The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill explicitly handles untrusted inputs from web content and external data sources, creating a potential surface for indirect prompt injection.
Ingestion points: The skill instructions in SKILL.md identify 'Web content and external data' ingested via tools like 'BrowserOS' as ingestion points.
Boundary markers: No explicit delimiters are defined in the provided markdown, though the skill mandates redaction of secrets.
Capability inventory: The skill possesses the capability to execute terminal commands (via the ExecutionProxy), write code diffs, and interact with persistent memory stores.
Sanitization: The instructions include a mandatory protocol to redact secrets and PII (Personally Identifiable Information) before sharing context with subagents.
[COMMAND_EXECUTION]: The skill is designed to execute terminal actions such as tests and build commands through an 'ExecutionProxy Interface' (using the rtk prefix), which allows the agent to run arbitrary shell commands to fulfill its debugging purpose.

build-fix