The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill clones the GitNexus repository from an untrusted GitHub user account (abhigyanpatwari) and executes build commands. This introduces unverified external code into the local environment. It also fetches the cloudflared binary from Cloudflare's official GitHub releases repository.
[REMOTE_CODE_EXECUTION]: The skill uses npx to execute the gitnexus package, which involves running code from a public registry that is not associated with a trusted organization.
[DATA_EXFILTRATION]: The skill initiates a public Cloudflare tunnel to expose the local knowledge graph service to the internet. This generates a public URL that grants remote access to indexed codebase data, potentially leading to unauthorized data exposure.
[DYNAMIC_EXECUTION]: The skill generates and executes a custom Node.js proxy script (proxy.mjs) that lacks proper path sanitization. This vulnerability could allow a remote attacker to perform a directory traversal attack by requesting paths like '../../', potentially gaining access to sensitive files outside the intended directory.
[INDIRECT_PROMPT_INJECTION]: The skill indices and processes codebase content, which represents a surface for indirect prompt injection attacks.
Ingestion points: Source code files in the target repository are read and analyzed by the gitnexus tool during the indexing process.
Boundary markers: The instructions do not define or implement boundary markers to isolate processed codebase content from the agent's control logic.
Capability inventory: The skill possesses capabilities for shell command execution (git, npm, npx, cloudflared), file system writes, and establishing network tunnels.
Sanitization: There is no evidence of content sanitization or validation of the indexed data before it is rendered in the web UI or processed by the agent.

gitnexus-explorer