feat: ✨ Implement Claude memory tool #8269

chezsmithy · 2025-10-15T22:42:22Z

Description

In the latest release anthropic has enabled a new memory tool solution that allows local repository memory to be automatically updated based on internal model decisions.
This PR implements the memory tool for the Bedrock provider, and OIA Bedrock adapter, as a tool.
For the vscode and intelliJ extensions it's enabled as an experimental tool.
for the CLI it's enabled as a default tool.

The tool is only enabled for the following models:

"claude-sonnet-4-5-20250929",      // Claude Sonnet 4.5
"claude-sonnet-4-20250514",        // Claude Sonnet 4
"claude-haiku-4-5-20251001",       // Claude Haiku 4.5
"claude-opus-4-1-20250805",        // Claude Opus 4.1
"claude-opus-4-20250514",          // Claude Opus 4

AI Code Review

Team members only: AI review runs automatically when PR is opened or marked ready for review
Team members can also trigger a review by commenting @continue-general-review or @continue-detailed-review

Checklist

I've read the contributing guide
The relevant docs, if any, have been updated or created
The relevant tests, if any, have been updated or created

Screen recording or screenshot

[ When applicable, please include a short screen recording or screenshot - this makes it much easier for us as contributors to review and understand your changes. See this PR as a good example. ]

Tests

Added tests as needed.

cubic-dev-ai

4 issues found across 18 files

Prompt for AI agents (all 4 issues)


Understand the root cause of the following 4 issues and fix them.


<file name="extensions/cli/src/tools/memory.ts">

<violation number="1" location="extensions/cli/src/tools/memory.ts:130">
Creating a memory file in a new directory always fails: after creating the missing parent directory we throw `Path not found`, so fs.writeFile never runs. Drop the throw so the command can proceed.</violation>
</file>

<file name="core/tools/implementations/memory.ts">

<violation number="1" location="core/tools/implementations/memory.ts:121">
Core memory command handling logic duplicates extensions/cli/src/tools/memory.ts:executeMemoryCommand() and its handlers</violation>

<violation number="2" location="core/tools/implementations/memory.ts:131">
After creating a missing directory, this branch still throws `Path not found`, so the create command always fails for new paths.</violation>
</file>

<file name="extensions/cli/src/tools/index.tsx">

<violation number="1" location="extensions/cli/src/tools/index.tsx:141">
This enables the memory tool for any model containing “claude”, but the feature must be restricted to the specific allowlisted models from the PR description.</violation>
</file>

Since this is your first cubic review, here's how it works:

cubic automatically reviews your code and comments on bugs and improvements
Teach cubic by replying to its comments. cubic learns from your replies and gets better over time
Ask questions if you need clarification on any suggestion

_{React with 👍 or 👎 to teach cubic. Mention @cubic-dev-ai to give feedback, ask questions, or re-run the review.}

extensions/cli/src/tools/memory.ts

core/tools/implementations/memory.ts

extensions/cli/src/tools/index.tsx

core/tools/implementations/memory.ts

RomneyDa · 2025-11-18T23:53:15Z

@chezsmithy this is neat!

Fixed TypeScript compilation errors in memory.vitest.ts by adding non-null assertions to all parameter property accesses. The Tool interface defines parameters as optional, but memoryTool always has parameters defined, making the assertions safe. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <[email protected]>

The test was incorrectly expecting an error when creating a file in a non-existent directory. The actual implementation (handleCreate in memoryShared.ts) automatically creates parent directories recursively before writing the file, which is the correct behavior. Updated the test to expect success and verify that mkdir is called with the recursive option, matching the actual implementation. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <[email protected]>

Apply Prettier formatting to memory tool test files. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <[email protected]>

Add CI_GITHUB_TOKEN environment variable to setup and install steps in the CLI PR checks workflow. This prevents GitHub API rate limiting issues when npm downloads packages with binaries from GitHub releases. Without authentication, GitHub API allows only 60 requests/hour. With authentication, the limit increases to 5,000 requests/hour. This matches the pattern used in other component workflows (VSCode, GUI) and references the same issue: microsoft/vscode-ripgrep#9 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <[email protected]>

