Type check: 75 TypeScript errors on main, classified into 5 fixable clusters

[prefrontalsys]

Summary

npm run check on main (currently @ 502b0f4) reports 75 TypeScript errors across 24 files. They're not blocking the build (vite build still succeeds via Svelte's loose-by-default mode), but they hide real type-drift between $lib/types and its consumers, plus a few missing dependencies. Cleaning them up gets the project to a green tsc --noEmit baseline so future drift is caught immediately.

I ran the classification while touching ModelConfigEditor for #213 — happy to send the fixes if useful. Wanted to surface the picture and check scope/priority before opening 5 PRs that might collide with work-in-progress.

Reproduction

git checkout main
npm install
npm run check 2>&1 | grep -c '^[0-9]\+ ERROR'
# → 75

Classification

I grouped errors by root cause. Each group is independently fixable; numbers are the count contributed to the 75 total.

Group	Errors	Files	Root cause
A. SubAgent / ParsedSubAgent type drift	26	SubAgentForm.svelte (1)	Component reads disallowedTools, maxTurns, memory, background, effort, isolation, hooks, mcpServers, initialPrompt — none declared on the SubAgent / ParsedSubAgent interfaces.
B. Containers feature	10	9	Missing DockerStatusType export, Container interface drift (isFavorite, dockerHostId etc. used at call sites but absent on type), missing @xterm/xterm and @xterm/addon-fit deps.
C. Auto-mode settings fields	9	AutoModeEditor.svelte, SettingsAutoModeTab.svelte	ClaudeSettings missing disableAutoMode, autoModeEnvironment, autoModeAllow, autoModeSoftDeny.
D. Lucide icon Component<…> mismatch	7	ProjectCard, ProjectList, McpLibrary	lucide-svelte types out of sync with current Svelte runes — RefreshCw, ExternalLink, Trash_2, Package, FolderOpen all flagged.
E. Keybindings test: "Input" context	6	keybindingsLibrary.test.ts	KeybindingContext union doesn't include "Input" but tests assert it.
F. ModelOverrides setting field	3	ModelOverridesEditor.svelte	ClaudeSettings.modelOverrides not declared on the interface.
G. Skill type (SkillType, skillType)	3	markdownParser.ts, skillLibrary.test.ts	SkillType not exported; skillType field used in test fixture but absent on CreateSkillRequest.
H. Sessions store rename	2	routes/sessions/+page.svelte	selectedSessionSummary + "sessionSummary" prop no longer exist on SessionStoreState / target component.
I. ProjectCard toast variant	1	ProjectCard.svelte	"info" not in toast variant union.
J. Spinner verbs test fixture	1	spinnerverbs.test.ts	Fixture passes sortOrder, missing from SpinnerVerb.
K. Tauri InvokeArgs mock	1	helpers/invokeMock.ts	Tauri tightened InvokeArgs; helper still uses Record<string, unknown>.
L. Sandbox type missing	1	SandboxFilesystemEditor.svelte	SandboxFiles* not exported from $lib/types.
M. Subagents selectedProject	1	AgentMemoryPanel.svelte	ProjectsState.selectedProject missing — store restructured.

Proposed PR sequence

I'd batch these into ~5 PRs to keep review surface small and reverts cheap if any fix turns out to be wrong about intent:

PR	Groups	Errors fixed	Risk	Notes
PR 1 — quick wins	C, F, G, I, J, L	18	Low	All additive: add missing fields/exports to $lib/types. No behavior change.
PR 2 — SubAgent type drift	A	26	Low	Extend SubAgent / ParsedSubAgent to match what the form already reads. Largest payoff per PR.
PR 3 — Lucide icon types	D	7	Low	Bump lucide-svelte to a version aligned with Svelte runes. Smoke-test icon rendering.
PR 4 — test-fixture cleanup	E, H, K, M	10	Low	Touches test files + a couple of call sites. Aligns fixtures with current types.
PR 5 — containers feature	B	10	Med	Touches in-progress containers code. Needs more context on intent. Last in line.

Total recoverable: 65 / 75 in PRs 1–4 (low risk). PR 5 deferred until you confirm the containers feature direction.

