The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates multiple local scripts to perform financial operations, which grants the agent authority to run Python code and shell commands.
Evidence: Execution of post_journals.py, roll_tb.py, generate_html_report.py, scripts/depth_gates.py, and scripts/close_engagement.py.
[REMOTE_CODE_EXECUTION]: The skill utilizes a package runner command to handle data extraction, which may involve downloading and executing external code from unverified sources.
Evidence: Use of npx … extract within the work sequence for bank statement processing.
[PROMPT_INJECTION]: The skill is designed to ingest and process arbitrary user-provided folders and financial documents, creating a surface for indirect prompt injection attacks.
Ingestion points: The skill reads from the current working directory (cwd), user-specified paths, and the clients/<slug>/source/** directory.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat data in source documents as untrusted or to ignore embedded instructions.
Capability inventory: The skill has the capability to write to the filesystem (engagement_state.json, transactions.json), execute Python scripts, and run Node.js packages.
Sanitization: The instructions do not specify any validation or sanitization steps for the contents of the ingested bank statements or folder dumps.

full-engagement-pipeline