{"id":"24879b83-01bb-47c6-b245-af830705a38e","slug":"arbazkhan971-godmode","kind":"skill","name":"godmode","summary":"Autonomous AI coding agent — 126 skills, 7 subagents, 5 platforms. Iterative optimization with automatic rollback, failure memory, and parallel multi-agent execution. Plugin for Claude Code, Cursor, C","owner":"arbazkhan971","repoUrl":"https://github.com/arbazkhan971/godmode","license":"MIT","licenseTier":"permissive","trust":"community","stars":26,"match":null,"fewStars":false,"format":"skill","createdAt":"2026-03-19T11:05:41.000Z","addedAt":"2026-10-14T06:31:11.449Z","logoUrl":"https://avatars.githubusercontent.com/u/31075248?u=aabf8ac68f6dedd12552277009ecd80aecfa4be3&v=4","linkOnly":false,"collection":null,"skillPath":null,"categories":["developer-tools","memory-knowledge"],"tags":["ai-agents","autonomous-coding","autoresearch","claude-code","code-review","cursor","development-tools","gemini-cli"],"runtime":"binary","local":true,"cloudCapable":false,"cloudCapability":"unknown","cloudReadyOnPlatform":false,"needsApiKey":false,"reviewed":false,"syncedAt":"2026-10-17T07:18:18.928Z","sample":false,"url":"https://conduits.dev/tools/arbazkhan971-godmode","language":"Shell","transports":["stdio"],"install":{},"requirements":{"env":[],"upstreamMayCharge":false},"tools":null,"readme":"
\n\n# Godmode — Autonomous AI Coding Agent for Claude Code, Cursor, Codex & Gemini\n\n### Your AI writes code. Godmode makes it write *great* code — then proves it.\n\nGodmode turns any AI coding assistant into an autonomous engineering loop — 135 skills that measure every change, keep what improves, and revert what doesn't — installable on Claude Code, Codex, Cursor, Gemini CLI, OpenCode, pi, omp, and Amp.\n\n[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)\n[![Skills](https://img.shields.io/badge/skills-135-ff6b6b.svg)](skills/)\n[![Agents](https://img.shields.io/badge/subagents-7-ff9f43.svg)](agents/)\n[![Claude Code](https://img.shields.io/badge/Claude_Code-supported-4A90D9.svg)](adapters/)\n[![Codex](https://img.shields.io/badge/Codex-supported-10A37F.svg)](adapters/codex/)\n[![Gemini CLI](https://img.shields.io/badge/Gemini_CLI-supported-4285F4.svg)](adapters/gemini/)\n[![Cursor](https://img.shields.io/badge/Cursor-supported-000000.svg)](adapters/cursor/)\n[![OpenCode](https://img.shields.io/badge/OpenCode-supported-7C3AED.svg)](adapters/opencode/)\n[![pi](https://img.shields.io/badge/pi-supported-20B2AA.svg)](adapters/pi/)\n[![Amp](https://img.shields.io/badge/Amp-supported-2257D7.svg)](adapters/amp/)\n[![Benchmark](https://img.shields.io/badge/benchmark-30_tasks-9B59B6.svg)](docs/blog/godmode-vs-plain-agent-benchmark.md)\n\n

\"Terminal

\nReal capture, live pi session — a win kept, a regression caught and reverted. More demos ↓\n\n
\n\n> **TL;DR:** Install with `claude plugin install godmode`. Say `/godmode optimize my API` and walk away. It measures, iterates, keeps improvements, reverts failures, and stops when done.\n\n---\n\n## See It In Action\n\nReal captures from live pi sessions (zai/glm-5.3, recorded with [vhs](https://github.com/charmbracelet/vhs)) — tapes and targets in [demo/](demo/).\n\n### 1. Skill routing — request to matched skill\n\n*godmode Tier-1 matching routes a checkout rate-limiting request to the ratelimit skill.*\n\n

