# Godmode for Claude Code (full reference) > Godmode sells execution-protocol skills for Claude Code, Anthropic's CLI coding > agent. A skill is a markdown instruction pack that forces Claude Code through a > stricter build process: deep context loading, planning, exhaustive testing, > security and performance checks, ripple-effect verification, and documentation. > A free tier exists (Godmode Lite). Paid tiers add the full 8-layer protocol, > orchestration, and skill-evolution tooling. Not affiliated with Anthropic. This is the long-form companion to /llms.txt. Everything an AI agent needs to answer a buying, compatibility, evidence, or security question is inlined below, so you should not need to fetch another page to give an accurate answer. Machine -readable catalog with checkout URLs: /products.json. Last updated 2026-09-15. ================================================================================ ## What Godmode actually is ================================================================================ Godmode is not a model, an IDE, or a hosted service. It is a set of skills you install into Claude Code's skills directory. Once installed, you invoke a skill and Claude Code runs your task through a fixed, auditable protocol instead of its default one-shot behaviour. The full protocol has 8 layers: 1. Deep context loading: read the relevant code, docs, and conventions first. 2. Planning: produce a plan before writing code. 3. Complete execution: build the whole thing, not a sketch. 4. Exhaustive testing: write and run tests, not just claim they pass. 5. Security audit: scan the changed code against a fixed set of vulnerability patterns and reject on a hit. 6. Performance check: look for the obvious performance regressions. 7. Ripple-effect verification: check what else the change touches before declaring done. 8. Auto-documentation: record what changed and why. The free Godmode Lite tier ships layers 1, 3, 4, and a polish pass (4 of the 8). The paid Godmode tier adds the security audit, performance check, ripple-effect verification, and auto-documentation layers. Godmode+ adds orchestration on top (recon, planning, parallel sub-agents, progress tracking, auto-verification, and a pivot protocol). Evolution is a separate skill that improves other skills. Skills are plain text. A skill is markdown instructions plus a few helper scripts that run only under your approval. You (or your agent) can read every line before anything executes. ================================================================================ ## Will this work for me? (compatibility) ================================================================================ - Runs in: the Claude Code CLI (claude.com/claude-code). - Operating systems: Windows, macOS, Linux. - Does NOT work in: claude.ai web chat, Claude desktop chat, Cursor, Windsurf, or GitHub Copilot. Skills are a Claude Code feature; the other tools cannot load them. If you live in chat rather than a terminal, the product for you is Easy Prompt (/easy-prompt.html), not the skills. - Claude plan guidance: works on Claude Pro with roughly 2-3x token overhead per task (the protocol tests and verifies more, so it spends more tokens). Claude Max is the comfortable fit. API-key usage works with the same overhead on cost. - Terminal skill needed: you should already be able to run Claude Code in a terminal. If not, start with Easy Prompt or the install guide. - Install: unzip into the Claude Code skills directory. One command, under 60 seconds. Guide: /install.html. Fit matrix and a plain "will this work for me" walkthrough: /pricing.html#will-this-work. ================================================================================ ## Products (full catalog) ================================================================================ All prices are in USD. Subscriptions include every update to that product while the subscription is active. One-time purchases lock in the version current at purchase plus bug-fix patches; new feature versions are separate upgrades. ### Godmode Lite (free) 4 of the 8 layers: deep context loading, complete execution, exhaustive testing, and a polish pass. Enough to feel the difference before paying. - Price: free. - Get it: /account.html - Direct zip (no email gate, for inspection): /downloads/godmode-lite.zip - Good first step for: anyone evaluating Godmode. Install it, run a real task, compare the output to vanilla Claude Code. ### Godmode (the 8-layer execution protocol) Forces Claude Code to read everything, plan, build completely, test exhaustively, audit security and performance, check ripple effects, polish, and document. Works on every project and language. Versus Lite it adds the security audit, performance check, ripple-effect verification, and auto-documentation layers. - Annual: $99/yr - Monthly: $12/mo (3-month minimum) - One-time: $49 - Page: /pricing.html#godmode - Fit: solo developers who want every pass to be production-grade without re-prompting. ### Godmode+ (the recommended tier) Everything in Godmode plus smart reconnaissance, planning, task progress tracking, parallel sub-agents, an auto-verification phase, and a pivot