The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/excel-siigo.sh generates a temporary batch file (.bat) at runtime containing a full command line, which it then executes using cmd.exe. While the script attempts to escape double quotes, it does not sanitize other shell metacharacters (such as &, |, ^, <, >) that might be present in environment variables like SIIGO_EXE or SIIGO_EMPRESA. This could allow for arbitrary command execution if these variables are maliciously crafted.
[CREDENTIALS_UNSAFE]: The scripts/excel-siigo.sh wrapper writes the SIIGO_CLAVE (user password) in plain text directly into the temporary batch file created on the file system. Although the script contains an rm command to delete the file after execution, the sensitive credential remains exposed on disk during the execution window and would persist if the process is terminated prematurely or crashes.
[COMMAND_EXECUTION]: Multiple Python scripts (scripts/excel-siigo.py, scripts/parse-filepath.py) utilize subprocess.run() to invoke external binaries like EXCELSIIGO.exe and cygpath. The paths for these executions are derived from environment variables, creating a risk of unauthorized command execution if the execution environment is compromised.
[DATA_EXPOSURE]: The skill instructions in SKILL.md and the implementation in scripts/excel-siigo.py facilitate the extraction of sensitive financial and corporate data from the SIIGO Pyme ERP into Excel files. While this is the intended functionality, the skill handles sensitive information such as accounting movements, third-party details, and inventory data without explicit data-handling guardrails for the AI agent.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from .xlsx files. This creates a surface for indirect prompt injection where malicious instructions embedded within spreadsheets could influence the agent's behavior during data aggregation, filtering, or downstream reporting tasks.
Ingestion points: Commands like scripts/excel-siigo.sh that ingest file paths and the downstream processing of .xlsx files via openpyxl or pandas as described in SKILL.md section 13.3.
Boundary markers: None identified. There are no explicit instructions for the agent to ignore or delimit instructions found within the processed Excel data.
Capability inventory: The skill possesses capabilities for shell command execution (subprocess.run, cmd.exe), file system writes (.xlsx, .log, .bat), and credential handling.
Sanitization: The Python wrapper uses argument lists for subprocess.run(), which provides some protection. However, the Bash wrapper uses string interpolation into a batch file with incomplete sanitization.

siigo-pyme-excel