The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from the Dinox CLI and external sources like note contents, which represents a potential surface for indirect prompt injection where malicious instructions could be embedded in data.
Ingestion points: The skill identifies note content, task text, tags, filenames, and all CLI output as untrusted data ingestion points in SKILL.md.
Boundary markers: The instructions explicitly mandate treating these sources as untrusted and provide a 'Runtime Contract' that forbids executing instructions found within Dinox data.
Capability inventory: The skill uses shell execution for dino CLI commands including status, login, and logout.
Sanitization: The skill requires using structured JSON output (--format json) and strictly forbids the interpolation of untrusted text into shell commands.
[COMMAND_EXECUTION]: The skill facilitates the execution of local shell commands via the dino CLI tool to manage authentication states.
Evidence includes instructions for dino auth status, dino auth login, and dino auth logout.
Risk is mitigated by guidelines requiring user-provided values to be passed as individual arguments and advocating for the use of stdin for sensitive token handling.

dino-auth