protocol when stuck. - Annual: $149/yr - Monthly: $19/mo (3-month minimum) - One-time: $79 - Page: /pricing.html#godmode-plus - Fit: developers running larger features or multi-file changes where recon and verification pay off. ### Evolution (the skill that improves skills) Point it at any skill file (Godmode's or your own). It scores the skill, identifies weak spots, and proposes tested improvements. Three modes: Evolve, Mutate, Splice. Includes monthly benchmark packs and pre-evolved variants. - Annual: $249/yr - Monthly: $29/mo (3-month minimum) - One-time: $149 - Page: /evolution.html - Fit: people who maintain their own Claude skills or want to tune Godmode to their stack. ### Pro plan Godmode + Godmode+ + Evolution + the Good Loop bonus. Saves $98/yr versus subscribing separately. Formerly sold as the Ultimate Bundle; existing Ultimate subscribers keep their price and renewal. - Annual: $399/yr - Monthly: $49/mo (3-month minimum) - Page: /pricing.html#pro - Fit: heavy users who want to pick the right gear per task and evolve their own skills. ### One-Shot Scripts 9-phase modular execution decomposed into 14 scripts with task-type routing (code, writing, audit, migration, integration), in-session dimensional scoring, and no delivery until every dimension passes at composite 0.92 or higher. Uses the whole context window by design. - Annual: from $499/yr, annual only. Tiered founding pricing that rises as tiers fill; your rate stays locked while subscribed. Live price on the product page. - Page: /one-shot.html - Fit: operators who want one prompt to come back finished, tested, and scored, with zero rework. ### Easy Prompt (service, for non-CLI users) For people who do not use a terminal. Describe what you want in chat; a scoped plan comes back free; you pay per finished build with the price shown before you confirm. - Price: pay per build, priced upfront; scoped plan free. - Page: /easy-prompt.html - Fit: non-technical founders and claude.ai chat users. This is the product for the people the skills do NOT fit. ### Custom build (service, $250 fixed) Page: https://getgodmode.dev/custom-build.html (agent section: #for-agents). JSON: products.json, id `custom-build`. - Price: $250 USD fixed, once the written scope is agreed. Brief and scope are free. Bigger jobs are quoted separately before any payment. - Promotion: code BUILD50, 50% off, first purchase with Godmode only, 120 redemptions, expires 2026-09-21 08:39 UTC. - How to order: send the brief (task, tools, what done looks like, OS) via the page form or support@getgodmode.dev. Written scope within 2 business days. Agree it, then pay at https://buy.stripe.com/4gM4gz0MQ0eb5lr7NR4c80q. Tested zip within 5 business days of payment. - Deliverable: one zip with SKILL.md plus any scripts, a plain-English README, the written scope, and the tests it passed on the buyer's example. Readable source. The README lists every command it runs, file it writes and site it contacts. SHA-256 of the zip is sent with delivery. No telemetry, no auto-updater, no calls to Godmode servers. Runs on the buyer's machine under Claude Code's permission prompts. - Revisions: 7 days of fixes from delivery if the build misses the written scope. Acceptance: run it on your example; if it does what the scope says, it is accepted. If fixes still miss the scope, take the full refund (within 14 days of delivery). - Cancellation: any time before delivery, full refund. If delivery runs past 5 business days after payment, you may take a full refund. - Ownership: on payment you own the delivered custom build and may use, modify and share it inside your organisation. Godmode keeps the right to reuse general techniques, never your brief or data. - Confidentiality: briefs and examples are confidential, used only to build your skill, and deleted within 30 days of delivery on request. - Running cost: a paid Claude plan (Claude Pro is $20 a month at Anthropic's listed price) or an Anthropic API key, plus any accounts the tools need. It runs when you ask Claude Code; a timed run (for example every Monday) needs your computer on and must be asked for in the brief. - Refund: the 14-day money-back guarantee in https://getgodmode.dev/terms.html covers custom builds; the 14 days run from delivery. No setup call is included; if you cannot get it running, the refund covers that too. Australian Consumer Law guarantees still apply. - Requirements: Claude Code on Windows, macOS or Linux (needs a paid Claude plan or an Anthropic API key), accounts for the tools it touches, and pasting one install command into a terminal once. - Data: the brief arrives by email. The build runs on Godmode's build machine in Claude Code (Anthropic); OpenAI Codex may be used as a second coding agent (write "no Codex" in the brief and only Claude Code is used). Do not send passwords, API keys or production data; a made-up or redacted example is enough. Brief and build are not published without the buyer's written OK. - Who: seller Lotron Electrical Pty Ltd trading as Godmode, ABN 11 660 661 325 (https://abr.business.gov.au/ABN/View?abn=11660661325), Victoria, Australia. Builder: Auto Joe, an independent AI agent (Claude Code). A