@dizzlkheinz/ynab-mcpb 0.13.1 → 0.15.1

package/.code/agents/95d9fbab-19a2-48af-83f9-c792566a347f/error.txt

@@ -0,0 +1 @@
+Failed to execute code-gpt-5.1-codex-mini: program not found

package/.code/agents/b0098cb8-cb32-4ada-9bc4-37c587518896/result.txt

@@ -0,0 +1,170 @@
+I'm using the writing-plans skill to create the implementation plan.
+
+# MCP Resilience Implementation Plan
+
+> **For Claude:** REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task.
+
+**Goal:** Harden the MCP server by validating every tool’s metadata schema and making `clear_cache` fully reset delta/caching state so knowledge doesn’t drift while keeping diagnostics accurate.
+
+**Architecture:** Validate metadata against generated `zod` schemas inside `ToolRegistry`, expose mismatches immediately, and delegate cache clearing to `DeltaCache.forceFullRefresh()` so both the LRU cache and `ServerKnowledgeStore` reset before diagnostics observe the new state.
+
+**Tech Stack:** TypeScript 5/Node 18+, `zod`/`zod-validation-error`, MCP SDK server wiring in `src/server`, Vitest for fast unit/integration suites, and Markdown docs in `docs/reference/TOOLS.md`.
+
+### Task 1: Tighten tool metadata schema validation
+
+**Files:**
+- Modify: `src/server/toolRegistry.ts`
+- Modify: `src/server/__tests__/toolRegistry.test.ts`
+
+**Step 1: Write the failing test**
+
+```ts
+it('rejects registration when metadata inputJsonSchema diverges from the handler schema', () => {
+  const registry = new ToolRegistry({
+    withSecurityWrapper: vi.fn().mockReturnValue(() => async (handler) => handler({})),
+    errorHandler: mockedErrorHandler,
+    responseFormatter: responseFormatter,
+  });
+  const mismatchedMetadata = { type: 'object', properties: { bogus: { type: 'string' } } };
+  expect(() =>
+    registry.register({
+      name: 'budget-sync',
+      description: 'Sync budget data',
+      inputSchema: z.object({ budget_id: z.string().min(1) }).strict(),
+      handler: async () => ({ content: [{ type: 'text', text: responseFormatter.format({}) }] }),
+      metadata: { inputJsonSchema: mismatchedMetadata },
+    }),
+  ).toThrow('metadata input schema must match generated schema for tool "budget-sync"');
+});
+```
+
+**Step 2: Run the test to verify it fails**
+
+Run: `npm run test:unit -- --runInBand src/server/__tests__/toolRegistry.test.ts --testNamePattern "metadata input schema"`
+Expected: FAIL because the registry currently accepts mismatched metadata.
+
+**Step 3: Implement the minimal enforcement**
+
+```ts
+private ensureMetadataSchemaMatches(definition: ToolDefinition): void {
+  const generatedSchema = this.generateJsonSchema(definition.inputSchema, 'input');
+  if (definition.metadata?.inputJsonSchema) {
+    if (JSON.stringify(generatedSchema) !== JSON.stringify(definition.metadata.inputJsonSchema)) {
+      throw new Error(
+        `metadata input schema must match generated schema for tool "${definition.name}"`,
+      );
+    }
+  } else {
+    definition.metadata = {
+      ...definition.metadata,
+      inputJsonSchema: generatedSchema,
+    };
+  }
+}
+```
+
+Call `this.ensureMetadataSchemaMatches(definition);` near the top of `register` before storing the tool so `metadata.inputJsonSchema` is always populated (and validated) for `listTools()`. This also means `ToolRegistry.listTools()` no longer needs to fall back to `generateJsonSchema` when metadata is present.
+
+**Step 4: Run the test to confirm it now passes**
+
+Run: `npm run test:unit -- --runInBand src/server/__tests__/toolRegistry.test.ts --testNamePattern "metadata input schema"`
+Expected: PASS.
+
+**Step 5: Commit**
+
+```bash
+git add src/server/toolRegistry.ts src/server/__tests__/toolRegistry.test.ts
+git commit -m "fix: validate tool metadata schemas at registration"
+```
+
+### Task 2: Reset delta knowledge when clearing the cache
+
+**Files:**
