The read-only proxy previously wrapped model delegates to block writes,
but left client-level raw/escape hatches ($transaction, $executeRaw,
$executeRawUnsafe, $queryRawUnsafe, $runCommandRaw) intact. A read-tool
could smuggle DML via raw SQL, or open an interactive $transaction whose
tx-scoped client (unproxied by construction) accepts writes.
- read-only-prisma: block $transaction, $executeRaw, $executeRawUnsafe,
$queryRawUnsafe, $runCommandRaw at the client level. Template-tagged
$queryRaw stays allowed (read-only by API contract).
- assistant-tools: add create_estimate to MUTATION_TOOLS — it uses
$transaction internally and was previously bypassing the proxy only
because $transaction wasn't blocked.
- shared: document isReadOnly flag on ToolContext so any scoped tRPC
caller a tool spawns keeps the proxied client.
- helpers: note the runtime wrap at assistant-tools.ts:739 is
authoritative; forwarding ctx.db verbatim is correct.
- tests: cover model writes, raw escapes, and the allowed $queryRaw
path (7 cases, all pass).
- loosen one estimate-detail test that compared the exact db instance
(fails once that instance is a proxy; the assertion's intent is the
estimate id).
Covers EGAI 4.1.1.2 / IAAI 3.6.22.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Tests fell behind source changes: lastTotpAt replay-attack prevention,
activeSession invalidation on password reset, select clauses in
permission updates, UNAUTHORIZED (anti-enumeration) for disabled TOTP,
and password minimum raised from 8 to 12 characters.
Also fix root eslint.config.mjs to ignore packages/ (linted via turbo)
and add --no-warn-ignored to lint-staged to suppress warnings for
ignored files.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Moves read, assignment-procedures, assignment-mutations, and demand
procedures into allocation/ so the domain boundary is discoverable
without grep.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Allocation bars that have active optimistic overrides (post-drag,
awaiting server confirmation) now pulse subtly via animate-pulse.
The pending set is derived from the existing optimisticAllocations
map keys, requiring no additional state.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The shoring indicator logic was backwards. In the business context,
higher offshore = more cost-efficient = GOOD.
Inverted logic:
- Green: offshore >= threshold (target met, e.g. >= 55%)
- Yellow: offshore close to threshold (threshold-10 to threshold)
- Red: offshore below threshold (too little offshore, too expensive)
Updated:
- ShoringIndicator: getSeverity() inverted, badge text updated
- ProjectModal: "Max Offshore" renamed to "Min Offshore" with new tooltip
- AI Tool: status text reflects "target met" vs "below target"
- Tool description: "higher offshore is better, threshold is minimum"
Co-Authored-By: claude-flow <ruv@ruv.net>
When generating multiple cover images in one conversation, the
accumulated tool results (each ~400KB base64) caused the OpenAI
conversation payload to exceed JSON parsing limits in the browser.
Fix:
- Strip coverImageUrl from invalidate action results (not needed by AI)
- Cap invalidate results to 4KB
- Cap all tool results to 8KB
- Prevents "JSON.parse: unexpected character" errors during batch ops
Co-Authored-By: claude-flow <ruv@ruv.net>
The tool was hardcoded to only check isDalleConfigured(), ignoring
Gemini even when it was the configured image provider. This caused
"DALL-E is not configured" errors for all 13 projects.
Fix: reads imageProvider from SystemSettings and routes to Gemini
or DALL-E accordingly (same logic as the generateCover mutation).
Co-Authored-By: claude-flow <ruv@ruv.net>
Corrected model names (per Google AI docs):
- gemini-2.5-flash-image (was gemini-2.0-flash-preview-image-generation)
- gemini-3-pro-image-preview (Nano Banana Pro)
- gemini-3.1-flash-image-preview (Nano Banana 2)
UI: replaced text input with dropdown selector showing all 3 models
with human-readable descriptions.
Default changed to gemini-2.5-flash-image (fast, high-volume).
AI Assistant: generate_project_cover tool description updated to be
provider-agnostic (works with both DALL-E and Gemini).
Co-Authored-By: claude-flow <ruv@ruv.net>
Engine (packages/engine):
- New checkDuplicateAssignment() pure function: detects same resource
assigned to same project with overlapping dates
- 15 unit tests covering: overlap, no-overlap, cancelled, self-exclude,
string dates, PROPOSED status
Application layer (packages/application):
- createAssignment: throws CONFLICT before DB write if duplicate found
- fillDemandRequirement: same check before entering transaction
AI Assistant (packages/api/router/assistant-tools.ts):
- create_allocation: checks before creating, returns helpful error message
- fill_demand: same check using demand's projectId
UI (apps/web):
- AllocationModal: amber warning when resource already assigned to
selected project with overlapping dates (non-blocking)
Database cleanup:
- Found and merged 1 duplicate: Wong Wong on Porsche Taycan Sport Film
(2 overlapping PROPOSED assignments merged into 1)
Regression: 298 engine tests pass (283 + 15 new). TypeScript clean.
Co-Authored-By: claude-flow <ruv@ruv.net>
Hartmut