RagLeapWith v0.7.0, ragleap-core ships all 9 standard agentic patterns: self-correction/reflection, chain-of-thought, tree-of-thought, supervisor routing, and sub-agent spawning — completing the set alongside the tool registry, observability, human-in-the-loop, and memory patterns that were already in place.
What's actually new
- Broad actionable agents — webhook, Slack, and email action senders, plus an LLM-driven action planner. Every action still routes through the same approval gate as everything else.
- Security hardening — 7 stack-trace-exposure alerts and 4 unflagged error leaks fixed, runtime-created roles now auto-detected as sensitive, Docker ports bound to localhost by default, and an optional
RAGLEAP_API_KEYglobal auth middleware. - Cost control — a usage ledger with token budget caps, global and per-role, covering streaming responses too.
- Contributor readiness — automated guard tests so a half-added role can't ship unguarded, plus a new "Adding an AI Employee" guide in CONTRIBUTING.md.
Full test suite: 477 passing. See the full release notes for the complete PR list.