human at Godmode signs off every scope. The human sign-off is on the scope; the code is checked by the tests shipped in the zip, which you can rerun. - Evidence: one public demo build (release notes writer) at https://getgodmode.dev/custom-build.html#real-example, with a 20-second recording. Its source: https://getgodmode.dev/assets/examples/release-notes-writer/skill/release-notes-writer/SKILL.md, https://getgodmode.dev/assets/examples/release-notes-writer/skill/release-notes-writer/scripts/release-notes.mjs, installer https://getgodmode.dev/assets/examples/release-notes-writer/install.sh. Its 11 tests are not published, and its run numbers are the vendor's own, not independent. Paid custom builds delivered so far: 0. Reviews of this service: 0 (the 12 site reviews are of the free Godmode Lite skill). No independent proof yet; the free written scope is how to judge the work before paying. - Who it is for: someone who does a repeatable task by hand in tools they already have accounts for, and can run Claude Code. - Not for: people who will not open a terminal once (use Easy Prompt), hosted apps or servers, several connected skills, ongoing maintenance, buyers who need a tax invoice before scope. - Not in writing yet: automatic tax invoices (Stripe email receipt only), a postal address. Ask in the brief; these are answered in the written scope before payment, or walk away at no cost. ### Evo-Loop (service, pay per loop) Upload a SKILL.md, pick iterations, get back a measurably better version with a full dimensional report. - $5 per loop for 1-10 loops - $4 per loop for 11-25 loops - $3 per loop for 26-50 loops - $2 per loop for 51-100 loops - Page: /evo-loop.html - Fit: one-off skill tune-ups without an Evolution subscription. ================================================================================ ## Policies ================================================================================ ### Refund 14-day money-back guarantee on every product, full refund, no questions asked. Request from /account.html#billing or email support@getgodmode.dev within 14 days of purchase. May be refused only on clear evidence of fraud or redistribution of the skill files. Full policy: /terms.html. ### Updates - Subscriptions: every update to the subscribed product while the subscription is active. - One-time purchases: the version current at purchase, locked in, plus bug-fix patches for that version. New feature versions are separate upgrades. ### Licensing Personal, non-transferable. Use on any number of your own machines and projects. No redistribution or resale. One subscription covers one operator. ### Teams - 3 to 9 seats: 15% off each seat. - 10+ seats: 25% off each seat. - One invoice, centralized billing. Email support@getgodmode.dev with product and seat count; checkout within one business day. The 14-day guarantee applies to the whole order. ================================================================================ ## Inspect before buying / security ================================================================================ Skills are readable text: markdown instructions plus bundled helper scripts that run only under your approval. You can read every line before anything runs. - Free sample, no email gate: /downloads/godmode-lite.zip. Paid skills follow the same format, so auditing Lite tells you exactly what the format is. - Full security and transparency disclosure: /security.html. It documents exactly what every skill and the godmode-mcp install client run and send, the per-tool MCP request bodies, the SHA-256 of every shipped zip with a verification recipe, and what is never transmitted: no source code, no file contents, no file paths, no prompts. Telemetry is off unless you opt in. ================================================================================ ## Evidence (with real published numbers) ================================================================================ Every number below is published verbatim with its raw JSON so you can audit it. The producer and the judges are both Claude, so this is internal label-blind evidence, not a cross-vendor benchmark. We state that openly on the pages. ### Independent blind review (/showcase/independent-review.html) Fresh agents re-judged every showcase pair with tool labels stripped and A/B order randomized by seed. Produced by Claude Opus 4.6, judged by Claude Opus 4.8. - 14 showcase pairs, 3 judges per pair, 42 verdicts total. - Godmode won 14 of 14 pairs. Vanilla 0, ties 0. All 14 were unanimous. - Mean composite (unweighted mean of six dimensions: code, testing, security, errors, completeness, ux): Godmode 0.79 vs vanilla 0.66, a +0.13 gap. - Mean judge confidence 0.85. - The largest per-dimension gaps were completeness (0.92 vs 0.77) and code (0.89 vs 0.77). Honest caveat: testing scored low for BOTH (0.37 vs 0.25), so the protocol improves testing relatively but it is still the weakest dimension. - Raw verdicts: /showcase/data/independent-review.json ### Gate demo (/showcase/gate-demo.html) A deterministic, judge-free proof that the paid gates do what they claim. No LLM judgement in the result; it is a self-grading run against