The Agent Skills Directory

[SAFE]: Secret Management
The scripts/crawlora.sh helper script securely handles the CRAWLORA_API_KEY by writing it to a temporary configuration file with restricted permissions (chmod 600). This ensures the secret is not exposed in process listings (ps) or via command-line arguments.
[SAFE]: Network Restriction and Path Validation
The skill implements a strict allowlist of 71 specific API endpoints and enforces the correct HTTP method (GET/POST) for each. It also validates the path structure to prevent directory traversal or URI smuggling.
[SAFE]: Local File Disclosure Prevention
The shell script explicitly rejects the @ character in query parameters and uses stdin (--data-binary @-) for POST bodies. These measures prevent attackers from using curl's local file upload shorthand to exfiltrate sensitive files from the agent's environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted search queries and returns data from third-party travel platforms. While this creates a surface for indirect prompt injection, the risk is mitigated by the skill's restricted output format (normalized JSON) and the absence of file-writing or arbitrary command-execution capabilities. Evidence found in SKILL.md (ingestion point: query param) and scripts/crawlora.sh (tool execution).

travel-hotel-research