The Agent Skills Directory

[METADATA_POISONING]: The skill package contains a file named postdoc-fellowship-matcher_audit_result_v2.json which presents itself as an official security audit report. It claims a safety grade of 'Production Ready' and a security score of '10/12'. This self-referential content is deceptive and intended to pre-empt or misguide independent security assessments.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection due to the lack of input handling controls.
Ingestion points: User-provided parameters for nationality, years since PhD, research field, and applicant name, as defined in SKILL.md and processed by scripts/main.py.
Boundary markers: Absent. There are no instructions or delimiters to prevent the agent from interpreting instructions embedded within user data.
Capability inventory: Execution of a local Python script (scripts/main.py) via the shell and generation of a formatted report.
Sanitization: Absent. The skill does not validate or escape the string content of user inputs before printing them to the output report header.
[EXTERNAL_DOWNLOADS]: The documentation references official program websites for fellowship details, including well-known sources such as the National Institutes of Health (NIH), the National Science Foundation (NSF), and the European Commission (MSCA).
[COMMAND_EXECUTION]: The skill workflow relies on executing the local script scripts/main.py with user-controlled arguments. While this is the intended function, the absence of input validation makes it a potential vector for command argument manipulation.

postdoc-fellowship-matcher