers/	mcp.NAME local server with command array.
OpenCode V2 plugins	https://opencode.ai/v2/docs/build/plugins/	V2 default id/setup; plural plugins; ctx.session.hook("context"); ctx.tool.hook("execute.after"). Context mutations affect outgoing model call, not durable history.
OpenCode migration	https://opencode.ai/v2/docs/build/plugins/migrate-v1	V1/V2 hooks are not interchangeable. The package supplies separate adapters.
OpenClaw plugin config	https://docs.openclaw.ai/tools/plugin	Local plugin paths, entries, enable/allow/deny policy; this package does not replace a memory slot or install policy.
OpenClaw hook reference	https://docs.openclaw.ai/plugins/hooks/reference	Synchronous registration, async lifecycle handlers where allowed; observer versus mutation distinctions.
OpenClaw prompt authority	https://docs.openclaw.ai/plugins/hooks/prompt-and-session	before_prompt_build, prependContext, requiresToolAuthority, allows, assertActive, optional current-user fields. Package fails closed without authority; no compatibility with older prompt APIs is claimed.
Hermes MCP	https://hermes-agent.nousresearch.com/docs/user-guide/features/mcp/	YAML mcp_servers command/args.
Hermes hooks	https://hermes-agent.nousresearch.com/docs/user-guide/features/hooks/	Shell hooks, JSON envelopes with extra, returned context, first-use approval. pre_llm_call is per user turn, not every tool-loop model request.
Gemini CLI	https://geminicli.com/docs/tools/mcp-server/	mcpServers in settings; package keeps trust: false. No automatic hooks installed.
Cursor	https://cursor.com/docs/context/mcp	User MCP server configuration; package supplies MCP only.
Copilot CLI	https://docs.github.com/en/copilot/how-tos/copilot-cli/customize-copilot/add-mcp-servers	~/.copilot/mcp-config.json, local command/args/tools. Exposure of tools does not bypass native execution approvals.
MCP transport	https://modelcontextprotocol.io/specification/2025-06-18/basic/transports	Stdio newline-delimited JSON-RPC; stdout contains protocol only. Package implements a minimal tools server, not the full SDK feature set.

Harness libraries and exact event shapes can change. Capability claims must be revalidated against the actual host version. A successful installer/config parser test proves file generation, not that a host loaded an API contract. A successful mock-host test proves the package's translation and bridge behavior against those fixtures, not end-to-end native integration.

All native host version fields remain unverified in the installation report. doctor checks availability and memory state, not compatibility certification.