The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection via the processing of image metadata.\n
Ingestion points: scripts/main.py extracts the XResolution TIFF tag (tag 282) from input files to determine scale calibration without user verification.\n
Boundary markers: The skill lacks instructions or markers to distinguish between legitimate calibration data and malicious strings embedded in metadata.\n
Capability inventory: The skill possesses file read (PIL.Image.open), file write (image.save), and console output capabilities.\n
Sanitization: No sanitization is performed on the data retrieved from metadata before it is rendered into the image output or printed to the terminal.\n- [METADATA_POISONING]: The skill includes misleading documentation and self-asserted security claims.\n
Evidence: The usage examples and parameter table in SKILL.md list flags like --image, --scale, and --unit, which do not match the --input, --scale-length, and --scale-unit arguments implemented in scripts/main.py. This would lead to execution failures if an agent follows the documentation.\n
Evidence: The inclusion of a pre-calculated audit file (microscopy-scale-bar-adder_audit_result_v4.json) that asserts a 'PASS' security status is a deceptive practice that may circumvent independent security reviews or mislead users.\n- [PRIVILEGE_ESCALATION]: The skill's path validation logic is insufficient to meet its stated security goals, allowing for potential boundary bypass.\n
Evidence: While SKILL.md claims to reject absolute paths outside the workspace, the check_path_traversal function in scripts/main.py only checks for ../ sequences and paths starting with ... It does not utilize os.path.isabs() or verify workspace boundaries, potentially allowing an agent to read from or write to arbitrary system locations if provided with absolute paths.

microscopy-scale-bar-adder