harness-runtime

Reviewed at commit 082fd08

Caution: Writes a log of your prompts, tool arguments and tool output previews, which can include secrets, to files in ~/.claude/mods/state/events outside your project.

Sends data to TypeSafe (api.typesafe.ai), only when you choose the jev judge backend and set a key

Shared Claude runtime adapter: normalises function-hook events into ClaudeRuntimeEvent, exposes $.runtime (emit bus, supports, judge, snapshot) to other…

Made by
Wes Sander
What we checked

A shared adapter that other plugins build on: it turns Claude Code events into a common event stream and writes every event to a log file in your Claude folder. The log holds prompt previews, tool arguments and short tool output previews. An optional judge setting can send session state to Haiku on your account or to TypeSafe.

Review details

Static review of the source at the pinned commit; not run. Hooks: engine.create (adds a runtime object with emit, supports, judge and snapshot to the engine for later plugins), runtime.judge, session.start, prompt.submit, turn.start, turn.step, tool.check, tool.call, agent.spawn, turn.complete. Every event goes to an in-memory ring of 500 and to a JSONL log at <CLAUDE_CONFIG_DIR or ~/.claude>/mods/state/events/<sessionId>.jsonl (or the eventLogDir option), flushed every 2 seconds and at each turn end by reading the whole file and rewriting it with $.fs.write. Logged content: first 120 characters of each prompt, tool arguments (600 characters), file paths for Read, Write, Edit, Grep, Glob and NotebookEdit, permission check inputs (400 characters) and decisions, first 160 characters of tool results or denial, token usage, cwd and model; nothing is redacted and the files are never deleted. It does not alter tool calls, results or permissions (it returns what next() gave). Judge backends: default stub is local rules; model calls $.model.complete with haiku and the session state (uses tokens on your account); jev posts the state and questions to https://api.typesafe.ai/v1/systemone with the TypeSafe key from the sensitive userConfig option, only when selected and a key is set, otherwise falls back to the stub. Both judge backends are described in plugin.json. It also reads policy settings with $.settings.read to probe features and writes a probe value with $.store. Only another plugin calling $.runtime.judge triggers the judge. npm dependencies: none.

What it uses
  • Reads your prompts: yes
  • Sees tool calls: yes
  • Blocks or changes tool calls: no
  • Reads or writes files: yes
  • Runs programs on your machine: no
  • Calls the model (uses tokens): yes
  • Sends data over the network: yes
  • Stores data: yes
  • Shows something on screen: no
Hooked events

Shared Claude runtime adapter: normalises function-hook events into ClaudeRuntimeEvent, exposes $.runtime (emit bus, supports, judge, snapshot) to other plugins, and writes a JSONL event log

Made this mod? Add a badge to your README

modsforClaude badge for harness-runtime

Paste this into your README. The badge says reviewed once the current version has been reviewed, and links readers to this page.

[![Listed on modsforClaude](https://modsforclaude.com/badge/harness-runtime.svg)](https://modsforclaude.com/mods/harness-runtime/)

← All mods