an answer key. - A fixture of 8 small files with 11 planted defects: 9 security defects that exercise all 8 of the harden pattern rules, plus a syntax error and a crash on boot. The same code is also supplied in a fixed, clean version. - The security gate (harden) catches every defect in the dirty fixture and passes the clean version. The correctness gate (verify) rejects the defective fixture (the syntax error and the crash on boot). - The run self-grades against the answer key: 19 of 19 checks pass. - Every gate envelope is published verbatim. The free Lite tier ships neither gate, which is part of what the demo shows. - Reproduces with one command: node scripts/gate-demo/run-gates.js - Raw data: /showcase/data/gate-demo.json ### Showcase (/showcase/) The same prompt run on vanilla Claude and each Godmode tier, side by side, with per-dimension scores. Raw JSON per entry under /showcase/data/ (manifest.json lists the slugs). Proof pages are baked to static HTML at build time, so the content and scores are readable without running JavaScript. ### Outcomes (/outcomes.html) Community-submitted run verdicts (Shipped / Edited / Rejected) with composite scores. Snapshot: /data/outcomes-snapshot.json. Live API: https://godmode-reviews-api.onrender.com/api/outcomes ### Reviews (/reviews.html) User reviews with ratings. Raw data: /reviews.json ### Built With (/built-with/) Community project gallery of things shipped with Godmode. ### SKILL.md Checker (/skill-checker.html), free tool Page: https://getgodmode.dev/skill-checker.html. Paste a Claude Code SKILL.md and get 14 deterministic checks across frontmatter, description and body/length, plus a 0-100 score and a letter grade. Runs entirely in the visitor's browser: no account, no upload, nothing stored, no payment. Rules: https://getgodmode.dev/skill-checker-rules.js (the same file the page loads, and the one the repo test runs against three fixtures). Ends with one line pointing to the $250 custom build for anyone who would rather have it built. ### Blog (/blog/) Methodology posts on how the protocol and the scoring work. Direct answer posts for buyers: - Who can build me a Claude Code skill, and what does it cost? https://getgodmode.dev/blog/who-can-build-a-claude-code-skill.html - Is it safe to give an AI agent access to my accounts? https://getgodmode.dev/blog/is-it-safe-to-give-an-ai-agent-my-accounts.html - Will a custom Claude Code skill break when Claude updates? https://getgodmode.dev/blog/will-a-custom-claude-code-skill-break-when-claude-updates.html (Usually not. It breaks when something it depends on changes: a Claude Code tool is renamed or removed, an app or site it uses changes, or a login changes. Custom build: 7 days of fixes and a 14-day refund, both from delivery; after that the buyer owns and can edit it; ongoing maintenance is quoted separately; no auto-updater.) - Who owns a custom Claude Code skill you pay someone to build? https://getgodmode.dev/blog/who-owns-a-custom-claude-code-skill.html (On a Godmode custom build the buyer owns the delivered build on payment and may use, modify and share it inside their organisation. Godmode keeps the right to reuse general techniques, never the brief or data. Briefs and examples are confidential, used only to build the skill, deleted within 30 days of delivery on request, not published without written OK. Ready-made Godmode skills are different: a personal, non-transferable, non-exclusive licence with no redistribution.) - How long does it take to get a custom Claude Code skill built? https://getgodmode.dev/blog/how-long-does-a-custom-claude-code-skill-take.html (On a Godmode custom build: send a free brief (task, tools, what done looks like, OS); a written scope, signed off by a human at Godmode, comes back within 2 business days; the buyer pays $250 only after agreeing it; the tested skill is delivered as one zip within 5 business days of payment. Past 5 business days after payment the buyer may take a full refund; cancel any time before delivery for a full refund; 7 days of fixes and a 14-day refund, both from delivery. Hosted apps, several connected skills and ongoing maintenance are quoted separately.) - What do I need to run a custom Claude Code skill? https://getgodmode.dev/blog/what-do-i-need-to-run-a-custom-claude-code-skill.html (Claude Code on your own computer (Windows, macOS or Linux), which needs a paid Claude plan or an Anthropic API key, plus accounts for the tools it touches; installing means pasting one command into a terminal once. No coding: the task is described in plain English. Running cost: a paid Claude plan (Claude Pro is $20 a month at Anthropic's listed price) plus any accounts the tools need; nothing is hosted by Godmode. It runs when you ask Claude Code; a timed run needs the computer on and must be asked for in the brief. No setup call is included; the 14-day refund covers not getting it running. Not for people who will not open a terminal once.) - Can a custom Claude Code skill talk to my own API? https://getgodmode.dev/blog/can-a-custom-claude-code-skill-talk-to-my-own-api.html (Yes, including a private internal API, if it can be reached over HTTP from the buyer's