Skip to content

This release includes breaking changes for platform teams planning a safe upgrade.

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

Summary

AI summary

Replace emoji logo with crisp teal-on-navy TG gate monogram across all customer‑facing surfaces.

Full changelog

[email protected]

Release Links

  • npm: https://www.npmjs.com/package/thumbgate/v/1.6.0
  • GitHub Release: https://github.com/IgorGanapolsky/ThumbGate/releases/tag/v1.6.0
  • Compare: https://github.com/IgorGanapolsky/ThumbGate/compare/v1.5.8...v1.6.0
  • Publish workflow: https://github.com/IgorGanapolsky/ThumbGate/actions/runs/24615891351
  • npm published at: 2026-04-18T23:04:06.366Z
  • npm shasum: 498206642719a266186dbdfce1b7f3f5dc15df07
  • npm tarball: https://registry.npmjs.org/thumbgate/-/thumbgate-1.6.0.tgz
  • Release ref: a5fc8b33f53b34e4db15a12903fae18dc221c629

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

No changed .changeset/*.md entries were detected for this release range.

CHANGELOG.md Entry

1.6.0

Minor Changes

  • #931 8161e51 Thanks @IgorGanapolsky! - Replace the 👍👎 emoji header logo with a crisp teal-on-navy TG gate monogram across every customer-facing surface (landing page, dashboard, lessons, Pro, Learn hub, Learn articles, SEO-GSD generated pages, and the post-checkout Context Gateway Activated page). Ships public/assets/brand/thumbgate-mark.svg, refreshed checkout PNGs, public/thumbgate-icon.png, and public/og.png; wires rel="icon", apple-touch-icon, and og:image tags on the main pages so tab icons, Stripe thumbnails, and link previews render the brand consistently instead of OS-dependent Unicode glyphs or the old chart-like mark. Hero-thumbs decorative art on the landing page is preserved intentionally.

  • #922 30cf554 Thanks @IgorGanapolsky! - Cursor plugin: fix broken promises and add real wiring. README claimed npx thumbgate init --agent cursor worked; it didn't. Added cursor detection + dispatcher + wireCursorHooks that writes .cursor/mcp.json with the ThumbGate MCP server (preserves other entries, idempotent). Added dedicated "🎯 Cursor plugin" card to the landing page Compatibility section with a real install URL. Added Cursor install link to the First-Dollar step 1 and hero secondary CTAs. 5 new tests guard the wiring. Also hardens landing-page pills into real <a> clickable links with hover/focus states.

  • #909 a9e0f0d Thanks @IgorGanapolsky! - Add Insights tab to dashboard with interactive Chart.js charts (feedback trend, lessons generated, gate effectiveness), clickable pipeline visualization, and data consistency fix across all stat paths.

  • #902 94d3882 Thanks @IgorGanapolsky! - Add ElevenLabs-based demo voiceover automation (scripts/generate-demo-voiceover.js) that extracts narration from the canonical demo video script and synthesizes an mp3 via the ElevenLabs TTS API. Promote the landing page demo video out of the collapsed <details> into a visible inline hero embed, add a 90-second demo section to the top of README.md, and rewrite the Show HN launch draft around the token-cost mission. Schedule reply-monitor.yml daily at 13:00 UTC with LinkedIn environment passthrough, and ship two LinkedIn ops docs: a 2-minute daily manual-check runbook and a fully-drafted LinkedIn Community Management API application package.

  • #926 d8d1047 Thanks @IgorGanapolsky! - Add lightweight durable-step helper (scripts/durability/step.js) inspired by Vercel Workflows' "use step" pattern. Wraps external I/O with uniform retry + idempotency semantics without pulling in a full durable-execution runtime:

    • runStep(name, opts, fn) — retry with exponential backoff, classifying transient vs permanent errors (HTTP 429/5xx retry, 4xx bail, socket codes retry, nonRetryable flag bails immediately)
    • idempotencyKey(...parts) — stable SHA-256-derived 32-char key for safe POST retry

    Wired into three highest-leverage call sites:

    1. Zernio publisher (publishPost, schedulePost) — adds Idempotency-Key header so retried POSTs collapse to one published post on Zernio's side. Plan-quota errors are tagged nonRetryable to avoid wasting retries on 402-equivalents.
    2. LanceDB vector write (upsertFeedback) — survives transient filesystem contention (EBUSY / lock timeouts) with 2-retry backoff; embedding is pure CPU so not retried.
    3. Anthropic SDK call (callClaude) — retries 429/5xx, bails on malformed-prompt / auth errors. Contract-preserving: callers still get null on permanent failure.

