Changelog

Notable changes to slashsynthex.com — the public marketing and documentation site. For Synthex plugin releases, see the upstream repository at

.

Versions follow SemVer with site-content-level granularity: major = visual/IA overhauls, minor = new pages or sections, patch = copy/style fixes.

#Next — Synthex 2.6.0: Cursor cloud agents

Tracks the upstream Synthex 2.6.0 release. On Cursor, /synthex:next-priority/synthex:next-priority can now run each parallel task as a cloud subagent with its own VM and branch, via --in-cloud or next_priority.execution_environment: cloud. Other hosts print a one-line notice and keep using local worktrees. Synthex also ships Cursor Task agent wrappers so its roles show up as named Cursor subagents. See the upstream plugin release at

.

#Added

• Cloud agents — new concept page for cloud execution of /synthex:next-priority/synthex:next-priority.
• Workspaces & config layers, Skills, and Hooks — new pages.
• Agents reference — a section listing the internal helper agents (multi-model orchestrator, consolidator, review adapters, standing-pool helpers) that commands invoke for you.

#Amended

• Configuration — now covers every key in the shipped defaults, including next_priority.execution_environment, the full multi_model_review and standing_pools blocks, lifecycle, and hooks.
• Getting started and Documentation — a "What's new" callout. Install steps now match the plugin README: the portable-skills/ bundle for Gemini CLI and OpenCode (with OpenCode's opencode.json), plus Hermes Agent and Cursor.
• Commands reference — the page description counts all 22 commands.

#Upstream fixes (no page changes)

• Parallel reviewers and multi-model proposers stay in the foreground on Cursor, so the consolidated verdict waits for every reviewer.
• A prompt-file reviewer's own report sections no longer leak into the Synthex report; the reviewer is told to return only the contracted fields.

#Acknowledgements

The plugin-side changes were authored and shipped by the upstream Synthex maintainers; this release covers only the documentation work to surface them on the public site.

#Next — Synthex 2.5.5

Tracks the upstream Synthex 2.5.5 release, a fix release. When the Product Manager runs as a sub-agent it can't ask you questions directly, because Claude Code doesn't give sub-agents the question tool. It now returns its questions, and

/synthex:write-implementation-plan/synthex:write-implementation-plan asks them for it and then resumes the Product Manager. Agent definitions also picked up newer sub-agent settings, such as turn limits on the review adapters. See the upstream plugin release at

.

No site pages changed.

#Next — Synthex 2.5.4

Tracks the upstream Synthex 2.5.4 release. It moved the repository's developer instructions from CLAUDE.md to AGENTS.md. Nothing changes for people using Synthex, so no site pages changed. See the upstream plugin release at

.

#Next — Synthex 2.5.3

Tracks the upstream Synthex 2.5.3 release, a fix release. The default plan and PRD reviewer rosters no longer name a designer agent that doesn't exist, the wizard behind

/synthex:configure-multi-model/synthex:configure-multi-model now actually turns multi-model review on, the Codex adapter's allow_api_key_billing setting now takes effect, and the commit-lint hook checks the git commit -F - heredocs that Synthex's own agents write. See the upstream plugin release at

.

#Amended

• Configuration — the plan reviewer roster now defaults to design-system-agent and the PRD roster to lead-frontend-engineer, with a note that a legacy agent: designer entry still works. The commit-conventions section covers git commit -F - messages written in the same command and the cases that still fail open.
• /synthex:write-implementation-plan/synthex:write-implementation-plan and /synthex:refine-requirements/synthex:refine-requirements reference pages — reviewer rosters updated to the real agents, with the legacy-alias note.
• /synthex:configure-multi-model/synthex:configure-multi-model reference page — Enable with detected CLIs now writes both per-command switches. The detection table lists the adapter each CLI maps to and the login check that adapter actually runs (Gemini API key, claude auth status, no check for llm).
• Multi-model review — "Turning it on" matches the wizard's new behavior. Billing shows where allow_api_key_billing goes and notes that each adapter now receives its whole per_reviewer block.
• Plugin architecture — the commit-lint.sh hook row mentions -F - messages.

#Acknowledgements

The plugin-side changes were authored and shipped by the upstream Synthex maintainers; this release covers only the documentation work to surface them on the public site.

#Next — Synthex 2.5.2

Tracks the upstream Synthex 2.5.2 release, a cleanup release. It removes leftover references to the retired synthex-plus plugin from Synthex's own command, agent, and config files. Nothing changes in how commands behave, so no site pages changed: the standing review pools pages already describe the current behavior. See the upstream plugin release at

.