+- Modify: `src/server/YNABMCPServer.ts`
+- Modify: `src/server/__tests__/YNABMCPServer.integration.test.ts`
+- Modify: `docs/reference/TOOLS.md`
+
+**Step 1: Write the failing integration test**
+
+```ts
+it(
+  'resets delta knowledge entries when clear_cache runs',
+  { meta: { tier: 'domain', domain: 'server' } },
+  async (ctx) => {
+    await skipOnRateLimit(async () => {
+      const token = accessToken();
+      await registry.executeTool({
+        name: 'list_accounts',
+        accessToken: token,
+        arguments: { budget_id: budgetsResult?.budgets?.[0]?.id },
+      });
+
+      const before = await registry.executeTool({
+        name: 'diagnostic_info',
+        accessToken: token,
+        arguments: { include_delta: true },
+      });
+      const beforeData = JSON.parse(before.content?.[0]?.text ?? '{}');
+      expect(beforeData.delta?.knowledge_entries).toBeGreaterThan(0);
+
+      await registry.executeTool({ name: 'clear_cache', accessToken: token, arguments: {} });
+
+      const after = await registry.executeTool({
+        name: 'diagnostic_info',
+        accessToken: token,
+        arguments: { include_delta: true },
+      });
+      const afterData = JSON.parse(after.content?.[0]?.text ?? '{}');
+      expect(afterData.delta?.knowledge_entries).toBe(0);
+    }, ctx);
+  },
+);
+```
+
+**Step 2: Run the integration test to verify it fails**
+
+Run: `npm run test:integration:domain -- --testNamePattern "clear cache runs"`
+Expected: FAIL because `clear_cache` currently only clears `cacheManager`, leaving `ServerKnowledgeStore` populated.
+
+**Step 3: Implement the minimal fix**
+
+- Update the `clear_cache` handler in `setupToolRegistry` to call `this.deltaCache.forceFullRefresh()` instead of directly `cacheManager.clear()`, ensuring both the cache and knowledge store reset:
+
+```ts
+handler: async () => {
+  this.deltaCache.forceFullRefresh();
+  return {
+    content: [{ type: 'text', text: responseFormatter.format({ success: true }) }],
+  };
+},
+```
+
+- Update `docs/reference/TOOLS.md` entry to explain the stronger guarantee:
+
+```md
+| `clear_cache` | Clear the cache and reset delta knowledge so clients can start fresh | None |
+```
+
+**Step 4: Run the integration test to confirm it passes**
+
+Run: `npm run test:integration:domain -- --testNamePattern "clear cache runs"`
+Expected: PASS.
+
+**Step 5: Commit**
+
+```bash
+git add src/server/YNABMCPServer.ts src/server/__tests__/YNABMCPServer.integration.test.ts docs/reference/TOOLS.md
+git commit -m "feat: reset delta knowledge when cache is cleared"
+```
+
+---
+
+Because the sandbox is read-only, I couldn’t save this plan to `docs/plans/2025-11-21-mcp-resilience.md`; please copy the above Markdown there in a workspace-write session before executing the tasks.
+
+- **Plan scope:** Adds schema validation in `src/server/toolRegistry.ts` plus a new regression test in `src/server/__tests__/toolRegistry.test.ts`, and modifies `clear_cache` in `src/server/YNABMCPServer.ts` with an integration test addition and doc update in `docs/reference/TOOLS.md`.
+- **Testing:** Run the targeted unit test via `npm run test:unit -- --runInBand src/server/__tests__/toolRegistry.test.ts --testNamePattern "metadata input schema"` and the integration slice via `npm run test:integration:domain -- --testNamePattern "clear cache runs"` after making the code changes.
+
+Next steps:
+1. Execute the described plan in a writable worktree (create `docs/plans/2025-11-21-mcp-resilience.md` with the above content first) and implement Task 1’s schema validation.
+2. Run the targeted unit test from Task 1 to confirm failure, then fix the code, rerun to confirm success.
+3. Implement Task 2, add the integration test, update `clear_cache`, refresh `docs/reference/TOOLS.md`, and run `npm run test:integration:domain -- --testNamePattern "clear cache runs"`.

