The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core function is to ingest and process large volumes of untrusted third-party documents (e.g., contracts, vendor files). This creates a surface for indirect prompt injection where malicious instructions hidden in documents could attempt to influence the agent's output or behavior.
Ingestion points: Local document folders, Virtual Data Rooms (VDR), and cloud document providers (Google Drive, Feishu, Nutstore) as specified in SKILL.md.
Boundary markers: The skill employs a strictly typed YAML schema (.review-schema.yaml) and mandates verbatim quoting of source text to distinguish between interpreted data and raw evidence.
Capability inventory: The agent can write various file types (.xlsx, .csv, .md, .yaml), access network APIs for Google Sheets, and execute Python snippets for spreadsheet generation.
Sanitization: The skill contains explicit 'Formula Injection Defense' instructions in references/excel-output.md and references/gsheets-output.md, requiring all document-sourced text to be prefixed with a single quote (') and properly escaped.
[DYNAMIC_EXECUTION]: The skill uses Python scripts (e.g., python3 -c "import openpyxl") and the openpyxl library to dynamically generate and format Excel workbooks based on the extracted data.
[COMMAND_EXECUTION]: The instructions include a step to check for and potentially install the openpyxl package via pip3 if it is not already present in the environment.
[DATA_EXPOSURE]: The skill accesses local configuration files within ~/.claude/plugins/config/ to load deal contexts, diligence structures, and thresholds. While this is standard for the 'Claude for Legal' framework, it involves reading sensitive internal configuration data.

tabular-review