telegram
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the Telegram API, creating a surface for indirect prompt injection. External messages are written to a shared inbox file and subsequently read by the agent.
- Ingestion points:
telegram_check.pyandtelegram_callback.pyread message text and callback data from~/.claudicle/daemon/inbox.jsonl. - Boundary markers: The skill documentation mentions sanitization but does not show explicit boundary delimiters in the script logic for inbox reading.
- Capability inventory: The skill can send messages via
telegram_send.pyand manage bot memory state throughtelegram_memory.py. - Sanitization:
SKILL.mdstates that user-provided text is sanitized for XML tags before being processed by the cognitive pipeline. - [DYNAMIC_EXECUTION]: The
telegram_memory.pyscript performs dynamic loading of modules by modifying the Python path at runtime. - Evidence: The script uses
sys.path.insertto include local directories such as~/.claudicle/adapters/shared,~/.claudicle/daemon, and~/.claude/skills/sharedto import internal framework components likeclaudicle_memoryandusermodel_resolver. - [COMMAND_EXECUTION]: The skill's architecture relies on the execution of multiple discrete Python scripts via subprocess calls to handle Telegram interactions.
- [EXTERNAL_DOWNLOADS]: The skill requires external Python libraries to function.
- Evidence: Documentation specifies prerequisites including
requestsfor the REST-based scripts andpython-telegram-botfor the optional listener component.
Audit Metadata