package/.code/agents/b4fe59a4-81df-42e2-a112-0153e504faca/error.txt

@@ -0,0 +1 @@
+Required agent 'gemini' is not installed or not in PATH

package/.code/agents/bf4ce152-f623-49d7-aa52-c18631625c3c/error.txt

@@ -0,0 +1,3 @@
+Command failed: Error loading configuration: config profile `Smoke test: from the current repo, run `pwd` and `git status -sb`, then list top-level files/directories and print a short summary. No file writes or edits.
+
+[Running in read-only mode - no modifications allowed]` not found

package/.code/agents/d7d1db75-d7eb-468e-adea-4ef4d916d187/status.txt

@@ -0,0 +1 @@
+Agent cancelled

package/.code/agents/e2baa9c8-bac3-49e3-a39d-024333e6a990/status.txt

@@ -0,0 +1 @@
+Agent cancelled

package/.code/agents/e350b8c3-8483-408c-b2bb-94515f492a11/error.txt

@@ -0,0 +1,3 @@
+Command failed: Error loading configuration: config profile `You are helping craft an implementation plan for the YNAB MCP Server repo (Model Context Protocol server for YNAB). Repo root at C:/Users/ksutk/projects/ynab-mcpb. Key info: TypeScript/Node 18+, ts source in src/, server wiring in src/server, tools in src/tools, utils in src/utils, entry at src/index.ts. Build: npm run build (lint+tsc+bundle) outputs dist/. Tests use Vitest with projects (unit, integration, e2e). package.json scripts include build:prod, lint:fix, test:unit, test:integration, test:e2e, package:mcpb. Docs in docs/. Current version 0.13.4. Goal: propose a concrete step-by-step implementation plan for the next iteration to improve the project. Include where to change code (paths), suggested tests to add/run, and rationale. Assume current tests passing (test-results show success). Focus on actionable tasks (2-5 min steps) using TDD mindset. Deliver plan structure we can hand to engineer. Do not assume write access.
+
+[Running in read-only mode - no modifications allowed]` not found

package/.code/agents/e63f9919-719f-4ad0-bccf-01b1a596e1e9/status.txt

@@ -0,0 +1 @@
+Agent cancelled

package/.code/agents/e71695a8-3044-478d-8f12-ed13d02884c7/status.txt

@@ -0,0 +1 @@
+Agent cancelled

package/.code/agents/f95b7464-3e25-4897-b153-c8dfd63fd605/error.txt

@@ -0,0 +1,5 @@
+Command failed: Error loading configuration: config profile `Context: cwd: C:\Users\ksutk\projects\ynab-mcpb. This is only to verify whether a gemini-3-pro backed agent can successfully start and complete.
+
+Agent: Minimal Gemini smoke test: from the ynab-mcpb repo, run `git status -sb` and list top-level directories, then print a short summary and exit. No file writes required.
+
+[Running in read-only mode - no modifications allowed]` not found

package/.code/agents/fa3c5ddf-cdf7-47a2-930a-b806c6363689/status.txt

@@ -0,0 +1 @@
+Agent cancelled

package/.github/workflows/publish.yml

@@ -10,6 +10,8 @@ jobs:
     permissions:
       contents: read
       id-token: write
+    env:
+      YNAB_ACCESS_TOKEN: ${{ secrets.YNAB_ACCESS_TOKEN }}
     steps:
       - uses: actions/checkout@v4
 
