The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill clones an external, unverified repository from GitHub (https://github.com/abhigyanpatwari/GitNexus.git) and proceeds to run npm install and npm run build on the downloaded content. This executes third-party code that is not from a trusted organization or well-known service.
[DATA_EXFILTRATION]: The skill uses cloudflared to create a public tunnel for the local GitNexus service (cloudflared tunnel --url http://localhost:8888). This exposes the indexed codebase data, including symbol call chains and architecture, to the public internet via a trycloudflare.com URL without any authentication or access control.
[COMMAND_EXECUTION]: The skill instructs the agent to perform several shell-level operations, including installing dependencies, building production assets, and managing background processes using pkill.
[INDIRECT_PROMPT_INJECTION]: The skill ingests local codebase data to generate a knowledge graph and provides an AI chat panel feature. This creates a surface where malicious instructions embedded in a processed codebase could potentially influence the agent's behavior during exploration or semantic search.
Ingestion points: Reads the target repository using npx gitnexus analyze (SKILL.md).
Boundary markers: None implemented to distinguish codebase content from instructions.
Capability inventory: Execution of shell commands, network tunneling, and local file reads (SKILL.md, scripts/proxy.mjs).
Sanitization: No evidence of sanitization for the data processed for the AI chat or embeddings features.
[EXTERNAL_DOWNLOADS]: Fetches the cloudflared binary directly from Cloudflare's official GitHub releases. While the source is a well-known service, the binary is used to facilitate the public exposure of local data described in the exfiltration finding.

gitnexus-explorer