The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied queries to perform vector searches and generate embeddings, creating a potential path for instructions embedded in data to influence the agent.
Ingestion points: The <query> argument defined in the argument-hint and processed by the embeddings_search and ruvllm_hnsw_route tools in SKILL.md.
Boundary markers: The skill does not define specific delimiters or instructions to the agent to treat the query content as data only.
Capability inventory: The skill has access to the Bash tool and multiple vector search/embedding management tools.
Sanitization: No explicit sanitization or input validation for the query string is documented.
[EXTERNAL_DOWNLOADS]: The skill instructions suggest using npx @claude-flow/cli@latest as an alternative interface.
Evidence: The "CLI alternative" section in SKILL.md contains commands like npx @claude-flow/cli@latest embeddings search.
This pattern downloads and executes code from an external package registry without pinning a specific version (@latest), which could lead to executing unexpected updates.
[COMMAND_EXECUTION]: The skill explicitly allows the use of the Bash tool and provides shell commands for the agent to execute.
Evidence: allowed-tools includes Bash and the CLI alternative section provides specific npx commands.

vector-search