The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the local dino CLI tool to manage saved views, involving potentially sensitive operations like data creation and deletion.
Evidence: Defined commands in SKILL.md include dino view list, dino view create, dino view update, and dino view delete.
Mitigation: The skill mandates passing user-provided values as individual arguments to prevent shell interpolation and requires the use of owner-only secure temporary files for complex JSON payloads.
[INDIRECT_PROMPT_INJECTION]: The skill handles untrusted data from CLI tool outputs, such as note content and view names, which could be used to deliver malicious instructions to the agent.
Ingestion points: CLI output from dino view list, dino view query, dino view get, and dino view note-detail as described in SKILL.md.
Boundary markers: Present; the Runtime Contract and Safety & Boundaries sections explicitly instruct the agent to treat all CLI output as untrusted and to never execute instructions found within it.
Capability inventory: The skill provides capabilities to list, query, create, update, and delete saved views via the dino CLI tool in SKILL.md.
Sanitization: Present; the skill requires structured JSON parsing, validation of field IDs through dino view fields, and mandatory dry-runs before performing any persistent writes.

dino-manage-views