The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes complex multi-step bash scripts to perform environment preflight checks, verify binaries, and manage global package installations.
[COMMAND_EXECUTION]: Recommends the use of sudo for global npm package installation to bypass permission restrictions, which involves privilege escalation.
[COMMAND_EXECUTION]: Generates and executes a dynamic Python one-liner script to programmatically parse and modify JSON configuration files for Cursor and OpenCode editors in the user's home directory.
[REMOTE_CODE_EXECUTION]: Downloads and installs the gitnexus package from the npm registry and executes its setup command, which installs additional unverified scripts and skills into editor directories.
[DATA_EXFILTRATION]: Accesses, reads, and modifies sensitive tool configuration files located in ~/.cursor/mcp.json, ~/.codex/config.toml, and ~/.config/opencode/config.json.
[DATA_EXFILTRATION]: Implements a repository-wide indexing process (gitnexus analyze) that reads all code and metadata to build a knowledge graph, creating a significant surface for indirect prompt injection.
Ingestion points: gitnexus analyze reads all files within the current repository path.
Boundary markers: Absent for the auto-generated skill files created during indexing.
Capability inventory: The skill possesses bash execution and file-write capabilities across multiple sensitive directories.
Sanitization: No sanitization or validation of the ingested repository content is performed before processing it into agent-ready skills.
[PROMPT_INJECTION]: Employs authoritative instructions such as "No asking the user what to do" and "Hard Rules" to enforce autonomous execution and bypass standard user-confirmation patterns.
[EXTERNAL_DOWNLOADS]: Fetches external code from the npm registry via npm install and npx during the setup and configuration phases.

gitnexus