The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources that could influence the agent's behavior during summarization tasks.
Ingestion points: User-submitted text messages and photo captions received via the Telegram bot implementation in SKILL.md are stored and subsequently retrieved for processing.
Boundary markers: The generate_daily_report function in SKILL.md interpolates raw JSON report data directly into an LLM prompt template without using delimiters (e.g., XML tags or triple quotes) or instructions to ignore embedded commands.
Capability inventory: The skill documentation includes patterns for filesystem writes (file.download_to_drive) and network operations (requests.post, Telegram API calls).
Sanitization: There is no evidence of sanitization, filtering, or validation of the message content before it is passed to the llm.summarize method.
[COMMAND_EXECUTION]: The skill provides functional Python code snippets that perform system-level operations.
The implementation in SKILL.md uses the telegram.ext library to handle commands and messages, involving local file storage and network communication.
The agent is instructed in instructions.md to follow these implementation patterns, which involve managing project data and files.
[EXTERNAL_DOWNLOADS]: The bot implementation facilitates the download of external media.
The handle_photo function in SKILL.md automatically downloads photos from Telegram's servers to the local filesystem using file.download_to_drive(photo_path).

telegram-field-bot