The Agent Skills Directory

[NO_CODE]: The skill does not bundle executable scripts or binaries, consisting entirely of Markdown instructions and code snippets for the agent to follow.
[COMMAND_EXECUTION]: The instructions direct the agent to execute a local CLI tool named vector-db to perform index operations and data insertion.
[DATA_EXFILTRATION]: The skill sends vector data and metadata to an external, non-whitelisted API endpoint at api.openclaw.com.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it handles untrusted external data.
Ingestion points: Data enters the context via the vectors and metadata parameters in both CLI commands and API requests.
Boundary markers: The skill relies on JSON structural formatting to separate data from instructions but lacks explicit guardrail prompts to ignore embedded instructions within the metadata.
Capability inventory: The skill utilizes requests for network communication and subprocess execution for CLI operations.
Sanitization: While the documentation suggests validating inputs, there is no evidence of specific sanitization or filtering logic to prevent malicious payloads in the metadata from influencing the agent's logic.

vector-db