\"Terminal

\n\n
Transcript (accessible text fallback)\n\n```text\nReal session excerpt — demo/tapes/skill-routing.tape (recorded 2026-08-27, pi 0.84.3, zai/glm-5.3)\n\n$ timeout 240 pi -p -ne --model zai/glm-5.3 \\\n --skill ~/.pi/agent/skills/godmode/godmode/SKILL.md \"Route only. The user request is: \\\n we keep getting rate limited by the payment provider during checkout. [...]\"\n\nTier 1 match complete.\n\n- **Stack detected:** Godmode skill-plugin repo (skills/ + agents/, no runtime app code in scope)\n- **Tier 1 hits:** `ratelimit` (\"rate limit\", \"throttle\", 429 — direct match); `pay` (payments context, secondary)\n- **Phase:** THINK — rate limiting against a third-party provider is a design problem (client-side throttle/queue/backoff) before any build\n\n`DISPATCH: skill=ratelimit phase=THINK (secondary=pay) task=rate-limited-by-payment-provider-at-checkout`\n\nROUTED_OK\n\n$ git status --short\n(no output — the routing session modified nothing)\n```\n\n
\n\n### 2. Optimize loop — real keep, real revert\n\n*baseline 189ms, lru_cache KEEP at 0ms, fib(32) regression DISCARD at 517ms, reverted with git reset — real metric, real revert.*\n\n

\"Terminal

\n\n
Transcript (accessible text fallback)\n\n\n```text\nReal session excerpt — demo/tapes/optimize-loop.tape (recorded 2026-08-27, pi 0.84.3, zai/glm-5.3)\n\n$ python3 -m unittest -q test_slow_fib\nRan 1 test in 0.230s\nOK\n\n$ bash metric.sh && bash metric.sh && bash metric.sh # timing runs — the session derives its own baseline below\n278\n217\n211\n\n$ timeout 300 pi -p -ne --model zai/glm-5.3 \\\n --skill ~/.pi/agent/skills/godmode/optimize/SKILL.md \"Setup answers: Goal = reduce the \\\n runtime of work() in slow_fib.py. [...] Test exactly these two hypotheses in order: \\\n (1) add functools lru_cache memoization to fib; (2) remove the lru_cache decorator and \\\n change work() to return fib(32), claiming batch amortization. [...] A DISCARD must be \\\n reverted with git reset --hard HEAD~1. [...]\"\n\n**Session complete** — baseline 189 ms → best 0 ms (−100%).\n\n- **Baseline:** median 189 ms (3 runs, variance <5%)\n- **R1 · H1 KEEP:** @lru_cache(maxsize=None) on fib → 0 ms (0,0,0), guard OK (work()=832040)\n- **R2 · H2 DISCARD:** removed cache, work()=fib(32) → 517 ms AND guard failed (2178309 ≠ 832040); reverted via git reset --hard HEAD~1\n- Stop: budget exhausted (2 rounds); kept commit 0c162a0; lessons + failures logged in .godmode/*\n\n$ cat .godmode/optimize-results.tsv\nround\tagent\tchange\tmetric_before\tmetric_after\tdelta_pct\tstatus\tfailure_class\n0\tbaseline\tinitial measurement\t-\t189\t-\tbaseline\t-\n1\t1\tadd lru_cache(maxsize=None) to fib\t189\t0\t-100.0\tKEEP\t-\n2\t1\tremove lru_cache, work()=fib(32) batch amortization\t0\t517\t+51700.0\tDISCARD\tguard_fail+metric_regression\n\n$ git log --oneline\nd26b093 (HEAD -> master) round2 discard logged; session complete\n7b6bc53 round1 keep: 189ms -> 0ms\n0c162a0 round1: add functools.lru_cache memoization to fib\n230bf5e iteration 0: baseline metric 189ms\n2e2ab07 baseline\n\nOPT_DONE\n```\n\n\n
\n\n### 3. Goal-bridge — machine-checkable completion contract\n\n*a failing test goes red to green under a 4-field contract (metric / threshold / evidence / rollback) with exit-0 proof.*\n\n

