dsh-record-replay

A DeepSeek Harness plugin that
turns the Open Record/Replay
macOS workflow recorder into first-class harness capabilities.

It registers the open-record-replay skill and six model-facing orr_*
tools, so an agent can learn a user-demonstrated desktop workflow: record the
user's real macOS actions, validate the evidence, and package it for the host
agent's native skill creator.

user demonstrates a workflow
  -> orr_record_start            (capture session.json + events.jsonl)
  -> orr_record_stop             (finalize)
  -> orr_session_validate        (check the evidence against the contract)
  -> orr_session_events          (read what the user actually did)
  -> orr_skill_prepare           (package a skill-input directory)
  -> host skill creator

Requirements

• macOS (the recorder's native backend is Swift; it needs Xcode Command Line Tools).
• Node.js >= 22.19 (the Harness runtime).
• A DeepSeek Harness installation.
• An open-record-replay
  checkout whose bin/orr.js the plugin invokes.

Install

Build the package and add it to a profile as a bundle:

git clone https://github.com/<you>/dsh-record-replay.git
cd dsh-record-replay
pnpm install
pnpm build
pnpm pack                      # produces dsh-record-replay-0.1.0.tgz
dsh plugin --profile web add ./dsh-record-replay-0.1.0.tgz

dsh plugin add records the package in the profile's package.json
dependencies and dsh.profile.bundles, and the harness heals the
profiles/node_modules fallback so the bundle resolves. The shipped
cordis.patch.yml mounts a neutral row; point it at your checkout by overlaying
the row from your profile's cordis.patch.yml:

- id: record-replay
  config:
    repoRoot: '/absolute/path/to/open-record-replay'
    runsOut: 'runs'
    skillInputsOut: 'skill-inputs'

The profile patch file is hot-reloaded, so the running GUI picks the plugin up
without a restart. Restart the Harness if you are not on a live profile.

Configuration

The CLI runs with the session workspace as its working directory, so recordings
and skill packages land where the agent's filesystem tools can read them.

Tools

Development

pnpm install
node scripts/link-dsh.mjs    # links @deepseek-ai/* from a DSH checkout/harness
pnpm build                   # tsc
pnpm test                    # vitest
pnpm validate                # build + test

Tests assert the registration contract and exercise the CLI runner through a
stubbed subprocess service; node scripts/link-dsh.mjs --path <dir> points the
@deepseek-ai scope at an explicit DSH checkout.

License

MIT — see LICENSE.