Add CI_GITHUB_TOKEN to the "Setup packages" step which was missing from the previous commit. This step runs build-packages.js which installs dependencies for 7 packages in parallel. Without the token, these parallel npm installs hit GitHub API rate limits (60 req/hour) when downloading binaries, causing processes to hang and eventually be killed with SIGTERM (exit code 143). The token increases the rate limit to 5,000 req/hour, preventing timeouts during the parallel package builds. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <[email protected]>

cubic-dev-ai

5 issues found across 18 files

Prompt for AI agents (all 5 issues)


Understand the root cause of the following 5 issues and fix them.


<file name="core/tools/index.test.ts">

<violation number="1" location="core/tools/index.test.ts:208">
This test claims it expects an empty array but only checks that the result is an array. Assert that no tools are returned to ensure the behavior is actually tested.</violation>
</file>

<file name="core/llm/llms/Bedrock.ts">

<violation number="1" location="core/llm/llms/Bedrock.ts:45">
Model memory tool support logic duplicates packages/openai-adapters/src/apis/Bedrock.ts:supportsMemoryTool() function. This duplication introduces a maintenance risk, as updates to the list of supported models might not be synchronized across both files, leading to inconsistent behavior.</violation>
</file>

<file name="packages/openai-adapters/src/apis/Bedrock.ts">

<violation number="1" location="packages/openai-adapters/src/apis/Bedrock.ts:608">
`tool_calls[].index` should be the zero-based ordinal of the tool call, but this change now copies Bedrock’s `contentBlockIndex`, which also counts text blocks. Any assistant text before a tool call will shift the index and cause downstream OpenAI consumers to mis-order or drop the tool call.</violation>
</file>

<file name="core/tools/implementations/memoryShared.ts">

<violation number="1" location="core/tools/implementations/memoryShared.ts:102">
Path validation allows escaping the memories root because it only checks `startsWith` on the absolute path prefix; include a path-separator-aware check (or compare `path.relative`) to keep operations inside the sandbox.</violation>

<violation number="2" location="core/tools/implementations/memoryShared.ts:286">
Root delete protection only matches the literal string `/memories`, so variants like `/memories/` delete the entire store; check the resolved path instead of the raw argument.</violation>
</file>

_{Reply to cubic to teach it or ask questions. Re-run a review with @cubic-dev-ai review this PR}

cubic-dev-ai · 2025-11-24T23:49:56Z

core/tools/index.test.ts

+      const tools = getConfigDependentToolDefinitions(
+        createParams({ modelName: "" }),
+      );
+      expect(Array.isArray(tools)).toBe(true);


This test claims it expects an empty array but only checks that the result is an array. Assert that no tools are returned to ensure the behavior is actually tested.

Prompt for AI agents

Address the following comment on core/tools/index.test.ts at line 208: <comment>This test claims it expects an empty array but only checks that the result is an array. Assert that no tools are returned to ensure the behavior is actually tested.</comment> <file context> @@ -0,0 +1,263 @@ + const tools = getConfigDependentToolDefinitions( + createParams({ modelName: "" }), + ); + expect(Array.isArray(tools)).toBe(true); + }); + </file context>

cubic-dev-ai · 2025-11-24T23:49:57Z

core/llm/llms/Bedrock.ts

+/**
+ * Checks if a model supports the memory tool (Claude 4+ models only)
+ */
+function supportsMemoryTool(modelName: string | undefined): boolean {


Model memory tool support logic duplicates packages/openai-adapters/src/apis/Bedrock.ts:supportsMemoryTool() function. This duplication introduces a maintenance risk, as updates to the list of supported models might not be synchronized across both files, leading to inconsistent behavior.