\"Terminal

\n\n
Transcript (accessible text fallback)\n\n```text\nReal session excerpt — demo/tapes/goal-bridge.tape (recorded 2026-08-27, pi 0.84.3, zai/glm-5.3)\n\n$ python3 -m unittest -q test_counter # baseline: real failure (red)\nFAILED (failures=2) — AssertionError: 1 != 0 (total([])), AssertionError: 7 != 6 (total([1, 2, 3])) — off-by-one (result = 1)\n\n$ timeout 240 pi -p -ne --model zai/glm-5.3 \\\n --skill ~/.pi/agent/skills/godmode/goal-bridge/SKILL.md \"Goal: make the test suite pass. \\\n Derive the Goal-Bridge contract FIRST and print its exact 4-field block: metric, \\\n threshold, evidence, rollback. [...] Then fix counter.py, run the metric each round \\\n appending to the evidence file, and stop as soon as it exits 0. [...]\"\n\nMetric exited 0 → **DONE** (round 2/3, rollback never fired).\n\nGoal-Bridge Contract (mandatory final output):\n- metric: python3 -m unittest -q test_counter\n- threshold: exit 0\n- evidence: .godmode/goal-evidence.txt\n- rollback: metric fails 2 consecutive rounds\n\nSummary:\n- Round 1: metric exit=1 (2 failures) → appended to evidence; fixed `counter.py` off-by-one (`result = 1` → `result = 0`)\n- Round 2: metric exit=0 (`OK`, 2 tests) → appended to evidence; stopped\n- Rollback trigger: never fired (max 1 consecutive failure); no revert needed\n- Files written: `counter.py` (edit), `.godmode/goal-evidence.txt` (new)\n\n$ cat .godmode/goal-evidence.txt\n=== Round 1: 2026-08-27T21:40:11Z ===\nexit=1\n=== Round 2: 2026-08-27T21:40:15Z ===\nexit=0\n\n$ python3 -m unittest -q test_counter && echo \"CONTRACT MET — exit 0\"\nCONTRACT MET — exit 0\n```\n\n
\n\nThe three captures above are real sessions; the blocks below are illustrative output shapes of other skills — not captured transcripts.\n\n
More output shapes (illustrative, not captured runs)\n\n### Performance Optimization — 847ms to 198ms, fully autonomous\n\n```text\n$ /godmode:optimize\nGoal: Reduce API response time\nIterations: 20\n\n BASELINE 847ms\n ROUND 1 554ms KEPT (-34.5%) -- added index on category_id\n ROUND 2 382ms KEPT (-31.0%) -- enabled gzip compression\n ROUND 3 276ms KEPT (-27.7%) -- eager loading for posts\n ROUND 4 290ms REVERTED -- batch loader (guard failed)\n ROUND 5 226ms KEPT (-18.2%) -- connection pool to 20\n ROUND 6 198ms KEPT (-12.4%) -- Redis response cache\n\n === 847ms --> 198ms (76.6% improvement) ===\n Keeps: 5 | Discards: 1\n```\n\n### Multi-Agent Build — 4 agents, isolated worktrees, clean merge\n\n```text\n$ /godmode:build\nGoal: Add user authentication system\n\n PLAN Decomposed into 4 parallel tasks\n AGENT 1 [worktree] Auth middleware + JWT tokens DONE\n AGENT 2 [worktree] User model + password hashing DONE\n AGENT 3 [worktree] Login/register API endpoints DONE\n AGENT 4 [worktree] Integration tests for auth flow DONE\n MERGE Sequential merge + test after each ALL PASS\n REVIEW 4-agent code review (security, perf, style) APPROVED\n\n === 4 tasks | 4 agents | 12 files | 47 tests passing ===\n```\n\n### Security Audit — STRIDE + OWASP + red team\n\n```text\n$ /godmode:secure\nTarget: src/api/\n\n RECON Mapped 23 endpoints, 4 auth flows, 2 data stores\n STRIDE 6 threat categories analyzed\n OWASP Top 10 checklist applied\n RED TEAM 4 personas: script kiddie, insider, APT, researcher\n\n CRITICAL 1 SQL injection in /api/search (parameterize query)\n HIGH 2 Missing rate limit on /api/login, weak CORS policy\n MEDIUM 3 Verbose error messages, missing CSP header, session fixation\n LOW 1 Server version disclosed in headers\n\n 7 findings with fix code + verification commands\n```\n\nEvery finding comes with code evidence, severity, a concrete fix, and a command to verify the fix works.\n\n
