The Agent Skills Directory

[CREDENTIALS_UNSAFE]: Plaintext credential exposure in logs. Both the PowerShell and Python scripts are designed to print the complete command-line string to the console before execution. This command string includes the database password if one is provided as an argument, making sensitive credentials visible in the agent's output history.
Evidence in scripts/db-dump-xml.ps1: Write-Host "Running: 1cv8.exe $($arguments -join ' ')"
Evidence in scripts/db-dump-xml.py: print(f"Running: 1cv8.exe {' '.join(arguments)}")
[COMMAND_EXECUTION]: The skill executes external 1C platform binaries via subprocesses. The execution path and command arguments are dynamically constructed based on user input and local project configuration files.
Evidence in scripts/db-dump-xml.py: subprocess.run([v8path] + arguments, capture_output=True, text=True)
Evidence in scripts/db-dump-xml.ps1: Start-Process -FilePath $V8Path -ArgumentList $arguments -NoNewWindow -Wait -PassThru
[PROMPT_INJECTION]: Indirect prompt injection surface through configuration ingestion. The skill reads and acts upon data from an external JSON file without explicit validation or safety boundaries.
Ingestion points: .v8-project.json (referenced in SKILL.md for platform paths and database aliases).
Boundary markers: Absent; no instructions are provided to the agent to distinguish between configuration data and embedded commands.
Capability inventory: Execution of local binaries and file system writes via bundled scripts.
Sanitization: Absent; the scripts process configuration values as raw strings for command-line construction.

db-dump-xml