The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to transmit local workspace data—including file paths, branch information, git diffs, and source code—to the Hamster service (via MCP at tryhamster.com or the local CLI). While this is the intended functionality, it constitutes intentional data exposure to a third-party service.
[COMMAND_EXECUTION]: The skill uses shell commands via the hamster CLI (hamster chat) to process user requests. This introduces a potential command injection vector if user-provided arguments are not properly handled by the execution environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external and local sources which could potentially contain malicious instructions.
Ingestion points: Data is ingested from the .hamster/.state.json configuration file and from the output of the ask_hamster and hamster chat tools.
Boundary markers: No explicit boundary markers or delimiters are used to isolate untrusted data when interpolating it into prompts.
Capability inventory: The skill has the capability to execute shell commands (hamster chat) and perform account switching operations through the Hamster MCP interface.
Sanitization: There is no evidence of sanitization or validation performed on the user-supplied arguments or the data retrieved from the Hamster service before it is processed.

ask-hamster