The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for creating a persistence mechanism on the host system via a scheduled task.
Evidence: The "Scheduling" section in SKILL.md explicitly instructs the configuration of a crontab entry to execute code at regular intervals: 0 7 * * 1 python3 run_skill.py newsletter-signal-scanner --client <client-name>.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes untrusted data from external newsletter emails.
Ingestion points: agentmail capability used to fetch raw email bodies from an external inbox (described in Phase 1 of SKILL.md).
Boundary markers: Absent. The skill extracts keyword context directly from email bodies and interpolates them into a markdown report without using XML tags, delimiters, or "ignore instruction" warnings.
Capability inventory: The skill possesses file-write capabilities, saving intelligence reports to the clients/<client-name>/intelligence/ directory.
Sanitization: The skill performs basic HTML-to-plain-text stripping, but lacks semantic sanitization to prevent embedded instructions in the newsletters from influencing the agent's logic during the digest creation or subsequent processing.

newsletter-signal-scanner