The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes a webhook receiver implementation that ingests untrusted JSON payloads from external sources and persists them to a database, which represents an attack surface for indirect prompt injection.\n
Ingestion points: The TypeScript implementation in SKILL.md accepts arbitrary input via an Express.js POST /ingest endpoint.\n
Boundary markers: The instructions do not include boundary markers or delimiters to signal to the agent that the ingested content should be treated as untrusted data.\n
Capability inventory: The skill leverages the @clickhouse/client library to perform bulk database insertions.\n
Sanitization: The provided code persists data to ClickHouse using basic JSON stringification without sanitizing the content for potentially malicious prompt instructions.\n- [DATA_EXFILTRATION]: The skill includes instructions for network operations to transfer data between the local environment and remote sources.\n
Evidence: The SKILL.md and references/ingestion-methods.md files provide examples using curl and the url() table function to ingest data from remote endpoints such as data.example.com.\n- [EXTERNAL_DOWNLOADS]: The skill requires the use of external software packages to facilitate its ingestion logic.\n
Evidence: The instructions specify the installation of the express and @clickhouse/client Node.js packages.

clickhouse-webhooks-events