The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill acts as a scout for Core Web Vitals, which involves processing telemetry data (hostnames, paths, and URLs) that is identified as attacker-controllable and a potential vector for malicious instructions.
Ingestion points: Untrusted telemetry data enters the agent context through event properties including properties.$host, properties.$pathname, properties.$current_url, and $web_vitals_INP_event.attribution retrieved via the execute-sql and read-data-schema tools in SKILL.md and references/onset-correlation.md.
Boundary markers: The skill contains explicit instructions in SKILL.md to treat these telemetry values as "opaque data" and "never instructions," specifically warning the agent to ignore any directives embedded within these strings.
Capability inventory: The skill possesses read and write capabilities, including execute-sql for querying events, scout-emit-report and scout-edit-report for writing to a report channel, and scout-scratchpad-remember for persisting data in internal memory.
Sanitization: Robust sanitization is implemented at the query layer. The instructions in SKILL.md and references/onset-correlation.md provide and require SQL regex patterns (replaceRegexpAll) to strip non-safe characters and length capping (substring) to neutralize malicious payloads before the data is processed by the agent.

signals-scout-web-vitals