The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external Telegram webhooks and Kafka event streams, creating a potential attack surface for indirect prompt injection.
Ingestion points: scripts/webhook-receiver.ts (accepts HTTP POST updates from Telegram) and references/route.md (processes structured event streams).
Boundary markers: The instructions do not explicitly mandate unique delimiters for all untrusted inputs, though they emphasize sanitization.
Capability inventory: The agent has access to bash (via curl), write_file, and replace_file_content across multiple files.
Sanitization: The references/route.md and references/hook.md files include explicit instructions to scrub sensitive payload fields and avoid dumping raw sensitive data into messages.
[DYNAMIC_EXECUTION]: The skill involves the dynamic generation of executable bash scripts and project scaffolding.
Script generation: The references/hook.md documentation instructs the agent to "Write a bash hook script per event" to facilitate Claude Code notifications.
Runtime scaffolding: The scripts/webhook-receiver.ts file includes a scaffoldReceiver function that programmatically writes a server.ts file and configuration templates to a user-specified directory.
Mitigation: These behaviors are localized, template-based, and intended for integration setup, representing a standard and low-risk implementation of the skill's primary purpose.

telegram