@@ -25,9 +27,7 @@ jobs:
         run: npm run build
 
       - name: Publish to npm
-        env:
-          NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
         run: |
-          echo "CI detected, using unit tests only"
+          echo "Using npm Trusted Publishing (OIDC)"
           npm run test:unit
           npm publish --provenance --access public

package/.github/workflows/release.yml

@@ -87,3 +87,7 @@ jobs:
           files: ${{ steps.meta.outputs.file }}
           generate_release_notes: true
           fail_on_unmatched_files: true
+          # Use WORKFLOW_PAT to allow triggering the publish workflow
+          # PAT needs: repo, workflow permissions
+          # Falls back to GITHUB_TOKEN if PAT not configured
+          token: ${{ secrets.WORKFLOW_PAT || github.token }}

package/CHANGELOG.md

@@ -7,6 +7,81 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.15.0] - 2025-11-30
+
+### Changed
+
+- **Version Bump** - Minor version increment
+
+## [0.14.0] - 2025-11-26
+
+### Added
+
+- **Reconciliation V2 Architecture** - Complete redesign of reconciliation system
+  - Canonical transaction types with milliunits-based amounts (eliminates float precision issues)
+  - New CSV parser using PapaParse with Canadian bank presets (TD, RBC, Scotiabank, Wealthsimple, Tangerine)
+  - Advanced fuzzy matching engine using fuzzball for merchant name matching
+  - Configurable scoring system with amount/date/payee weights and bonuses
+  - Enhanced date parsing with chrono-node supporting multiple formats
+  - Auto-detection of CSV formats with comprehensive error reporting
+  - Support for debit/credit column formats and European number formats
+  - New dependencies: `chrono-node`, `dayjs`, `fuzzball`, `papaparse`
+
+### Changed
+
+- **Matching Algorithm** - Improved accuracy and configurability
+  - Default amount tolerance: 1 cent (10 milliunits, down from 50)
+  - Default date tolerance: 7 days (up from 2 days for bank posting delays)
+  - Rebalanced weights: amount 50%, payee 35%, date 15%
+  - Auto-match threshold: 85% (down from 90% for better match rates)
+  - Token-set-ratio matching for payee names handles bank merchant variations
+
+- **Code Quality** - Eliminated duplication and improved maintainability
+  - Exported `normalizeConfig` function from matcher for reuse
+  - Removed 23 lines of duplicated config construction in analyzer
+  - Updated all documentation to reflect V2 implementation
+
+### Fixed
+
+- **Documentation** - Corrected outdated references and mismatches
+  - Fixed file path references (removed non-existent `matcher.v2.ts`)
+  - Corrected config value documentation (amountToleranceMilliunits)
+  - Fixed Markdown table formatting in reconciliation plan
+
+## [0.13.4] - 2025-11-21
+
+### Changed
+
+- **npm Trusted Publishing** - Switched to OIDC-based authentication
+  - More secure than token-based authentication
+  - No secrets to manage or rotate
+  - Automatic authentication via GitHub Actions OIDC
+
+## [0.13.3] - 2025-11-21
+
+### Fixed
+
+- **Publish Workflow** - Added YNAB_ACCESS_TOKEN to npm publish workflow
+  - Performance tests now have required token during publish
+  - Ensures all unit tests pass before npm publish
+
+## [0.13.2] - 2025-11-21
+
+### Changed
+
+- **Improved CI/CD Workflows** - Enhanced reliability and automation
+  - Integration tests now optional with `continue-on-error` (won't block merges)
+  - Increased integration test timeout to 90 minutes for rate-limited YNAB API calls
+  - Added WORKFLOW_PAT support for automatic npm publish on release
+  - Release workflow can now trigger publish workflow automatically
+
+### Fixed
+
+- **CI Timeout Issues** - Resolved integration test timeouts in GitHub Actions
+  - Tests hitting YNAB API rate limits (200/hour) no longer cause failures
+  - 60-minute rate limit reset waits are now accommodated
+  - CI provides visibility without blocking development
+
 ## [0.13.1] - 2025-11-21
 
 ### Fixed

package/NUL

@@ -0,0 +1 @@
+dir: cannot access '/B': No such file or directory