The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions and metadata describe downloading and installing the airbyte-admin-pp-cli binary and MCP server from GitHub and NPM repositories associated with the mvanhorn organization.
[COMMAND_EXECUTION]: The skill is primarily used to execute the airbyte-admin-pp-cli command-line tool with various subcommands to interact with the Airbyte API.
[DATA_EXFILTRATION]: The skill documentation describes a --deliver webhook:<url> feature. This capability allows the output of any command execution to be sent to an arbitrary external URL via an HTTP POST request, which could be used to transmit sensitive information extracted from the Airbyte environment.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface. It ingests data from the Airbyte Public API (such as source descriptions or user names) which could contain malicious instructions. This data is processed in the agent's context and can be outputted to files or network destinations via the delivery system.
Ingestion points: Data returned from airbyte-admin-pp-cli public list-* commands in SKILL.md.
Boundary markers: The CLI wraps results in a JSON envelope (.results), but instructions do not explicitly warn the agent to ignore instructions embedded within these results.
Capability inventory: Subprocess execution (via allowed-tools: Bash), file writing (--deliver file:<path>), and network operations (--deliver webhook:<url>).
Sanitization: There is no evidence of sanitization or filtering of the content received from the Airbyte API before it is placed into the agent's context.

pp-airbyte-admin