codebase-memory-mcp
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill documents that the upstream project's Go wrapper (in
pkg/go/cmd/codebase-memory-mcp/main.go) downloads a prebuilt platform binary from GitHub releases during its first execution. This behavior is explicitly noted in the documentation to ensure the agent requests user approval before the download occurs. - [INDIRECT_PROMPT_INJECTION]: The skill processes local codebase files, which creates a potential surface for malicious instructions embedded in source code to influence agent behavior.
- Ingestion points: Reads files and directory structures from a local repository checkout using
Bash,Read,Grep, andGlobtools. - Boundary markers: The skill uses a manifest file (
references/source.json) and a dedicated verification script (scripts/preflight.py) to confirm the integrity of specific files via SHA-256 hashes before indexing or operating on them. - Capability inventory: The skill uses
Bashfor command execution and filesystem operations, and is capable of writing logs, indexes, and architecture decisions. - Sanitization: The
scripts/preflight.pyscript performs significant sanitization, including validating that paths are absolute and non-symbolic to prevent traversal attacks, and explicitly blocking access to sensitive directories and files such as.ssh,.env,.aws, and credential-related stems.
Audit Metadata