    21 unit tests cover success/retry/exhaustion/nonRetryable paths and idempotency-key stability.

    Not a Vercel Workflows migration — deliberately scoped to capture ~70% of the reliability benefit with ~60 lines of code and zero new infrastructure.

  • #912 f1fccae Thanks @IgorGanapolsky! - PreToolUse hook now injects semantically-relevant past negative lessons into additionalContext before every tool call. Turns ThumbGate from a passive log into an active governor: captured lessons surface at decision time so the agent sees its past mistakes BEFORE executing, not after. Shipped by default via thumbgate init --agent claude-code|codex — users already running that get the enforcement automatically on next hook invocation.

  • #952 dadf4ba Thanks @IgorGanapolsky! - Add buildRecentCorrectiveActionsContext to scripts/gates-engine.js: surfaces the 3 most recent captured mistakes (from memory-log.jsonl, last 24h) as hookSpecificOutput.additionalContext on every tool call. Plugs the cold-start gap where a just-captured mistake would otherwise wait for semantic match or the recurring-pattern threshold before reaching the agent's context.

  • #889 bc79ae2 Thanks @IgorGanapolsky! - Reposition ThumbGate around a single sharp mission: stop your AI from making the same mistake twice. Repeated AI mistakes cost real money in tokens — one thumbs-down captures the lesson and ThumbGate blocks that exact pattern on every future call, across every agent.

    • New hero copy everywhere — plain-English, pain-point-in-one-sentence, no buzzword cadence. Applied to landing page, README, meta/OG tags, JSON-LD, package.json, plugin.json, and config/github-about.json.
    • Live "💸 Tokens Saved" counter on the dashboard. New scripts/token-savings.js helper (21 tests, Sonnet-blended default) turns blocked-gate + bot-deflection counts into a live token + dollar estimate. Swap in your own model mix to honestly reflect your Anthropic / OpenAI bill.
    • New ClawHub / OpenClaw distribution skilldist/clawhub-skill/SKILL.md — ready for npm run clawhub:publish once authenticated. Expands the distribution surface to the OpenClaw skill marketplace alongside the Claude Extension, Codex plugin, npm, and MCP marketplaces.
    • SEO blog post docs/marketing/blog-token-cost-mission.md ranking on "save Claude tokens" / "reduce LLM cost" / "AI agent token waste."
    • Pre-validated social pack docs/marketing/token-cost-mission-social-pack.md (X/Threads/LinkedIn/HN/Reddit/TikTok) under every platform's char limit.
  • #922 30cf554 Thanks @IgorGanapolsky! - Real token-savings on the dashboard — no more hardcoded numbers. The Insights tab now shows $ saved computed from actual gate-stats.blocked count × conservative tokens/block × published Sonnet/Opus/Haiku prices. Zero blocks → shows $0.00 honestly (not a marketing placeholder). Methodology (input/output tokens per block, model mix, blended price) is disclosed inline. Landing page hero still uses the "Sample" demo — dashboard now uses real data.

  • #931 8161e51 Thanks @IgorGanapolsky! - Send a branded welcome email with the license key and activation command whenever
    checkout.session.completed fires. Uses Resend (RESEND_API_KEY) with
    [email protected] as the default sender so the webhook keeps working
    without a verified domain. If the key is unset, the webhook logs a warning and
    continues — the license key is always persisted regardless of email state.

Patch Changes

  • #919 7be5cc6 Thanks @IgorGanapolsky! - Fix recurring regression: add public/pro.html, public/blog.html, public/learn.html to npm files whitelist so they actually ship. New tests/public-package-parity.test.js asserts (a) every HTML in public/ is in whitelist, (b) every whitelist entry exists on disk, (c) no stale $99/seat Team pricing ships. Prevents the packaging-bug pattern that hit 1.5.0, 1.5.1, 1.5.3.

