The Agent Skills Directory

[COMMAND_EXECUTION]: The SKILL.md preamble executes two Python scripts, telemetry/version_check.py and telemetry/telemetry_init.py, whenever the skill is initialized. These files are not provided in the skill package, making their specific behavior unverifiable. Although they are described as local telemetry logging tools, their absence from the source prevents a full safety audit of the startup routine.
[COMMAND_EXECUTION]: The scripts/cold-outbound-sender.py script uses subprocess.run to execute a CLI tool (defaulting to gog gmail send) for email delivery. The specific CLI command used is configurable via script arguments, which could be misused if execution happens in a shell context with untrusted inputs.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because it ingests untrusted data from external sources and processes it through an LLM scoring panel.
Ingestion points: scripts/competitive-monitor.py performs web scraping of competitor pricing pages and blogs; scripts/lead-pipeline.py fetches lead data from the Apollo API.
Boundary markers: The instructions do not define clear delimiters or "ignore instructions" wrappers for the content scraped from external competitor websites.
Capability inventory: The agent has the capability to write to the Instantly API, send emails via SMTP, and execute CLI commands via the sender script.
Sanitization: The skill includes a significant positive security feature in scripts/cold-outbound-sender.py named validate_outbound, which uses regular expressions to detect and block emails containing API keys (e.g., sk- patterns), authorization headers, or local file paths. Additionally, scripts/competitive-monitor.py truncates large scraped content to 500k characters.

outbound-engine