#Upstream changes (no page changes)

• /synthex:start-review-team/synthex:start-review-team no longer runs a separate "Synthex plugin not detected" check. That check came from synthex-plus, which depended on Synthex being installed. The remaining preflight checks are the same as before.
• The plugin's bundled standing-pools guide now uses the same flag names as the command and as this site: --name, --multi_model, and --ttl_minutes. It also says that idle TTL is enforced lazily.
• The multi_model_review.per_command.team_review.enabled key in the shipped defaults is now marked unused. /team-review was retired along with synthex-plus, and nothing reads the key.
• Internal docs and test fixtures for the retired synthex-plus team commands were removed.

#Next — Catch-up for Synthex 1.0–2.2

Between Synthex 1.0.0 and 2.2.0, fourteen upstream releases shipped without site docs: the automated docs step was failing silently. This entry covers what those releases added, plus two gaps flagged since.

#Added

• Multi-model review — new concept page: Codex, Gemini, Claude, Ollama, llm, and Bedrock reviewers alongside the native ones, when the multi-model path runs, the multi_model_review config, and failure behavior. Plus a reference page for /synthex:configure-multi-model/synthex:configure-multi-model.
• Standing review pools — new concept page for warm reviewer pools (folded into Synthex from the removed synthex-plus plugin), plus reference pages for /synthex:start-review-team/synthex:start-review-team, /synthex:stop-review-team/synthex:stop-review-team, /synthex:list-teams/synthex:list-teams, and /synthex:configure-teams/synthex:configure-teams.
• Workflow engines — new concept page for the opt-in Claude Code engines behind /synthex:review-code/synthex:review-code and --loop verdicts.
• Command reference — pages for /synthex:refine-requirements/synthex:refine-requirements (previously a 404), /synthex:dismiss-upgrade-nudge/synthex:dismiss-upgrade-nudge, and /synthex:star/synthex:star.

#Amended

• Configuration — model profiles, the verification pass, workflow engines, commit conventions, .synthex/facts.md, and pointers to the multi-model and pool config.
• review-code — corrected parameters and verdict rules; an empty reviewer roster now fails; prompt-file reviewers fall back to their directory name.
• Native looping — the default ALLDONE<session_id> promise, idle iterations, compaction recovery, and the retired synthex-plus team example replaced.
• Plugin architecture — the hooks Synthex ships and the utility agents that became scripts.

#Next — Synthex 2.5.0

Tracking the upstream Synthex 2.5.0 release. /synthex:next-priority/synthex:next-priority's --auto-decide flag now actually skips questions that already have a recommendation. On Claude Code it arms a session-scoped hook that answers single-(Recommended) questions, and every decision is recorded so you can review it later. See the upstream plugin release at

.

#Amended

• next-priority reference page — the Auto-decide section now covers the auto-decide-gate PreToolUse hook (what it answers and what it always leaves to you, including [H] approval). It also covers how to audit decisions afterward: the .synthex/auto-decide/ log, Auto-Decision: commit trailers, and the bold Auto-decisions section at the top of the PR.

#Acknowledgements

The plugin-side change was authored and shipped by the upstream Synthex maintainers; this release covers only the documentation work to surface it on the public site.

#Next — Synthex 2.4.0: prompt-file reviewers

Tracks the upstream Synthex 2.4.0 release. Review-loop reviewers are no longer limited to Synthex agents: a reviewers entry can point at a markdown prompt file in the project, so an existing review prompt (a CI guardian, a house checklist) runs inside a Synthex review and its findings dedupe and rank with everyone else's. See the upstream plugin release at

.

#Added

• Configuration — new "Prompt-file reviewers" section covering the - prompt: <path> entry shape, the id / enabled / focus keys, frontmatter handling, the read-only convention, failure modes, and standing-pool behavior.

#Amended

• /synthex:review-code/synthex:review-code reference page — the Reviewer roster gained a "Prompt-file reviewers" subsection with an example and multi-model / standing-pool notes.
• /synthex:refine-requirements/synthex:refine-requirements and /synthex:write-implementation-plan/synthex:write-implementation-plan reference pages — the Reviewer roster now notes that an entry may be a prompt file.

#Acknowledgements

The plugin-side feature was authored and shipped by the upstream Synthex maintainers; this release covers only the documentation work to surface it on the public site.

#Next — Synthex 2.3.0

Covers upstream Synthex 2.3.0. Synthex config can now live in user-level and project-local files as well as the shared project file. The Codex adapter for multi-model review now works with the current Codex CLI. See the upstream plugin release at

.