  • #949 c8b31e9 Thanks @IgorGanapolsky! - Add an Autoresearch Safety Pack acquisition wedge with a buyer guide, landing-page CTAs, LLM context, SEO/GEO seeds, and regression tests for self-improving agent safety discovery.

  • #918 f063c1a Thanks @IgorGanapolsky! - Version bump to 1.5.3 — publish the landing page congruence fixes, dashboard deep-linking, and README corrections that merged as #914 after 1.5.2 had already been published from #911.

  • #858 204dbbe Thanks @IgorGanapolsky! - Expose the ChatGPT Actions OpenAPI YAML import before bearer auth and document the GPT Builder bearer key setup.

  • #869 5bac711 Thanks @IgorGanapolsky! - Deflect checkout crawlers and link-preview bots before creating Stripe sessions so revenue telemetry reflects real buyer intent.

  • #932 bc9f0c0 Thanks @IgorGanapolsky! - Replace stale checkout logo assets with ThumbGate brand marks and add activation email delivery instrumentation for trial provisioning.

  • #877 1c7140e Thanks @IgorGanapolsky! - Add prominent "Install Claude Extension →" CTA to the landing page hero section, matching the existing Codex plugin link. Links to the .mcpb bundle download with PostHog tracking.

  • #922 30cf554 Thanks @IgorGanapolsky! - Compat cards that promise a download now link directly to the release asset instead of a docs/source page. Codex plugin card was linking to INSTALL.md source despite saying "download the zip"; Claude Desktop Extension card was linking to a guide page despite saying "install the .mcpb bundle today". Both now go straight to the .zip / .mcpb on GitHub Releases. Setup-instruction secondary links preserved inline. New test landing-page-claims.test.js guards against regression: any compat card with "Download" in the arrow MUST have href pointing at releases/.../download/.

  • #935 1785ca9 Thanks @IgorGanapolsky! - Clarify the Codex plugin marketing card so it sends users to the install page and keeps MCP directory install copy on ThumbGate's npx path.

  • #927 4742253 Thanks @IgorGanapolsky! - Codex MCP installs now resolve thumbgate@latest when Codex starts the MCP server or hook bundle, instead of preferring a stale already-installed runtime binary. The repo-local Codex plugin, standalone bundle config, README, landing page, and distribution docs now advertise the auto-updating Codex plugin path truthfully while preserving local source fallback for unpublished development builds.

  • #895 fbc66c9 Thanks @IgorGanapolsky! - Wire the full Codex hook bundle during init and add the Codex status line target to the generated local config.

  • #880 7ddf48f Thanks @IgorGanapolsky! - Landing page conversion overhaul: restructure visual hierarchy for conversion

    • Hero: single dominant CTA (install command + Install Free CLI), secondary CTAs grouped and visually demoted
    • Terminal demo: moved immediately after hero to show the product before any explanation
    • Trust bar: added above-the-fold honest social proof (MIT, GitHub stars, local-first, 6 integrations)
    • Hero headline: rewritten for clarity ("Stop expensive AI agent mistakes before they happen")
    • Nav: simplified to 4 visible links (How It Works, Pricing, FAQ, GitHub) + Install Free CTA
    • Enterprise intake form: collapsed behind a details/summary toggle to reduce page overwhelm
    • Newsletter section: simplified headline, removed internal jargon ("Buyer Follow-Up" → "Stay Updated")
    • Final CTA: simplified to 2 primary actions, secondary CTAs visually demoted
    • CSS: added conversion hierarchy styles to reduce visual weight of secondary sections
    • Pro pricing card: added email capture input (pro-email) for 7-day trial flow
    • All 36 landing page tests pass
  • #906 6db3ab1 Thanks @IgorGanapolsky! - Rewrite postinstall banner to drive first-dollar conversion. Lead with concrete token-waste pain point, add tracked /go/pro click-through (UTM: source=npm, medium=postinstall, campaign=first_dollar) alongside direct Stripe link, clean up ragged box formatting. Every npm install sees this banner — making it the highest-leverage conversion touchpoint.

