The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to download and install a CLI tool and an MCP server from a third-party vendor using npx and go install. While these resources belong to the vendor 'mvanhorn', they represent remote code execution at installation time.- [INDIRECT_PROMPT_INJECTION]: The 'Automatic learning' loop instructions create a surface for indirect injection. The agent is directed to run the recall command and then follow the resulting playbook steps or next_action candidates verbatim. Because these steps are derived from data returned by the CLI, it allows the tool's output to drive agent behavior without explicit sanitization or safety boundaries.- [DYNAMIC_EXECUTION]: The agent is instructed to execute trial commands and multi-step playbooks that are dynamically generated by the CLI at runtime. Specifically, the agent is told to follow the next_action field's commands verbatim, which constitutes dynamic generation and execution of shell commands from external data.- [DATA_EXFILTRATION]: The CLI tool supports a --deliver webhook:<url> flag, allowing the agent to route the results of its actions (which may include private messages or account details from the iNaturalist API) to arbitrary external HTTP endpoints.- [COMMAND_EXECUTION]: The skill relies extensively on the Read Bash capability to run the inaturalist-pp-cli binary and various subcommands to perform API operations, manage authentication, and handle the local learning database.

pp-inaturalist