The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the @zvec/zvec-grep package via the Node Package Manager (npm). The instructions emphasize pinning the installation to version 0.2.2 from the public registry.
[COMMAND_EXECUTION]: The skill utilizes several CLI commands through the zg binary and the provided zvec-grep.sh helper script to perform workspace indexing, hybrid searches, and status checks. It also uses npm for installation tasks and node for runtime verification.
[PERSISTENCE]: The zg install command is designed to modify local agent configuration files (such as ~/.claude.json, ~/.cursor/mcp.json, and others) to register the MCP server. This is the primary mechanism for maintaining the tool's integration across agent sessions. The skill specifies the use of managed blocks to preserve unrelated user settings during these modifications.
[INDIRECT_PROMPT_INJECTION]: The skill provides retrieval capabilities over untrusted local workspace data, which creates a potential surface for indirect prompt injection if malicious instructions are embedded in indexed files.
Ingestion points: The zg query command reads and processes content from the local workspace, including code files, Markdown, text, and structured data (JSON, YAML, CSV).
Boundary markers: The instructions recommend that the agent use result limits (--limit) for semantic probes and verify findings with exact matches using rg or direct file reads.
Capability inventory: The skill environment allows for the use of Bash, Read, Write, Edit, Glob, Grep, and WebFetch tools.
Sanitization: The skill instructions do not implement specific data sanitization logic; it relies on the agent's internal processing of retrieved search results.

zvec-grep