  • #924 3a8ec38 Thanks @IgorGanapolsky! - Unlock the full dashboard demo (no blur-wall paywall), point GSD-brief CTAs directly at /checkout/pro instead of the homepage 301 hop, and fix the sticky sidebar overflow so long right-rails scroll internally on GSD-brief pages.

  • #893 e699073 Thanks @IgorGanapolsky! - Add decision-trace module for full gate evaluation observability. Logs passes, blocks, and near-misses (constraints that almost matched). Includes session trace summaries showing safety posture at a glance — inspired by Ethan Mollick's observation that operators need to see agent thinking traces.

  • #910 b1c4c28 Thanks @IgorGanapolsky! - Replace the landing-page explainer video with a reproducible 90-second animated
    walkthrough that actually explains the mechanism — same-mistake-different-session
    pain, 👎 → Pre-Action Gate extraction, gate fires on the next bad call,
    compounding token savings, one-line install. Adds an offline render pipeline
    (scripts/render-demo-video/) that drives a scripted 1920×1080 HTML animation
    through headless Playwright and muxes an ElevenLabs/say narration track —
    byte-reproducible on every re-render, no live agent session required. New
    npm scripts: demo:narration, demo:render, demo:render:full.

  • #924 3a8ec38 Thanks @IgorGanapolsky! - Replace legacy "MCP Memory Gateway" green logo in docs/logo-400x400.png with the proper ThumbGate brand mark (cyan thumbs-up + wordmark on dark background). Also detached the stale image from the Stripe Product (prod_UE7SR5NFBkumEp) so checkout no longer shows the legacy asset. Fixes CEO-reported "weird MCP logo on Stripe annual checkout" bug.

  • #866 8a62372 Thanks @IgorGanapolsky! - Fix wire-hooks to clean stale project-level Claude Code hooks referencing missing files. Previously only cleaned user-level settings, leaving broken hooks in .claude/settings.json that caused "UserPromptSubmit hook error".

  • #902 94d3882 Thanks @IgorGanapolsky! - Fix: serve public static assets (/assets/*, /favicon.ico, /thumbgate-logo.png, /og.png, /apple-touch-icon.png) without requiring an API key. Before this change the landing page rendered but every image, video, and icon fell through to the /v1/* API-key guard and returned 401, leaving visitors with an empty video player and broken poster images. Adds path-traversal-safe asset routing with correct MIME types, Cache-Control: public, max-age=86400, immutable, and HEAD-request support. Covered by tests/public-static-assets.test.js.

  • #903 689a9bd Thanks @IgorGanapolsky! - Add gate-coherence analyzer to detect pseudo-unification across enforcement layers. Runs 20 probes across spec-gate and gate-config layers, detects contradictions (one blocks, another allows), coverage gaps (dangerous input passes all layers), and false positives. Reports coherence score and grade (unified/divergent/over-blocking). Inspired by entropy-probing research on pseudo-unification in multimodal models.

  • #898 bc67f55 Thanks @IgorGanapolsky! - Add gate-eval module for systematic evaluation of gate effectiveness. Operators define eval suites (expected block/pass outcomes), run them against specs, get precision/recall/F1 metrics, compare spec versions A/B, and track effectiveness trends over time. Ships with 16-case agent-safety eval suite. Inspired by Anthropic's prompt evaluation framework.

  • #941 fdcbb13 Thanks @IgorGanapolsky! - Fix header logo rendering as tiny iOS-launcher tile across all site surfaces. The existing /assets/brand/thumbgate-mark.svg is designed as an app-icon (full 512×512 canvas with a #0a0d12 rounded-square backdrop filling the entire viewBox). When inlined in headers at 28–32px next to the wordmark it read as "a dark tile with a microscopic icon inside" rather than as a clean brand mark. Adds a new transparent full-bleed companion /assets/brand/thumbgate-mark-inline.svg and repoints every header <img src=…> (landing, dashboard, lessons, pro, learn hub + 5 learn articles, post-checkout success page, SEO-GSD generator — 12 surfaces) to the inline variant. apple-touch-icon / PWA / OG link tags intentionally still reference the app-icon tile — that is the correct asset for iOS home-screen bookmarks. Adds a regression-guard in brand-assets.test.js that fails if the app-icon tile is ever re-inlined in a header, and an inline-mark transparency assertion that blocks reintroducing a full-canvas dark rectangle.

