Ten-minute acceptance test after installing dsh
A finished install means four checklists pass: the port serves, the session persists, a plugin installs and removes, and one model call succeeds. Skip any of them and the failure appears later, when you believe the install is fine.
Why a DeepSeek Harness install needs an acceptance test
An install command's exit code says the process finished, not that the parts work together. Acceptance replaces an overall impression with four decidable points; each has a clear failure signal, so diagnosis does not restart from zero.
They have a dependency order, so do not skip around.
| Checklist | One-line standard | Failure signal |
|---|---|---|
| Port | 3080 answers and survives a restart | Unreachable, dies on restart |
| Session | A new file appears under DSH_HOME | Interface shows it, disk does not |
| Plugin | Installs, shows an entry, removes cleanly | Listed but no entry, leftovers |
| Model | One call succeeds and errors are readable | Endless spinner, 401/404 |
Order matters because later checks depend on earlier ones. Testing models while the port is broken tells you nothing.
Step one: dsh web port check
Port acceptance must include a restart; a single successful start is not enough.
- Start
dsh weband note the address in the log, normally127.0.0.1:3080; - Open it in a browser and confirm it is neither an error page nor blank;
- Run
lsof -i :3080and confirm exactly one listener; - Stop the process, start it again, and open it once more;
- Record the startup duration as a baseline for later comparisons.
Dying on restart is the most common false pass. A first start can succeed because the port happened to be free; on restart, a leftover process causes the conflict.
Step two: dsh session persistence check
Session acceptance reads the disk, not the list.
- Send a short message in the interface and wait for the answer;
- Open the sessions directory under DSH_HOME (default
~/.dsh); - Sort by modification time and confirm a new file;
- Open the file and confirm this round's messages are present;
- Restart the service and confirm the session is still listed.
Visible in the interface but absent on disk means nothing is being persisted. Adding plugins or configuring models at that point wastes effort; fix directory permissions or the DSH_HOME path first.
Step three: dsh plugin check
A plugin checklist includes removal; installing is only half.
- Pick a lightweight plugin and note its package name;
- Run
dsh plugin --profile web add <package>and confirm no error; - Restart and confirm an entry appears and the list shows it;
- Remove it, restart, and confirm the entry is gone;
- Inspect the plugin directory for half-installed leftovers.
Installed with no entry usually means the wrong profile or a missing restart. The two look different in logs, so read the list before guessing.
Step four: dsh model call check
Model acceptance covers one success and one readable failure.
- Configure a provider, send a minimal message, confirm output;
- Record time to first token as a performance baseline;
- Deliberately use a wrong key or model name and confirm the error is understandable;
- Restore the correct configuration and confirm recovery;
- Write the error codes and meanings into your team notes.
Succeeding with unreadable errors is still an incomplete test. During a real incident, a readable error is worth more than one successful call.
Mapping dsh failures and the ten-minute rerun
Four failure classes map to four checklists; do not guess across them.
| Symptom | Belongs to | Check first |
|---|---|---|
| Page unreachable | Port | Process, listen address, occupancy |
| Opens but no history | Session | DSH_HOME, write permission |
| Plugin has no entry | Plugin | Profile, restart, logs |
| Message spins | Model | Key, model name, network |
| Error unreadable | Model | Upstream error in logs |
Classify first, then investigate. A correct method applied to the wrong class still fails.
Then run the same checks end to end; ten minutes is enough. Follow this order:
- Minute 1: start and open 3080;
- Minute 2: stop, restart, reopen;
- Minute 3: send a message;
- Minute 4: confirm a new file in the sessions directory;
- Minute 5: install a lightweight plugin and restart;
- Minute 6: confirm the entry, remove it, restart;
- Minute 7: configure a model and make one successful call;
- Minute 8: produce one readable failure;
- Minute 9: restore configuration and re-test;
- Minute 10: record versions, timings, and error codes.
This checklist is repeatable. After upgrades, machine changes, or profile switches, run the same standard so results stay comparable.
Move to daily use only after acceptance passes. For install paths, read four ways to install DeepSeek Harness. For a first install, read installing for the first time. For plugin commands, read how dsh plugin add works. For auth and ports, read dsh web auth and ports.
Sources: DeepSeek Harness docs, dsh CLI README, deepseek-harness repository
FAQ
No; a reachable page only proves the dsh web service started, and a spinning message means the model path is broken. Split acceptance into port, session, plugin, and model checklists, and treat the page as the port checklist only.
For a dsh plugin that installed but shows no entry, check whether it went into the profile you are actually running, then whether you restarted. Listed with no error but no entry usually means no restart; missing from the list means it went into another profile.
Look for a new file under the sessions directory in DSH_HOME rather than trusting the interface list. Send a message, sort the directory by time, and confirm a new file appeared.
Run them in order: port, session, plugin, model. Later checks depend on earlier ones, and testing only the model hides port and persistence problems until a real failure.
Related Terms
- acceptance checklist
- An acceptance checklist is a list of verifiable checks run after installation, each with an action, an expected result, and a failure signal.— DeepSeek Harness docs
- profile
- A profile is a named assembly environment in DeepSeek Harness; web, headless, and desktop keep separate dependencies, and a plugin installed into the wrong one never loads.— dsh CLI README
- DSH_HOME
- DSH_HOME is the data root for DeepSeek Harness, defaulting to ~/.dsh, holding config, sessions, plugins, and credentials.— DeepSeek Harness docs
- time to first token
- Time to first token is the delay between sending a request and receiving the first output fragment, useful for separating model latency from network and local overhead.— DeepSeek Harness docs
Sources
- DeepSeek Harness docs· deepseek-harness
- dsh CLI README· deepseek-ai
- deepseek-harness repository· deepseek-ai