The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon data from external sources, specifically Taskboard issue descriptions and comments. It explicitly instructs the agent to treat these comments as current requirements, creating a vulnerability where an attacker could embed malicious instructions in a task to manipulate the agent's behavior.
Ingestion points: SKILL.md (Core workflow step 1) utilizes taskctl issue get and taskctl comment list to retrieve data.
Boundary markers: Absent. There are no instructions to delimit or ignore instructions within the retrieved data.
Capability inventory: The skill has the ability to read and write files via taskctl attachment upload/download and perform network operations through the Taskboard API.
Sanitization: Absent. The skill does not describe any sanitization or validation of the external content before processing.
[DATA_EXFILTRATION]: The skill provides the ability to upload local files to a remote server using the taskctl attachment upload --file PATH command. While intended for task management, this capability could be abused if the agent is misled by an indirect prompt injection to upload sensitive local files (such as .env or credentials) to the Taskboard.
[COMMAND_EXECUTION]: The skill relies on the execution of the taskctl CLI binary. It provides specific system paths for this tool on macOS and Linux, which is a powerful capability that allows the agent to interact with the underlying system.
[CREDENTIALS_UNSAFE]: The documentation in references/cli.md describes a cloud login workflow that handles authentication secrets, including a "Shared key" prompt and the use of Basic Authentication via a local proxy service.

manage-taskboard