The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data, specifically cluster JSON and command text, which could potentially contain malicious instructions.
Ingestion points: The skill accepts <cluster-json-or-command-text> as input via the argument hint.
Boundary markers: The instructions lack specific prompt delimiters (e.g., XML tags or triple quotes) to isolate user-supplied data from the instructions, though they do provide logic for validating fields.
Capability inventory: The skill is limited to the Read tool and the specific syntax parser mcp__triage__parse_review_command.
Sanitization: The skill contains explicit instructions to refuse the display of unredacted sensitive text and to validate input records before rendering.

triage-display