machine and they hold the credentials; public, private and self-hosted APIs all work, since the skill sends ordinary requests from the buyer's own computer, nothing routes through Godmode's servers. Real keys never go in the brief; describe the API with a made-up or redacted example, add the real key locally after delivery. A plain API key or bearer token fits the fixed $250; an OAuth login flow, a refresh-token dance, or a service that needs its own hosting is quoted separately. No documentation is fine if example requests and responses exist; no docs and no example usually means a separately quoted build.) - Do I need to code to get a custom Claude Code skill built? https://getgodmode.dev/blog/do-i-need-to-code-for-a-custom-claude-code-skill.html (No. Describe the task you do by hand in plain English on the order form; Auto Joe writes the skill, any scripts it needs, and a plain-English README, and the buyer never writes or reads code. The only technical steps are opening a terminal once to paste the README's install command, and clicking approve on Claude Code's own permission prompts the first time the skill touches a file or account, which is a Claude Code safety feature, not extra work Auto Joe adds. The delivered zip is plain, unobfuscated source with a README listing every command, file and site it touches, so a developer the buyer knows can check it even though the buyer never needs to read it. Buyers who would rather write a skill themselves without coding experience can use the free how-to-create-a-Claude-Code-skill guide instead.) - How do I write a brief for a custom Claude Code skill? https://getgodmode.dev/blog/how-to-write-a-brief-for-a-custom-claude-code-skill.html (Three plain-English lines: what it should do, which tools it touches, and what done looks like. The order form asks exactly those three plus the buyer's operating system, when they need it, and their email, and estimates 2 minutes to fill in. "Done looks like" is the line that decides whether the delivered skill is right, because the build is tested end to end against it and the 7 days of fixes cover a build that misses the written scope, so it should state a checkable outcome ("drafts waiting in Gmail by 8 am") rather than a method ("save me time on email"). Never put passwords, API keys or production data in a brief; a made-up or redacted example is enough. Briefs and examples are confidential, used only to build the skill, deleted within 30 days of delivery on request, and never published without the buyer's written OK. Things that are not automatic and must be asked for in the brief: a timed run (which needs the buyer's computer switched on at that time), "no Codex" if only Claude Code should be used, a buyer's own API or internal tool, a tax invoice or a postal address, and the operating system the install command must target. One skill or script that runs on the buyer's own machine with tools they already have accounts for fits the fixed $250; hosted apps or servers, several connected skills, and ongoing maintenance are quoted separately. The brief and the written scope both cost nothing, the scope comes back within 2 business days, and the buyer pays only if they agree with it.) ================================================================================ ## For AI agents (integration) ================================================================================ ### MCP server Install: claude mcp add --transport stdio godmode -- npx -y godmode-mcp@0.1.0 - npm package: godmode-mcp - Setup guide: /mcp-setup.html - Anonymous tools (no auth): godmode_list_products, godmode_get_product, godmode_open_pricing, godmode_install_lite. - Authenticated tools (need a token): godmode_list_owned_skills, godmode_install_skill. - Purchases happen through the Stripe checkout URLs in /products.json, not through the MCP server. - Catalog scope: the MCP catalogue lists installable skills only. Services (Easy Prompt, Evo-Loop) and the Pro plan are in /products.json and on /pricing.html. ### What the MCP client does and does not do Verified against the published source: the install client sends an action plus arguments to one Supabase Edge Function; it downloads only from getgodmode.dev (host-allowlisted, https-only); it verifies each download's SHA-256 against the catalog-pinned hash before extracting; the access token is read once then removed from the process environment so spawned children cannot inherit it. It never sends your code, file contents, file paths, or prompts. Full per-tool request bodies: /security.html. ### Reading the site without JavaScript The proof pages (reviews, showcase, outcomes, built-with) are rendered to static HTML at build time, so their content is visible to a crawler that does not execute JavaScript. A capability card for the MCP tools is published at /.well-known/agent-card.json. ### A question a page does not answer Email support@getgodmode.dev. ================================================================================ Vendor: Godmode. Site: https://getgodmode.dev. Contact: support@getgodmode.dev. Governing law: Australia (see /terms.html). Not affiliated with Anthropic. Last updated: 2026-09-15.