Introspection Review Tracker — Reference Documentation

This skill is reference documentation for the automated review outcome persistence hook. The llmem track-review CLI command fires automatically after every adversarial code review, creating self_assessment memories without requiring manual agent intervention. This skill documents how the hook works, the severity-to-category mapping, and the content format.

How Persistence Works

Persistence happens mechanically via llmem track-review (CLI). The agent must run this command after every adversarial code review — it is not optional.

CLI: llmem track-review

The primary mechanism. Run after every review:

# For a review with findings (batch mode):
llmem track-review --finding-file /tmp/review-findings.json --context "handler.py"

# For a single finding:
llmem track-review --category NULL_SAFETY --what-happened "missing null check" --context "handler.py:42" --severity Required --caught-by self-review