# AI use cases for DSH Crew (ZSeven-W/dsh-crew) Repo: [ZSeven-W/dsh-crew](https://github.com/ZSeven-W/dsh-crew) | Use cases: 4 | Page: https://dedupedai.com/use-cases/zseven-w-dsh-crew/ DSH Crew is a plugin that lets Claude Code, Codex, Antigravity and Grok dispatch work to DeepSeek Harness (DSH) agents while keeping the host's native subagent progress UI. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Send mechanical coding tasks from Claude Code to DSH flash workers: https://dedupedai.com/use-cases/zseven-w-dsh-crew/offload-mechanical-tasks-to-flash-workers/ - Run many translation or batch jobs in parallel and track them on one board: https://dedupedai.com/use-cases/zseven-w-dsh-crew/fan-out-parallel-translation-jobs/ - Let text-only DSH agents describe images through a vision fallback: https://dedupedai.com/use-cases/zseven-w-dsh-crew/give-text-only-harness-image-understanding/ - Run DSH workers in CI or on a machine with no DSH instance running: https://dedupedai.com/use-cases/zseven-w-dsh-crew/run-dsh-workers-headless-in-ci/ ## How to send mechanical coding tasks from Claude Code to DSH flash workers with DSH Crew Repo: [ZSeven-W/dsh-crew](https://github.com/ZSeven-W/dsh-crew) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README Hand simple, well-scoped jobs to cheap DSH flash workers from Claude Code and watch them in the host's task panel. **Who it is for:** A developer in Claude Code who has routine edits, renames, or formatting work and wants it done without spending the orchestrator's own model on it. **The problem:** Simple tasks tie up the main session and cost more than they need to, and there is no clear view of what a delegated job is doing. **The outcome:** Routine tasks run on a DSH flash worker, the dispatch count and progress appear in Claude Code's task UI, and the orchestrator keeps its own model. ### Steps 1. Install the plugin into a DSH profile: `dsh plugin --profile web add @zseven-w/dsh-crew@latest` 2. Start the DSH web instance: `dsh web` 3. Run the smoke test and confirm it prints `smoke test passed, configuration OK`: `node scripts/smoke.mjs` (inferred, check the README) 4. Install the Claude Code integration: `node src/install/cli.mjs claude` 5. Restart the Claude Code session, then say "dispatch to ds-flash" and check the task panel for progress ### Prompt to paste into your agent ```text In the dsh-crew repo (ZSeven-W/dsh-crew), install the Claude Code integration with `node src/install/cli.mjs claude`, restart the session, and dispatch to ds-flash. Report the worker's tier, progress, and final result. Verify the job appears in Claude Code's task UI before reporting success. ``` ### Caveat The README says to restart the Claude Code session after installation, and changes do not take effect until you do. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/zseven-w-dsh-crew/offload-mechanical-tasks-to-flash-workers/ ## How to run many translation or batch jobs in parallel and track them on one board with DSH Crew Repo: [ZSeven-W/dsh-crew](https://github.com/ZSeven-W/dsh-crew) | Difficulty: Intermediate | Time: 45 minutes | Basis: From the README Split a large batch of text into independent jobs, dispatch them at once, and monitor them in the DSH Crew panel. **Who it is for:** A maintainer with a large document or docs set to translate or transform into several languages, who needs the work done in parallel rather than one file at a time. **The problem:** Running many similar jobs one by one is slow, and it is hard to see which jobs are still running, failed, or were lost. **The outcome:** Each file or language becomes its own worker job, all run at once, and the DSH Crew panel shows tier, progress, tokens, and any orphaned jobs. ### Steps 1. Install and verify the plugin as in the install steps: `dsh plugin --profile web add @zseven-w/dsh-crew@latest`, then `dsh web` 2. Install the Claude Code or Codex integration: `node src/install/cli.mjs claude` or `node src/install/cli.mjs codex` 3. Write one self-contained brief per target language or file so each worker can run without extra context 4. Dispatch all briefs at once from the host, for example by saying "dispatch to ds-flash" for each one 5. Open Settings → DSH Crew or the DSH Crew panel and check that every job is listed with progress and tokens ### Prompt to paste into your agent ```text Using the dsh-crew repo (ZSeven-W/dsh-crew), dispatch one independent translation job per target file in to ds-flash, all in parallel. Expected output: one translated file per target and a list of job statuses. Verify each job is listed in the DSH Crew panel and that no job was refused by the workspace guardrail. ``` ### Caveat Dispatch guardrails refuse a second worker on a workspace another job already holds, so parallel jobs must be scoped to different working areas. