The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as email exports (.eml files) and incoming files using AI prompts, which could contain hidden instructions.
Ingestion points: Untrusted data enters the agent context through the code-along/01-data-intake-review/incoming/ and code-along/03-offline-communications-triage/eml/ directories.
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are identified in the provided prompt templates.
Capability inventory: Example shell scripts include file reading operations (cat) and interaction with AI CLI tools.
Sanitization: No explicit sanitization, escaping, or filtering of the processed content is documented in the workshop instructions.
[PERSISTENCE]: The workshop provides explicit instructions and examples for setting up cron jobs to automate log generation, which creates a persistence mechanism on the user's system.
Evidence: Cron entry examples provided in the Operational Memory section for running log generation scripts at scheduled intervals.
[EXTERNAL_DOWNLOADS]: The skill documentation instructs users to clone a repository from a third-party GitHub account to begin the workshop.
Evidence: git clone https://github.com/dandenney/webinars-build-your-own-executive-assistant-with-openclaw.git in the Installation & Setup section.
[COMMAND_EXECUTION]: The skill includes shell script templates that demonstrate how to automate workflows by executing commands and piping data to CLI tools.
Evidence: The generate-daily-log.sh pattern in the Operational Memory section uses shell variables and pipes content from files into external AI CLI tools.

openclaw-executive-assistant-workshop