  • #931 8161e51 Thanks @IgorGanapolsky! - Rewrite the post-checkout "Hosted API setup" section on the Context Gateway Activated page with a plain-English value prop: what it is, when teams and CI users need it, when solo-laptop users can skip it, then the setup steps. Fixes the feedback that customers finish checkout and see jargon with no explanation of why the Hosted API matters.

  • #904 c5b5204 Thanks @IgorGanapolsky! - Add incremental dashboard review checkpoints so operators can mark the current state as reviewed and then see only new feedback, promoted lessons, and gate blocks that landed afterward. This ships the persisted review baseline, the dashboard checkpoint controls, and the /v1/dashboard/review-state API for reading and resetting the current checkpoint.

  • #943 7ac112c Thanks @IgorGanapolsky! - Replace the header inline logo and legacy favicon SVGs with the TG gate monogram so checkout, dashboard, and marketing headers use the same professional ThumbGate identity.

  • #879 5f3e1fc Thanks @IgorGanapolsky! - Fix Instagram publishing end-to-end. post-video.js now uses the Zernio presign upload flow + shared publishPost, matching the { url, key, size, contentType, type } media-item shape Instagram requires (legacy /media multipart + minimal { url, type } payload was silently rejected). Added instagram dispatcher to post-everywhere.js (previously a silent no-op). Added daily instagram-autopilot.yml workflow that posts a ThumbGate card via publish-instagram-thumbgate.js.

  • #945 2f8e670 Thanks @IgorGanapolsky! - Welcome email v2: consolidate the trial welcome email through the scripts/mailer/resend-mailer.js module and upgrade the template. Adds personalized greeting (first name from Stripe customer_details.name), explicit trial-end date (from Stripe subscription.trial_end), branded header mark, founder signoff, quickstart P.S., reply_to: [email protected], and a CAN-SPAM footer (business name, physical address, unsubscribe mailto) on every send. handleWebhook now threads customerName and trialEndAt through to the mailer. The legacy inline transport remains as a fallback and its no_api_key skip reason is normalized to missing_resend_api_key so dashboards and support tooling see a stable vocabulary regardless of which transport produced the skip.

  • #878 927e3ca Thanks @IgorGanapolsky! - feat: Claude-first landing page overhaul

    Restructures the entire landing page to prominently feature Claude plugin, Claude Extension, and Claude Code alongside (and above) the GPT promotion:

    • Hero section: rewrites subtitle from GPT-first to agent-agnostic, adds "Install Claude Extension" as a primary amber CTA button
    • New dedicated Claude Code section added before the ChatGPT GPT section
    • Compatibility grid reordered: Claude Desktop Extension first, Claude Code Skill second, ChatGPT demoted to last
    • First-Dollar Activation Path rewritten from GPT-centric to agent-agnostic install flow
    • Proof bar reordered with Claude links first
    • Final CTA adds Claude Extension button
    • Nav bar adds Claude link and Claude Extension CTA
    • GPT section renamed to "Also Available" to reduce GPT-first impression
  • #914 e6c6012 Thanks @IgorGanapolsky! - Landing page congruence fixes and dashboard deep-linking:

    • Remove misleading "1 agent" Free tier bullet (no per-agent enforcement exists in rate-limiter)
    • Rephrase Free tier bullets to match actual code behavior (1 auto-promoted prevention rule, built-in safety gates)
    • Add hash-based deep-linking to dashboard: /dashboard#insights, /dashboard#gates, /dashboard#export now auto-switch tabs
    • "Visual gate debugger" link on Pro tier now deep-links to #insights (was pointing to root /dashboard)
    • "DPO training data export" link on Pro tier now deep-links to #export
    • Add public/dashboard.html, scripts/prompt-eval.js, bench/prompt-eval-suite.json, CHANGELOG.md to npm files whitelist — these were missing, breaking the dashboard for users running npx thumbgate pro
    • New tests: 19 landing-page-claims (code-backed claim audit), 3 dashboard-deeplink-e2e (real server + HTTP fetch + hash validation)
  • #913 7dddb46 Thanks @IgorGanapolsky! - Add LinkedIn one-shot comment engagement: publishComment publisher
    (scripts/social-analytics/publishers/linkedin-comment.js) that posts a comment
    on a specified activity URN via the socialActions endpoint, plus a
    linkedin-comment-engage.yml workflow_dispatch that runs it with the
    LINKEDIN_ACCESS_TOKEN / LINKEDIN_PERSON_URN secrets. Used for
    high-signal targeted engagements on prospect / thought-leader posts
    whose audience overlaps ThumbGate's ICP; bulk / scheduled engagement
    still flows through Ralph Loop.

  • #924 3a8ec38 Thanks @IgorGanapolsky! - Add LinkedIn Post Dispatch workflow — first-party post publisher with optional article link-preview card. Fallback path when Comment API and Quote-Post reshare are blocked by LinkedIn's permission model.

  • #920 bb7a1f8 Thanks @IgorGanapolsky! - Add LinkedIn quote-post engagement pivot: linkedin-quote-post.js publisher + linkedin-quote-post-engage.yml workflow_dispatch. Publishes a standalone post on the authenticated member's feed with reshareContext.parent referencing the target activity URN, so we can engage with thought-leader posts when the Community Management API (socialActions/{urn}/comments) is not available on the app. Uses only w_member_social — already granted via the existing "Share on LinkedIn" product — no additional LinkedIn Developer Portal approvals required. The original author receives a mention-style notification through the reshare reference.

  • #886 f72d242 Thanks @IgorGanapolsky! - Marketing assets and README overhaul: conversion-optimized README with architecture diagrams, SEO tutorial article, Manus AI skill, and technical architecture diagrams (MCP flow, feedback pipeline, agent integration).

  • #863 2a048e2 Thanks @IgorGanapolsky! - Wire Stripe pricing calls to action into the marketing autopilot and scheduled X revenue loop.

  • #881 91e971d Thanks @IgorGanapolsky! - fix(monetization): enforce lifetime free-tier caps, reduce Team pricing to $49/seat

    • Rate limiter switched from daily resets to lifetime caps (3 captures, 1 rule, recall blocked)
    • Team plan reduced from $99 to $49/seat/month with new Stripe price ID
    • Landing page rewritten with pain-first copy, hard limits visible, updated CTAs
  • #921 a97ef8e Thanks @IgorGanapolsky! - Add pre-commit + pre-push git hooks to catch regressions before CI. Hooks live in .githooks/ (no new npm deps), auto-activate via prepare npm script, enforce: public/ HTML package parity, version sync, check-congruence, landing-page-claims, gates-engine regression tests, npm pack dry-run, internal link validation. Also adds CI publish-guard that fails when a merge leaves shipped content un-bumped (prevents the "1.5.2 already on npm, content didn't ship" silent no-op that forced 1.5.3/1.5.4).

  • #917 d33b81f Thanks @IgorGanapolsky! - Put the Pro pricing card INSIDE the homepage hero (between subtitle and dashboard preview) so $19/mo and $149/yr never get buried. The card shows both Monthly and Annual plans side-by-side with dedicated "Choose monthly / Choose annual" buttons and a "SAVE 35%" pill on annual — visible in pixel #1 on any viewport, not hidden behind scroll. /pro is now a permanent 301 redirect to /#pro-pitch (the id of the in-hero pricing card), so every README, plugin manifest, guide, and compare page link still works and passes link equity onto a single canonical landing page. /pro also removed from the sitemap entry list and from the JSON root-endpoint listing so search engines index / directly instead of chasing the redirect.

  • #896 cb1657f Thanks @IgorGanapolsky! - Add prompt-evaluation positioning to the README and landing page so ThumbGate explains that prompt engineering is only the start, and proof lanes plus self-heal checks are how behavior gets measured and enforced.