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/zseven-w-dsh-crew/fan-out-parallel-translation-jobs/ ## How to let text-only DSH agents describe images through a vision fallback with DSH Crew Repo: [ZSeven-W/dsh-crew](https://github.com/ZSeven-W/dsh-crew) | Difficulty: Advanced | Time: 1 hour | Basis: From the README Use the multimodal bridge so DSH workers can describe images by calling DeepSeek's vision model or an installed CLI. **Who it is for:** A developer whose DSH workers need to read screenshots or diagrams, but whose DSH models are text-only. **The problem:** DSH's models cannot read images, so a worker given a screenshot or diagram has no way to understand it. **The outcome:** The worker calls describe_image, which uses DeepSeek's VL model when a key is available or falls back to a CLI or OpenAI-compatible API, and returns a text description. ### Steps 1. Install and start the plugin: `dsh plugin --profile web add @zseven-w/dsh-crew@latest`, then `dsh web` 2. Open Settings → DSH Crew and review the multimodal bridge and custom provider settings 3. Add an OpenAI-compatible endpoint (Base URL, API key, models) under Custom Providers if you want one, then run its connectivity test 4. Dispatch a worker with an image path and ask it to call describe_image on that image 5. Check that the returned text description matches the image content ### Prompt to paste into your agent ```text Using the dsh-crew repo (ZSeven-W/dsh-crew), dispatch a worker that calls describe_image on . Expected output: a text description of the image. Verify the connectivity test for the configured provider passed before relying on the result. ``` ### Caveat describe_image prefers DeepSeek's VL model when a key is available, so the behavior depends on which keys and CLIs you have configured. Risk, credentials: A DeepSeek key or a custom provider API key is needed for the vision path, so store keys carefully and do not paste them into prompts. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/zseven-w-dsh-crew/give-text-only-harness-image-understanding/ ## How to run DSH workers in CI or on a machine with no DSH instance running with DSH Crew Repo: [ZSeven-W/dsh-crew](https://github.com/ZSeven-W/dsh-crew) | Difficulty: Advanced | Time: 45 minutes | Basis: From the README Set up the standalone fallback with a DeepSeek key so dispatch works in headless environments. **Who it is for:** A DevOps engineer who wants scripted or automated coding jobs on a build server or CI runner where no DSH web instance is running. **The problem:** Without a running DSH instance, the hub path is unavailable, so dispatch needs its own runtime and credentials. **The outcome:** Dispatch falls back to a standalone DSH runtime, and the smoke test reports which path it used and whether configuration is OK. ### Steps 1. Obtain an API key from platform.deepseek.com 2. Write the key to `~/.config/dsh-crew/.env` as `DEEPSEEK_API_KEY=sk-...` 3. Install the plugin from npm: `dsh plugin --profile web add @zseven-w/dsh-crew@latest` 4. Run the smoke test: `node scripts/smoke.mjs` 5. Confirm the output says `smoke test passed, configuration OK` and note which path was used (inferred, check the README) ### Prompt to paste into your agent ```text In the dsh-crew repo (ZSeven-W/dsh-crew), configure the standalone fallback with a DEEPSEEK_API_KEY in ~/.config/dsh-crew/.env and run `node scripts/smoke.mjs`. Expected output: the smoke test line that shows which path ran and whether configuration passed. Verify it reports `smoke test passed` before any scripted dispatch. ``` ### Caveat The README says the standalone fallback needs its own key, while the hub mode uses the DSH instance's existing credentials. Risk, credentials: The DeepSeek API key is stored in a plain .env file on the machine, so restrict its file permissions and keep it out of version control. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/zseven-w-dsh-crew/run-dsh-workers-headless-in-ci/