#Added

• Configuration: user and project-local layers. Settings can now come from ~/.config/synthex/config.yaml (every repo, honors $XDG_CONFIG_HOME) and .synthex/config.local.yaml (one repo, gitignored), as well as the shared .synthex/config.yaml. Project settings beat user settings. Two narrow multi-model review key groups resolve differently: per_reviewer.* is user-owned, and enabled flags are restrictive-wins.

#Amended

• Getting started: the init step notes that .synthex/config.local.yaml is added to .gitignore.
• Reviewer pipeline: the multi-model section explains that disabling external review in any config layer wins.

#Upstream fixes (no page changes)

• Codex adapter matches the real Codex CLI. The multi-model Codex reviewer now uses codex login status for its auth check and codex exec --sandbox read-only for review runs. It drops flags that Codex CLI doesn't support.
• Codex adapter defaults reach codex exec. The shipped parent-mediated default for Codex now runs read-only and logs one warning. An empty or missing Codex result counts as a parse failure and triggers a retry. It is no longer treated as a clean review.
• Per-token API-key billing is opt-in for Codex. Unless the adapter's allow_api_key_billing setting is true, Synthex unsets CODEX_API_KEY and OPENAI_API_KEY for Codex calls and uses your stored codex login.

#Next — Grok install

The public install guides now include Grok beside the other supported harnesses. Grok uses the same LumenAI marketplace and the same generated Agent Skills as Codex. The canonical command and agent Markdown is unchanged.

#Added

• Agent compatibility and Getting started — native Grok install (grok plugin marketplace add, grok plugin install synthex --trust, grok plugin enable synthex) and /synthex:<command> invocation examples.
• FAQ, links, marketplace, plugin architecture, and llms.txt — Grok is listed with the native plugin installs. The links page points at the Grok skills and marketplace docs.

#Next — Synthex 1.0

Synthex is becoming Claude-first, now for any agent. Claude Code remains the original and primary plugin experience; version 1.0 also distributes the same workflows through Agent Skills for Codex CLI, Gemini CLI, OpenCode, and future compatible harnesses.

#Added

• Agent compatibility — install and invocation instructions for Claude Code, Codex CLI, Gemini CLI, and OpenCode, plus the boundary between portable workflows and Claude-native features.
• Shared-source architecture documentation — explains how canonical agents/ and commands/ Markdown drives generated SKILL.md entrypoints without copying prompts.

#Amended

• Site-wide product language — preserves Synthex's Claude Code roots while accurately describing its Agent Skills distribution and supported harnesses.
• Marketplace, FAQ, links, search, sitemap, and AI-readable documentation — now point to the multi-harness installation guide and supported agent ecosystem.

#0.2.1 — 2026-06-08

Native-looping reliability fix, tracked from upstream. Synthex 0.9.10 + synthex-plus 0.5.10 hardens how a loop binds to its Claude Code session so iteration keeps being driven to completion instead of silently stalling at the first turn-end. This release clarifies that session binding on the concept page. See the upstream plugin release at

.

#Amended

• Native looping concept page — the Multi-session section now notes that each loop is stamped with the id of the session that starts it (and re-stamped on resume). That binding is what lets the framework keep driving a loop turn after turn; the full state-schema and gate-ownership rules live in the upstream spec.

#Acknowledgements

The plugin-side fix was authored and shipped by the upstream Synthex maintainers; this release covers only the documentation work to surface it on the public site.

#0.2.0 — 2026-05-14

Native-looping coverage. The Synthex plugin shipped a native iteration primitive (synthex 0.8.0 + synthex-plus 0.4.0); this release documents and showcases it across the site. See the upstream plugin release at

.

#Added

• Concept page — /docs/concepts/native-looping. Canonical explainer for the iteration framework: state file schema, shared-context vs. isolation mode, multi-session semantics, promise emission, and coexistence with the official ralph-loop plugin.
• Command reference pages (3 new) — full overlays for /synthex:loop/synthex:loop, /synthex:list-loops/synthex:list-loops, and /synthex:cancel-loop/synthex:cancel-loop. Parameter tables, examples, refusal paths, and cross-references to the concept page.
• Command reference page (1 new) — /synthex:refine-requirements/synthex:refine-requirements finally has its own overlay (it previously shipped with only the structured fallback).
• Landing-page feature card — a "Native looping" section between Lifecycle and the Org Chart, with a terminal-style example and a CTA to the concept page. Static SSG; no client JS added.

