The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external task descriptions and CLI search results, which represents a potential surface for indirect prompt injection.
Ingestion points: Data enters the agent's context through the output of dino todo search and user-provided task text (File: SKILL.md).
Boundary markers: The skill contains explicit instructions to treat all CLI output and task text as untrusted data and strictly forbids executing any instructions found within the data (File: SKILL.md, Runtime Contract).
Capability inventory: The skill possesses the capability to execute dino CLI commands to create, search, and update tasks, which involves reading and writing data (Files: references/todo-mutations.md, references/todo-search.md).
Sanitization: To mitigate risks, the instructions mandate using host-native secure temporary-file APIs or stdin for free-form content and explicitly warn against string interpolation in shell commands (File: SKILL.md).
[COMMAND_EXECUTION]: The skill relies on executing the dino CLI binary to perform its tasks. The instructions strictly limit execution to this specific tool for todo-related operations and emphasize maintaining user-provided values as separate arguments to prevent command injection.

dino-manage-todo