Prompt for AI agents

Address the following comment on core/llm/llms/Bedrock.ts at line 45: <comment>Model memory tool support logic duplicates packages/openai-adapters/src/apis/Bedrock.ts:supportsMemoryTool() function. This duplication introduces a maintenance risk, as updates to the list of supported models might not be synchronized across both files, leading to inconsistent behavior.</comment> <file context> @@ -39,6 +39,27 @@ interface PromptCachingMetrics { +/** + * Checks if a model supports the memory tool (Claude 4+ models only) + */ +function supportsMemoryTool(modelName: string | undefined): boolean { + if (!modelName) { + return false; </file context>

cubic-dev-ai · 2025-11-24T23:49:57Z

packages/openai-adapters/src/apis/Bedrock.ts

                tool_calls: [
                  {
-                    index: 0,
+                    index: blockIndex,


tool_calls[].index should be the zero-based ordinal of the tool call, but this change now copies Bedrock’s contentBlockIndex, which also counts text blocks. Any assistant text before a tool call will shift the index and cause downstream OpenAI consumers to mis-order or drop the tool call.

Prompt for AI agents

Address the following comment on packages/openai-adapters/src/apis/Bedrock.ts at line 608: <comment>`tool_calls[].index` should be the zero-based ordinal of the tool call, but this change now copies Bedrock’s `contentBlockIndex`, which also counts text blocks. Any assistant text before a tool call will shift the index and cause downstream OpenAI consumers to mis-order or drop the tool call.</comment> <file context> @@ -518,7 +605,7 @@ export class BedrockApi implements BaseLlmApi { tool_calls: [ { - index: 0, + index: blockIndex, id: delta.toolUse.toolUseId, type: "function", </file context>

cubic-dev-ai · 2025-11-24T23:49:57Z

core/tools/implementations/memoryShared.ts

+): Promise<MemoryCommandResult> {
+  const fullPath = validateMemoryPath(args.path, memoryRoot);
+
+  if (args.path === `/${MEMORIES_ROOT_NAME}`) {


Root delete protection only matches the literal string /memories, so variants like /memories/ delete the entire store; check the resolved path instead of the raw argument.

Prompt for AI agents

Address the following comment on core/tools/implementations/memoryShared.ts at line 286: <comment>Root delete protection only matches the literal string `/memories`, so variants like `/memories/` delete the entire store; check the resolved path instead of the raw argument.</comment> <file context> @@ -0,0 +1,364 @@ +): Promise<MemoryCommandResult> { + const fullPath = validateMemoryPath(args.path, memoryRoot); + + if (args.path === `/${MEMORIES_ROOT_NAME}`) { + throw new Error( + `Cannot delete the /${MEMORIES_ROOT_NAME} directory itself`, </file context>

Suggested change

if (args.path === `/${MEMORIES_ROOT_NAME}`) {

if (path.resolve(fullPath) === path.resolve(memoryRoot)) {

cubic-dev-ai · 2025-11-24T23:49:57Z

core/tools/implementations/memoryShared.ts

+  const resolved = path.resolve(memoryRoot, relative.length ? relative : ".");
+  const normalizedRoot = path.resolve(memoryRoot);
+
+  if (!resolved.startsWith(normalizedRoot)) {


Path validation allows escaping the memories root because it only checks startsWith on the absolute path prefix; include a path-separator-aware check (or compare path.relative) to keep operations inside the sandbox.

Prompt for AI agents

Address the following comment on core/tools/implementations/memoryShared.ts at line 102: <comment>Path validation allows escaping the memories root because it only checks `startsWith` on the absolute path prefix; include a path-separator-aware check (or compare `path.relative`) to keep operations inside the sandbox.</comment> <file context> @@ -0,0 +1,364 @@ + const resolved = path.resolve(memoryRoot, relative.length ? relative : "."); + const normalizedRoot = path.resolve(memoryRoot); + + if (!resolved.startsWith(normalizedRoot)) { + throw new Error( + `Path ${memoryPath} would escape /${MEMORIES_ROOT_NAME} directory`, </file context>