#Amended

• /synthex:next-priority/synthex:next-priority, /synthex:review-code/synthex:review-code, and /synthex:write-implementation-plan/synthex:write-implementation-plan reference pages each gained a "Native looping" subsection with that command's emission rule and an example.
• Sidebar nav and sitemap include the new concept page and reference pages.

#Acknowledgements

The plugin-side feature was authored and shipped by the upstream Synthex maintainers; this release covers only the documentation work to surface it on the public site.

#0.1.0 — 2026-05-11

Initial public release. The site goes live at slashsynthex.com with the full code-side deliverable across the five planned phases.

#Pages

• Landing — marketing hero, lifecycle, 15-agent org chart with three tiers, command grid, quickstart, footer with Bluminal sponsor credit.
• Docs — three-pane shell with sidebar / right-TOC / scroll-spy. Group ordering: Get started → Concepts → Reference → Extending → Resources. Cmd+K palette across all routes.
• Concepts (7 pages) — Philosophy (excerpt + canonical link to AJ's essay), Lifecycle, Quality gates, Reviewer pipeline, Parallel execution, Marketplace, plus the standalone Configuration reference.
• Reference (7 pages) — includes an Agents directory with per-agent slug routes (15), a Commands directory with per-command slug routes (11), Documents, Specs, ADRs, RFCs, and Plugin architecture.
• Extending (2 pages) — Authoring agents, Authoring commands.
• Top-level — Examples (4 Coming-Soon slots), FAQ (6 Q&A), Links.

#Per-route OG cards

Every route ships with a tailored OG image generated at the edge via @vercel/og. Landing, docs-section, and reference-detail templates cover the full surface. All four dynamic OG handlers wrap their primary render in try/catch with a system-font fallback.

#Programmatic affordances

• <SlashCommand> — single typographic source of truth for the /synthex:<verb> mark across NavBar, OrgChart, Commands grid, MDX, Cmd+K palette, and OG templates.
• <AnchorButton> — left-margin # glyph on every H2/H3 with one-click deep-link copy and <Toast> feedback.
• <PageShare> — Copy URL / Copy as markdown / Share to X / Share to LinkedIn dropdown on every docs page.
• <Toast> — aria-live="polite" provider, FIFO stack of 3, opacity-only animation under prefers-reduced-motion.
• CodeBlock copy — every fenced code block gets a copy button; raw-code copy strips leading prompts ($, ▸, >).

#SEO + AI-search

• JSON-LD on every route: WebSite + Organization site-wide, plus SoftwareApplication, HowTo, TechArticle, BreadcrumbList, CollectionPage, FAQPage as appropriate.
• dateModified derived from git log -1 per source file (no rebuild-time noise).
• public/llms.txt short summary; public/llms-full.txt regenerated at build from all docs MDX.
• Sitemap with 47+ entries; robots.txt allows all crawlers, disallows /design/.
• Lighthouse: /docs and /docs/getting-started pass all four thresholds (Perf 95+, A11y 96, BP 95+, SEO 100).

#Quality gates (CI-enforced)

• ESLint with custom synthex/no-styled-jsx-in-loops rule (D13).
• TypeScript strict mode (noUncheckedIndexedAccess, exactOptionalPropertyTypes).
• Vitest 169/169 unit tests.
• Playwright e2e + a11y; npm run test:a11y passes 15/15 on the local build + 3/3 on the live deploy.
• size-limit budgets per D19 (HTML ≤25KB, JS ≤240KB, CSS ≤30KB — gzipped).
• Local structured-data audit validates every JSON-LD block against minimal Schema.org rules.
• Local internal-link audit asserts zero broken hrefs.
• llms-full.txt staleness check.

#Infrastructure

• Hosted on Vercel (project bluminal-labs/slashsynthex-web) with auto-deploy from main.
• Apex slashsynthex.com + www redirect; HSTS preload-eligible.
• Plugin data sync (data/agents.json + data/commands.json) refreshed from the upstream LumenAI repo at build, with committed-snapshot fallback for rate-limit / outage resilience.

#Acknowledgements

Sponsored and developed by Bluminal. The Ralph Loop driving the authoring iterations is itself a Claude Code plugin pattern.

#Upcoming

• Pagefind search (post-launch, replacing the current static Cmd+K index)
• Examples pages (the four Coming-Soon slots filled in)
• AJ's "Agentic Software Delivery" essay republished where appropriate

The implementation plan tracks per-task status at the source level; this changelog records material site changes as they ship.