self-improving-agent

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The enhanced-failure-logger.js explicitly records WebFetch tool inputs/results (it extracts toolInput.url and logs hookData.result/error into failures_detailed.jsonl and learnings.jsonl), and those learnings are shown at session start (the "Learning Review"), so the skill can ingest and display arbitrary/untrusted external web/API content provided via WebFetch calls.