Skip to content

This release adds 3 notable features for engineering teams evaluating rollout.

Published 1mo MCP Developer Tools
✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

agent-reliability ai-agents ai-cost-optimization ai-safety amp claude-code
+14 more
codex cursor developer-tools feedback-loop gemini guardrails mcp mcp-server opencode pre-action-checks reduce-llm-cost save-llm-tokens thompson-sampling thumbgate

Affected surfaces

deps

Summary

AI summary

Adds MCP leader‑agent coordination tools for evidence gating, context distribution, and observability reporting.

Full changelog

[email protected]

Release Links

  • npm: https://www.npmjs.com/package/thumbgate/v/1.9.0
  • GitHub Release: https://github.com/IgorGanapolsky/ThumbGate/releases/tag/v1.9.0
  • Compare: https://github.com/IgorGanapolsky/ThumbGate/compare/v1.8.0...v1.9.0
  • Publish workflow: https://github.com/IgorGanapolsky/ThumbGate/actions/runs/24645400274
  • npm published at: 2026-04-20T02:20:15.658Z
  • npm shasum: f98b34ef347380f42f7a52ed6e119e24f8513768
  • npm tarball: https://registry.npmjs.org/thumbgate/-/thumbgate-1.9.0.tgz
  • Release ref: 279dfd57320b341d9659847b1b6a014452721099

npm Email Companion

npm controls the native "Successfully published" email template, so the email itself stays short. Treat this generated artifact as the full release-note companion for that email: it carries the Changeset summaries, CHANGELOG entry, publish workflow, npm tarball, and shasum when available.

Full Changeset Release Notes

Minor Changes

.changeset/bright-mcp-research-gates.md

Add high-ROI MCP agent-discovery and research-loop surfaces.

  • Publish progressive MCP discovery manifests under .well-known/mcp, including a compact tool index, per-tool schema URLs, skill manifests, and application manifests so AI agents and crawlers can load ThumbGate without stuffing every tool into context.
  • Add run_autoresearch as a bounded MCP tool for Shopify-style baseline, hypothesis, holdout, and keep/discard loops around revenue and reliability metrics.
  • Add plan_multimodal_retrieval so operators can plan screenshot, PDF, dashboard, and proof-artifact retrieval using multimodal sentence-transformer guidance, Matryoshka-style dimensions, reranker metrics, and hard-negative holdouts before spending GPU time.

CHANGELOG.md Entry

1.9.0

Minor Changes

  • #957 68b3de3 Thanks @IgorGanapolsky! - Agentic-engineering Leader Agent endpoints: completion gate, swarm coordinator, and unified observability.

    Adds three MCP tools that lift ThumbGate from a bag of primitives into a Leader-Agent coordination layer (per the LangChain agentic-engineering framing — worker agents consume, leader endpoints coordinate and verify):

    • require_evidence_for_claim — completion gate. Wraps verifyClaimEvidence with a first-class blocking boolean and mode (blocking default, advisory). Records the decision to the audit trail under gateId: completion_claim. Agents call this before declaring done/fixed/shipped; hooks honor the blocking flag to stop evidence-free completion claims.
    • distribute_context_to_agents — swarm coordinator. Constructs one context pack via constructContextPack and records a context_pack_distributed provenance event per named agent (dedup'd, capped at MAX_AGENTS=32, TTL defaults to 15 minutes). Replaces N independent context derivations by auto-agents (perplexity-bug-resolver, codex-reviewer, grok-x-intelligence, etc.) with one shared pack.
    • session_report — unified observability rollup. Aggregates feedback stats, gate stats, and windowed provenance into a single LangSmith-style report. windowHours clamps to [1, 720]; invalid/missing input falls back to the 24h default. Errors in any section are isolated via a per-section errors map so one broken source doesn't sink the report.

    Exposed in default, essential, readonly, and dispatch MCP profiles. No OpenAPI surface changes (MCP-only). Ships with 24 new tests across tests/swarm-coordinator.test.js, tests/session-report.test.js, and tests/require-evidence-gate.test.js; regression runs clean across test:api (834), test:gates (198), test:tool-registry (11), test:proof (96), test:deployment (55), test:e2e (29), test:workflow (98), test:schema (8), and test:mcp-config (9).

Patch Changes

  • #925 e0c89bc Thanks @dependabot! - Bump transitive dependency protobufjs from 7.5.4 to 7.5.5 (security/bugfix release). Lockfile-only change via Dependabot.

  • #947 b326963 Thanks @IgorGanapolsky! - Mailer module now accepts THUMBGATE_RESEND_API_KEY as a fallback for the bare RESEND_API_KEY, matching the dual-read behavior already implemented in scripts/billing.js. Prevents a silent "skipped: no_api_key" regression if an operator sets only the prefixed variable name. Adds a positive unit test that sends with only the prefixed variant set.

  • Fix repo bootstrap so worktree checkouts can create local MCP wiring and info exclude entries without failing on .git pointer files.

Verification Standard

  • Publish only runs from main after version sync, tests, and runtime proof pass.
  • The npm package is smoke-tested after publish by installing thumbgate@VERSION in a clean runtime.
  • GitHub Release notes are generated from Changesets, not only GitHub auto-generated PR titles.

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Track IgorGanapolsky/mcp-memory-gateway

Get notified when new releases ship.

Sign up free

About IgorGanapolsky/mcp-memory-gateway

Pre-action gates that prevent AI coding agents from repeating known mistakes. Captures explicit feedback, auto-promotes failures into prevention rules, and enforces them via hooks.

All releases →

Beta — feedback welcome: [email protected]