Every plugin test in this folder follows this methodology.
Each per-plugin file (Documentation/Testing/<package>.md) is an instance of this template and uses
the README.md, SCHEME.md, Source/Variable/Config.ts, cordis.patch.yml, and profile row of
that package as its base.
The 35 new packages are enabled in the desktop profile (~/.dsh/profiles/desktop).
Each package name appears in the profile package.json under dsh.profile.bundles, its row exists
in the profile cordis.yml (resolved from the insert: block of each bundle's cordis.patch.yml),
and the app was restarted so that the bundle layers composed at host boot.
The live app runs them.
A file: dependency alone is not activation; the restart is.
Before any behavior test, prove that the plugin activated:
-
Config schema.
cordis_inspect_querywith platformhost, providerConfig, methodlistConfigs, input{"name": "@playform/<pkg>"}→ exactly one entry whose status is"schema"(the Schemastery schema was validated at load, and every default was filled). -
Tools.
For tool plugins, the Tool inspect provider lists the tool under its tool surface (
cordis_inspect_listfirst, then the list method of the tool provider).A missing tool means that the tool registration did not run or the plugin is disabled.
-
Ledger line.
The
activated (...)string of the package (quoted byte-exact in each per-plugin file) appears in its logFile (thelogFileConfig default, usually~/.dsh/<name>.log) or in the logger.A cold mount instead logs the byte-exact
disabled (enabled=false)line, and that line means the row is present but the plugin is disabled. -
Services.
For service-provider bundles (
plugin-dsh-feedback,plugin-dsh-ingester,plugin-dsh-pass-on,plugin-dsh-queue,plugin-dsh-route,plugin-dsh-workspace-*),cordis_inspect_queryhost/Service/listService<service>(e.g.feedback,workspaceGraph,dshQueue) must resolve.
| Tier | Meaning | Examples |
|---|---|---|
| T1 | Read-only | inspect queries, read-only tools, git status-class commands |
| T2 | Safe writes | a /tmp scratch file through batch_edit; a temp repo for git |
| T3 | Guards | the govern gate on a tool call, the suggesters' notes, the observe stamps |
| T4 | Flows | the batch decision → pass-on → ingester chain |
| T5 | Client | the UI tabs - the user checks these by eye |
-
Failure.
The failure → capture the error text verbatim.
-
Likely cause.
The likely cause → read the Source of the failing package.
-
Fix file.
The fix file → use the package
Source/(one definition per file; never the builtTarget/). -
Rebuild.
Rebuild, and the flag is mandatory:
pnpm --config.verify-deps-before-run=false --dir Classic/packages/<pkg> run prepublishOnly -
Smokes.
Keep the smokes green: for each test folder,
node Classic/packages/<pkg>/Target/Test/<Folder>/Test.jsmust print the exactN PASS - <Name>count from its README. -
Types.
Keep the types green:
pnpm test:types(the honest repo-level type gate) must reportPASSfor the package; seeDocumentation/Plans/Completion-Spec/06-VERIFICATION.mdcheck 3b. -
Restart.
Restart the app, because a code fix does not hot-reload, and then re-run the live test.
- Plugin. Package name, profile row id, logFile.
- Tests run. Pass/fail per test, with the observed output quoted.
- Defects.
file:line - cause - fix. - Rebuild/smoke result. The build command outcome and the smoke counts.
- Residuals. What is still broken or untested, and why.
- Never run a destructive git command to test the gitguard: use a scratch git repo under
/tmp, or a harmless deny-listed command. batch_editwrites only/tmpscratch files during testing.- Never test against the live profile config of the user: the Config checks are read-only inspect queries.
- The Lead disables a plugin that crashes the boot; the tester does not.
plugin-dsh-batch-editinject defect - FIXED. The first dogfood failed every call withcannot get property "fs" without inject;"fs"is now in the inject list, and the batch engine runs.- Workspace graph indexing - FIXED. The graph no longer relies on a missing trigger, because
plugin-dsh-workspaceexposes theworkspace_indextool (path?, maxFiles?) that drivesindexWorkspaceof the graph service, soworkspace_map/workspace_findreturn real data after an index. batch_decisionworks. The empty pending list is the correct answer.