The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external sources to generate a structured brief for a downstream agent.
Ingestion points: Processes user-provided PR links, issue links, and session context (Step 1), as well as outputs from research tools like /how and /why (Step 3).
Boundary markers: The skill creates a structured Markdown brief to hand over to the /to-artifact tool, but does not include explicit instructions to the downstream agent to disregard potentially malicious instructions embedded within the gathered descriptions or research data.
Capability inventory: The skill has the ability to execute code via /run, write capture files to the disk, and dispatch new agents with specific contexts.
Sanitization: There is no evidence of sanitization or escaping of the content gathered from research tools or external links before it is interpolated into the final brief.
[COMMAND_EXECUTION]: The skill uses platform tools to interact with and execute project-related code.
Evidence: Step 3 instructions require the agent to 'Invoke /run to get the app serving, then drive the flow from that URL' to generate visual captures. This utilizes the platform's capability to execute the local application environment.

showcase