extract-skill-from-conversation

This file itself appears non-malicious and functions as an orchestration wrapper for topic-inference workflows. The main security risk is supply-chain and data-exfiltration through external dependencies: the 'fabric' CLI and the sibling parse_conversation.sh and prompt files. These external components can transmit unredacted conversation text and any embedded secrets/PII to remote providers. Operational issues (brittle JSON extraction, inconsistent cleanup, lack of timeouts/concurrency limits) increase chances of data leakage or residual sensitive artifacts. Recommended mitigations: validate and pin the fabric binary (path, checksum, or signature), audit parse_conversation.sh and prompt files, add pre-send redaction or secret detection, improve JSON extraction (use jq to parse JSON output robustly), ensure reliable trap/cleanup semantics (use EXIT consistently), add timeouts and concurrency limits, and avoid writing sensitive data to disk when possible (use securely handled pipes or in-memory streams).