The Agent Skills Directory

[PROMPT_INJECTION]: The skill incorporates a proactive defense against indirect prompt injection from retrieved data.
Ingestion points: Visitor data retrieved from analytics endpoints defined in references/api.md.
Boundary markers: Explicit 'Do NOT' constraint in SKILL.md instructing the agent to ignore directives within strings.
Capability inventory: SQL execution and network mutation calls (POST/PATCH/PUT/DELETE) defined in references/api.md.
Sanitization: Instructions in SKILL.md ensure external data is treated as literal content rather than executable directives.
[DATA_EXFILTRATION]: The skill interacts with the vendor's official domain (https://dash.tinyanalytics.io).
In SKILL.md, it identifies that project identifiers (siteId) may be located in the local .env file, which is a standard configuration practice. All network operations are scoped to the vendor's platform.
[COMMAND_EXECUTION]: The skill facilitates database query execution via a managed API.
In SKILL.md and references/api.md, the agent is directed to use the /dashboards/run-card endpoint for dry-running ClickHouse SELECT queries to validate them before persistence.

tinyanalytics-dashboards