The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes a file titled journal-matchmaker_audit_result_v2.json which contains simulated audit results, claiming 'security: PASS' and 'stability: PASS'. This is a self-referential deceptive pattern intended to influence security analysis tools or the agent's perception of its own safety by providing false authoritative claims about its integrity.
[DATA_EXPOSURE]: The script scripts/main.py provides a --file argument that allows the user to specify a file path to be read as the 'abstract'.
Ingestion points: The main() function in scripts/main.py reads file contents via open(args.abstract, 'r', encoding='utf-8') if the --file flag is set.
Boundary markers: The skill does not implement restrictions on the paths provided, allowing potential access to sensitive system or configuration files if the agent is manipulated into specifying them.
Capability inventory: The script performs local file reads and extracts keywords. While it does not exfiltrate data over a network, the keyword extraction process could inadvertently reveal information from sensitive files (like .env or SSH keys) if they are processed as abstracts.
Sanitization: The script performs basic character filtering using regex re.sub(r'[^\w\s]', ' ', text), which is intended for text processing rather than preventing unauthorized file access.

journal-matchmaker