The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface because the sub-agent processes data from external trade logs and memory files.
Ingestion points: Reads from daily memory logs, dsl-state JSON files, and trade history via mcporter.
Boundary markers: The analysis prompt in references/analysis-prompt.md lacks delimiters or specific instructions to ignore embedded commands in the processed data.
Capability inventory: The sub-agent can write to the filesystem, query financial APIs, and send messages via the Telegram message tool.
Sanitization: There is no evidence of data sanitization or validation before interpolation into the sub-agent context.
[COMMAND_EXECUTION]: The scripts/howl-setup.py script facilitates the creation of a persistent cron job that executes agent turns on a daily schedule.
[DATA_EXFILTRATION]: The skill collects and transmits sensitive account data, including wallet addresses and trade performance metrics, to a remote Telegram chat ID.

wolf-howl