RomneyDa · 2025-11-26T22:58:38Z

@chezsmithy bumping that some of the cubic nitpicks seems valid!

chezsmithy added 2 commits October 12, 2025 18:36

Ensure memory directories are created

4e538dd

feat: 🐛 Update tool seeding for converse API

64b2b71

github-project-automation bot added this to Issues and PRs Oct 15, 2025

github-project-automation bot moved this to Todo in Issues and PRs Oct 15, 2025

fix: ✨ Additional changes and fixes

978a330

cubic-dev-ai bot reviewed Oct 21, 2025

View reviewed changes

extensions/cli/src/tools/memory.ts Outdated Show resolved Hide resolved

core/tools/implementations/memory.ts Outdated Show resolved Hide resolved

extensions/cli/src/tools/index.tsx Outdated Show resolved Hide resolved

core/tools/implementations/memory.ts Show resolved Hide resolved

chezsmithy added 3 commits November 12, 2025 21:50

fix: 🐛 fix cubic recommendations

f133d92

Merge branch 'main' into codex/implement-memory-tool-in-cli-and-core

4c8916a

fix: 🎨 Prettier fixes

f193bb3

chezsmithy and others added 5 commits November 24, 2025 09:44

fix: 🎨 Prettier fixes

76b08eb

Apply Prettier formatting to memory tool test files. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <[email protected]>

chezsmithy marked this pull request as ready for review November 24, 2025 23:34

chezsmithy requested a review from a team as a code owner November 24, 2025 23:34

chezsmithy requested review from tingwai and removed request for a team November 24, 2025 23:34

dosubot bot added the size:XXL This PR changes 1000+ lines, ignoring generated files. label Nov 24, 2025

cubic-dev-ai bot reviewed Nov 24, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat: ✨ Implement Claude memory tool #8269

feat: ✨ Implement Claude memory tool #8269

chezsmithy commented Oct 15, 2025 •

edited

Loading

Uh oh!

cubic-dev-ai bot left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

RomneyDa commented Nov 18, 2025

Uh oh!

cubic-dev-ai bot left a comment

Uh oh!

cubic-dev-ai bot Nov 24, 2025 •

edited

Loading

Uh oh!

cubic-dev-ai bot Nov 24, 2025 •

edited

Loading

Uh oh!

cubic-dev-ai bot Nov 24, 2025 •

edited

Loading

Uh oh!

cubic-dev-ai bot Nov 24, 2025 •

edited

Loading

Uh oh!

cubic-dev-ai bot Nov 24, 2025 •

edited

Loading

Uh oh!

RomneyDa commented Nov 26, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	if (args.path === `/${MEMORIES_ROOT_NAME}`) {
	if (path.resolve(fullPath) === path.resolve(memoryRoot)) {

feat: ✨ Implement Claude memory tool #8269

Are you sure you want to change the base?

feat: ✨ Implement Claude memory tool #8269

Conversation

chezsmithy commented Oct 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

AI Code Review

Checklist

Screen recording or screenshot

Tests

Uh oh!

cubic-dev-ai bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

RomneyDa commented Nov 18, 2025

Uh oh!

cubic-dev-ai bot left a comment

Choose a reason for hiding this comment

Uh oh!

cubic-dev-ai bot Nov 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cubic-dev-ai bot Nov 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cubic-dev-ai bot Nov 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cubic-dev-ai bot Nov 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cubic-dev-ai bot Nov 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

RomneyDa commented Nov 26, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

chezsmithy commented Oct 15, 2025 •

edited

Loading

cubic-dev-ai bot Nov 24, 2025 •

edited

Loading

cubic-dev-ai bot Nov 24, 2025 •

edited

Loading

cubic-dev-ai bot Nov 24, 2025 •

edited

Loading

cubic-dev-ai bot Nov 24, 2025 •

edited

Loading

cubic-dev-ai bot Nov 24, 2025 •

edited

Loading