Offer

Happy to open these one at a time against main, smallest first, each with its own test plan. Will hold off until I hear back — if any of these are already in flight on your end, just say which groups to skip. Also fine to close this if npm run check is intentionally not gating; in that case the only thing worth doing is updating package.json script to make that explicit.

(For context: I noticed this while sending #213. None of those 75 errors are caused by that PR.)

[tylergraydev]

Thanks for the careful classification — really helpful framing.

Greenlight to send PRs 1–4 in order, smallest first:

• PR 1 (quick wins, groups C/F/G/I/J/L) — go ahead. One thing to double-check on group C: disableAutoMode, autoModeEnvironment, autoModeAllow, autoModeSoftDeny are part of an in-flight Auto Mode settings tab. If those names match what the editor actually reads, just adding them to ClaudeSettings is fine.
• PR 2 (SubAgent type drift, group A) — yes. Big payoff and the form is effectively the source of truth for the shape.
• PR 3 (Lucide icon types, group D) — yes, but coordinate with Dependabot Bump lucide-svelte from 0.577.0 to 1.0.1 #180 which bumps lucide-svelte to 1.0.1. If the bump alone fixes the runes mismatch, your PR may collapse into a smoke-test confirming icons still render. Otherwise rebase on top of Bump lucide-svelte from 0.577.0 to 1.0.1 #180 once it lands.
• PR 4 (test fixture cleanup, groups E/H/K/M) — yes.

Hold on PR 5 (containers, group B). Containers + docker hosts are still scaffolded — they don't return real data yet. Errors there reflect type drift on a feature that's in flux, so fixing them now would just re-drift when the feature lands. Defer until that work has direction.

If anything else surfaces while you're in there (ModelOverridesEditor / AttributionEditor have their own state_referenced_locally warnings I noticed during PR #213 testing), feel free to flag separately rather than expand scope.

Thanks for offering — appreciate it.

[tylergraydev]

Status update — 64/75 errors fixed

Working through the groups from your classification. Net: 75 → 11 errors on main. Every "type drift / missing field / fixture" group is now resolved. What remains is real work that doesn't belong in this issue's scope.

Merged

PR	Group(s)	Errors closed	Approach
#221	C, F, G, I, J, L	22	Additive type fields, plus fixed a spinnerverbs fixture typo (sortOrder → displayOrder) — declined to add a phantom field nothing else reads
#226	A	26	Added 9 fields to SubAgent / CreateSubAgentRequest and 7 fields to ParsedSubAgent. All match the existing Rust struct and round-tripped data
#227	H, K	3	Removed dead sessionSummary prop from the call site; widened invokeMock callback to Tauri's InvokeArgs
#236	E	6	The keybindings tests already pass at runtime — they mock $lib/types to add an Input context. Added one localized Input cast alias in the test instead of adding a phantom value to the production KeybindingContext union
#180	D	7	Bumped lucide-svelte 0.577 → 1.0.1. The lucide bump dropped the Github brand icon, so I added a small GithubIcon.svelte to replace it. Also loosened ActionMenuItem / EmptyState icon prop type — lucide 1.x still exports icons as the legacy SvelteComponentTyped shape, which never satisfies Svelte 5's strict Component<Props>

Remaining (11 errors)

These are not type drift — they're either guarded work or latent bugs:

• Group B — 10 errors, all in containers/*. The containers feature is scaffolded but returns empty data (per project memory: "Containers, docker hosts, templates are scaffolded but return empty data"). These errors will resolve naturally when the feature is built out — fixing them now would require either mocking real shapes or papering over the unfinished code. Recommend leaving until the feature ships.

• Group M — 1 error, AgentMemoryPanel.svelte: projectsStore.selectedProject doesn't exist. This isn't drift — ProjectsState has no project-selection concept at all. The panel's projectPath derives from projectsStore.selectedProject?.path ?? null, which currently always evaluates to null — the project-scoped agent memory silently never receives a path. This is a real runtime bug, not a missing field; the fix is a data-flow decision (prop pass-through vs route param) that belongs in its own focused PR.

Decision

If you want a green npm run check gate, the next step is to either decide on the AgentMemoryPanel data flow (closes M) and add a containers exclusion or progress that feature (closes B). Otherwise I'd consider this issue effectively resolved and close it — the remaining items are tracked by their natural homes (containers feature work, a focused AgentMemoryPanel fix).

🤖 Generated with Claude Code