The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests job postings directly from LinkedIn via BrightData. These postings are attacker-controlled external sources that may contain malicious instructions designed to hijack the agent's execution flow.
Ingestion points: Data is fetched through scripts/fetch_linkedin_job.py and scripts/search_linkedin_jobs.py which communicate with the BrightData API.
Boundary markers: The skill lacks delimiters (such as XML tags or unique markers) to separate the untrusted job data from the agent's instructions. It prints raw JSON directly to stdout.
Capability inventory: The agent context typically includes access to shell commands, file system writes, and further network operations, which could be targeted by an injection attack.
Sanitization: There is no evidence of sanitization, filtering, or escaping of the job_description_formatted, job_summary, or other text fields before they are presented to the agent.
[EXTERNAL_DOWNLOADS]: The skill fetches data from api.brightdata.com. BrightData is a well-known technology service, and the network operations are consistent with the skill's stated purpose of data retrieval.
[COMMAND_EXECUTION]: The skill defines local Python scripts (scripts/search_linkedin_jobs.py and scripts/fetch_linkedin_job.py) meant to be executed by the agent to perform network requests and data processing.

linkedin-jobs