The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/batch_generate.py uses subprocess.check_call to execute the secondary script scripts/generate_qr.py using arguments derived from untrusted CSV data. While it uses a list for arguments which prevents shell injection, it incorporates unvalidated external input.
[COMMAND_EXECUTION]: Path traversal vulnerability in scripts/batch_generate.py. The id field from the CSV input is used directly to construct file paths (os.path.join(args.outdir, f\"{_id}.png\")) without sanitization. An attacker could provide a CSV with an id like ../../filename to write files outside the designated output directory.
[PROMPT_INJECTION]: Indirect prompt injection surface identified.
Ingestion points: scripts/batch_generate.py reads user-provided CSV files and processes each row.
Boundary markers: Absent; there are no delimiters or instructions to the agent to ignore content within the CSV data.
Capability inventory: The skill allows file system writes and subprocess execution across both Python scripts.
Sanitization: While URL validation and XML escaping are present in scripts/generate_qr.py, the skill lacks path sanitization for file names in scripts/batch_generate.py.

qr-code-generator