  • #929 29bb812 Thanks @IgorGanapolsky! - Make the landing-page proof-bar links individually clickable with padded hit targets and keyboard focus states, and show both thumbs-up reinforcement and thumbs-down correction examples in the first-dollar activation path.

  • #857 2f3fa15 Thanks @IgorGanapolsky! - Fix public landing page version synchronization so multiple release markers update in one pass.

  • #911 1d36bab Thanks @IgorGanapolsky! - Include public/dashboard.html, scripts/prompt-eval.js, and bench/prompt-eval-suite.json in the published npm package. The 1.5.1 release shipped without dashboard.html, breaking the local Pro dashboard for users who ran npx thumbgate pro. This patch restores the dashboard and ships the prompt evaluation framework.

  • #868 e42391d Thanks @IgorGanapolsky! - Harden revenue observability by preferring hosted billing-summary truth over local fallback when THUMBGATE_API_KEY is available, adding machine-readable Stripe live status diagnostics, and wiring the daily revenue loop to audit hosted revenue, Stripe, and Plausible checkout attribution with artifacts.

  • #855 69157d2 Thanks @IgorGanapolsky! - Add a local sales pipeline ledger for first-dollar workflow hardening outbound, and update GTM targeting so direct outreach leads with the Workflow Hardening Sprint before self-serve Pro follow-up.

  • #905 d3f7195 Thanks @IgorGanapolsky! - Add sandbox scope to spec-gate constraints for secure code execution environments. Adds 2 sandbox-specific constraints (no-sandbox-network, no-sandbox-fs-escape) to agent-safety spec. Also adds workflow-gate-checkpoint module for persisting gate state across long-running workflow restarts. Inspired by Vercel's Open Agents infrastructure.

  • #888 9fcc0a0 Thanks @IgorGanapolsky! - Add session health sensor and episodic session store for real-time and cross-session agent degradation detection. Tracks repeat errors, negative feedback density, stagnation, context amnesia, time-of-day risk, category risk, recurring errors, and feedback effectiveness trends.

  • #892 86152fa Thanks @IgorGanapolsky! - Add spec-gate module for proactive correctness enforcement. Operators define specs (constraints + invariants) upfront as JSON; gates enforce them from session start, not just from learned failures. Ships with agent-safety spec covering force-push, secrets, destructive ops, and test-before-commit invariants.

  • #939 adcc368 Thanks @IgorGanapolsky! - Fix broken logo on /success (Context Gateway Activated) page. After PR #932 moved brand assets to /assets/brand/, the HTML templates from PR #931 still referenced the legacy /brand/thumbgate-mark.svg path — which Railway's route guard now returns 401 for. Migrates all 15 customer-facing surfaces (landing, dashboard, lessons, pro, learn hub + 5 learn articles, post-checkout success page, SEO-GSD generator) to the correct /assets/brand/thumbgate-mark.svg path (serves 200). Also migrates favicon link from the 401ing /favicon.svg to the 200ing /thumbgate-icon.png, and og:image from /brand/thumbgate-og.svg to /og.png, with correct MIME types. Updates brand-assets test suite to pin the new paths so this can't regress.

  • #865 81dac4e Thanks @IgorGanapolsky! - Enforce ThumbGate-only launch, GPT Actions, analytics, and outreach surfaces so legacy repository names cannot leak into active product guidance.

  • #940 5a39d1c Thanks @IgorGanapolsky! - Polish the ThumbGate Pro trial email so checkout activation uses conversion-ready copy, a clear dashboard call to action, Pre-Action Gates positioning, and Resend sender configuration synced into Railway deploys.

  • #924 3a8ec38 Thanks @IgorGanapolsky! - Enforce per-platform character limits in the Zernio publisher before posting or scheduling. The previous path blasted identical content to every connected platform — a 315-char post silently failed at Bluesky's 300-char ceiling (CEO-reported post 69d939ba88955f0579e44fa7, 2026-04-16). New platform-limits.js module maps canonical limits (Bluesky 300, X/Twitter 280, LinkedIn 3000, etc.) and rejects over-limit targets with actionable { reason, platform, limit, length, overBy } detail rather than letting the provider eat the failure.

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]