v1.106.0

What's Changed

Added

• Unified Session Store (SQLite + FTS5) — New ll-session CLI and .ll/session.db provide a queryable, full-text-indexed store of Claude Code session events. Supports search --fts, recent --kind, and backfill subcommands. (FEAT-1112)
• Context Window Analytics (ll-ctx-stats) — New CLI command and PostToolUse hook capture per-tool byte vs. context savings from the session DB, giving visibility into context consumption across tools. (FEAT-1160, FEAT-1623, FEAT-1624, FEAT-1625, EPIC-1626)
• ll-history and analyze-workflows query session DB — Both commands now read from the unified session store instead of re-parsing raw log files. (ENH-1621)
• ll-session wired across docs, help, and config — Reference docs, init permissions, and CLAUDE.md now surface the new session store. (ENH-1619)

Fixed

• Nested loops discoverable in ll-loop list and doc verifier — Recursive discovery now finds runnable loops under loops/oracles/ and other subdirectories. (BUG-1633, BUG-1634)
• FSM failure-terminal diagnostic convention — Tooling, docs, create-loop wizard, and validation now enforce a diagnose state before failure terminals to surface root causes. (BUG-1607)
• SKILL.md block-scalar description parsing — doc_counts and two CLI parsers now resolve YAML block-scalar descriptions via yaml.safe_load instead of regex.

Changed

• README agent description — Rephrased to be drift-proof rather than naming a fixed agent count. (ENH-1441)
• Status one-shot cleanup — Normalized non-canonical status: values (e.g. completed → done) across all issue files via ll-migrate-status. (ENH-1551)

Other

• Multiple issue refinements and plan documents (ctx-stats epic, ll-loop list readability, Hermes Agent PRD).
• ruff format passes across doc_counts, ctx-stats files, and tests.

Full Changelog: v1.105.0...v1.106.0