The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could potentially contain malicious instructions aimed at influencing the agent's behavior during the referral mining process.\n
Ingestion points: The skill ingests data from a LinkedIn Connections.csv export and an icp.json configuration file via the bundled scripts/score_connections.py script.\n
Boundary markers: There are no explicit boundary markers, delimiters, or 'ignore instructions' warnings used when the processed data (the shortlist) is presented to the agent for further reasoning.\n
Capability inventory: The scripts/score_connections.py script performs local file system operations (read/write) and utilizes the Python standard library to process user-supplied data.\n
Sanitization: The skill lacks explicit sanitization or filtering of external text content (such as job titles or company names from the CSV) before it is used by the agent to generate 'why now' outreach justifications.\n- [METADATA_POISONING]: The NOTICE.md file contains an anomalous future date ('2026-10-06') cited as the date the skill was copied from its source repository. While this is likely a typographical error and not a malicious attempt to deceive, it represents inconsistent metadata that could technically be used for versioning deception.

referrals