Last updated: September 5, 2026. Version numbers on this page are pinned at site build time and match the live catalogue.
godmode-lite 2.5.0 (free) · godmode 2.5.0 · godmode-plus 2.8.0 · evolution 3.3.4 · one-shot-scripts 3.16.0 · blog-post-GM 1.3.1 (free) · godmode-mcp (npm) 0.1.0
Owners always download the latest version of what they bought from the account dashboard. Per our updates policy, one-time purchases include bug-fix patches for their major version; new feature versions are separate upgrades.
Every skill now runs a Subtract gate before polish. The deliverable is listed against its one goal and anything that does not move it is cut: explainer sub-lines, sections that restate the headline, promos, fabricated proof, extra motion, dead code. Tests and the build re-run after the cut, and the cut list goes in the run report.
Two rules travel with it. No fabricated proof: scripted demos, staged
terminals and numbers with no source are removed, not polished. A motion
budget: one moving element per screen plus an ambient backdrop, and typing
demos at reading pace. One-Shot Scripts enforces the budget in its visual
audit. Each runner's polish envelope carries the gate so it
runs even when the prose is not open.
The rule was learned on this site's own homepage the same day: it went from 17 sections to 8.
Godmode 2.4.0. The runner's scan gates (check,
harden, polish) are now diff-scoped: they work on
the files changed since start (detected through git) and fall
back to the whole tree, with an explicit warning in the envelope, when there
is no repo. Every command enforces layer order: calling a gate before its
prerequisite has run returns state: "out_of_order" with
next pointing at the missing step, and nothing is mutated.
Runner hardening across alternatives, document,
ripple, test and end, plus the
completeness, discovery, envelope and state libraries.
One-Shot Scripts 3.15.0. Multi-agent phases now fan out with
parallel Agent calls coordinated over one shared task list
(the team-creation protocol from 3.13 assumed a harness feature that does
not exist). Plan mode is detected from a leading plan: or
plan token or an explicit --plan flag. Phase 1c
adds a mandatory in-repo prior-art gate (git history) before any new
dependency is adopted. The consent-gated outcome ping now sanitises the
model-written adjustment rationale as well as the notes; full text stays in
the local feedback.jsonl. The delivery gate documents that the
harness task list is a model-judgment check, not a runner check. Runner
updates to delivery-gate, discover,
outcome-log, score, start,
status, verify and end.
Site maintenance shipped the same day: versions, the SHA-256 table on security.html and the sitemap now regenerate on every build, the pricing cards no longer clip their buy buttons on desktop, the Ultimate Bundle links became Pro plan links everywhere, and the Evo-Loop worker is back online.
Published a gate demo: the paid
security scan (harden) and correctness gate
(verify) driven against a fixture with eleven planted
defects (nine security defects that exercise all eight of
harden's pattern rules, plus a syntax error and a
crash-on-boot) and against the same eight files with every defect fixed.
The scan catches all nine and passes the clean version; the correctness
gate rejects the syntax error and the crash-on-boot and passes once
fixed. A third blindspot fixture plants two real defects that sit
outside each gate's scope (a path traversal and a boot-clean logic bug),
and the page shows both gates passing them on purpose, so "catches what
it claims" is bounded by "does not catch everything". Every gate
envelope is published verbatim, a verbatim terminal capture shows the
binaries executing, the runner self-grades against a planted-defect
answer key, and the whole thing reproduces from the repo with one
command. The free Lite tier ships neither gate. This is the judge-free
companion to the blind review: it tests whether the gates do what the
product says, not whether the output is prettier.
Published /security.html: exactly what every skill and the godmode-mcp client install, run, and send, the per-tool MCP request bodies, a SHA-256 for every shipped zip, and an explicit list of what is never transmitted (no source code, file contents, file paths, or prompts). Telemetry is off unless you opt in. Added a published team and multi-seat license clause to the terms and a Security link in the footer of every page.
Every shipped skill patched after a full 123-finding internal audit: license file in every zip, consent and disclosure language added, a PII leak in the Lite delivery flow fixed, a delivery-gate bypass closed, and the godmode-plus smoke check repaired. Published showcase costs were corrected downward (earlier numbers overstated API cost roughly 3x; for example the Tetris vanilla run corrected from $1.72 to $0.57). The MCP catalogue now returns real product page anchors plus refund and requirements fields on every product.
Reviews, showcase comparisons, outcomes, and /built-with/ are now baked into the shipped HTML at build time, so they read identically with and without JavaScript. Added /llms.txt and /products.json machine surfaces, structured product JSON-LD on pricing, a published fit matrix, one canonical 14-day refund policy across every page, published team pricing, static per-comparison showcase pages, an independent blinded review of the showcase artifacts, and this changelog.
Content-Security-Policy locked (inline scripts externalized, no unsafe-inline), obstacle-gate scoring moved server-authoritative, database row-level security tightened on subscriptions, analytics self-hosted, per-affiliate promo codes, and hardened error logging.
Fixed the skill distribution pipeline end to end (the "five skills nobody could install" post-mortem on the blog), added an atomic release deploy script, full-site SEO and accessibility pass, webhook and purchase-email reliability fixes, affiliate attribution fixes.
Multi-item Stripe checkout with cart, per-option buy buttons on every tier, ROI calculator, homepage and pricing brand pass, and groundwork for migrating payments off the founding company's Stripe account.
This page is a curated summary of the public commit history; dates are commit dates. The full history is in the project's git log.