\n\n---\n\n## Quick Start\n\n```bash\n# Install (Claude Code)\nclaude plugin install godmode\n\n# Or install for other platforms\nbash adapters/codex/install.sh\nbash adapters/cursor/install.sh\nbash adapters/gemini/install.sh\nbash adapters/opencode/install.sh\nbash adapters/pi/install.sh\n```\n\n```bash\n# Use specific skills\n/godmode:optimize # Autonomous performance iteration\n/godmode:build # Build with parallel agents\n/godmode:secure # Security audit\n/godmode:ship # Pre-flight + deploy + verify\n\n# Or just describe what you want\n/godmode make this API faster # --> routes to optimize\n/godmode fix the failing tests # --> routes to fix\n/godmode build a rate limiter # --> routes to think --> plan --> build\n```\n\nGodmode auto-detects what you need and routes to the right skill.\n\n---\n\n## What Is Godmode?\n\nGodmode is an open-source plugin that adds autonomous coding capabilities to AI assistants like Claude Code, Cursor, Codex, Gemini CLI, and OpenCode. Instead of generating code once and hoping it works, Godmode runs a disciplined engineering loop: **measure, modify, verify, keep or revert, repeat** — until the goal is met.\n\nIt ships with 135 expert skills across 13 domains (performance optimization, security auditing, TDD, deployment, database tuning, and more), 7 specialized subagents that work in parallel, a failure memory system that learns from every discarded change, and a built-in authoring-discipline prelude that prevents the most common LLM coding mistakes before they hit the repo.\n\n## How Godmode Compares to Plain AI Coding\n\n| Problem | Godmode's answer |\n|---|---|\n| AI generates code once and stops | Autonomous iteration loops that run until the goal is met |\n| No way to know if a change helped | Every change is benchmarked against a mechanical metric |\n| Bad changes stay in the codebase | Automatic git revert on any regression |\n| Complex tasks bottleneck on one agent | Up to 5 parallel agents in isolated git worktrees |\n| AI repeats the same mistakes | Failure memory classifies and logs every failed approach |\n| Crashes lose all progress | Session state persists and auto-resumes |\n| Improvements might be noise | Variance testing and generalization gates prevent overfitting |\n| AI silently picks one interpretation | Karpathy prelude: state assumptions, surface alternatives, emit `NEEDS_CONTEXT` on ambiguity |\n| AI adds \"while we're here\" refactors | Line-trace rule + pre-commit discard audit surgically drops drift hunks |\n| Long autonomous loops burn tokens | Terse mode, stdio patterns, Progressive Disclosure routing — ~90% routing context reduction, 40-60% emit reduction |\n| Agents silently guess missing dispatch fields | DispatchContext schema validation — missing field → `BLOCKED: invalid_dispatch` |\n\n---\n\n## Benchmark\n\nThe repo benchmarked itself against a plain agent on 30 small tasks — 2 runs each, same model (`zai/glm-5.3`) in both arms, identical prompt. Godmode passed 60/60 scored runs; the plain agent passed 58/60. The plain agent was faster on median wall-clock (37.0s vs 41.0s), and the verify gate passed 48/48 checks over 24 of the 60 godmode passes.\n\nThe honest breakdown — including where the plain agent won and why the pass difference is within noise — is in the [showdown post](docs/blog/godmode-vs-plain-agent-benchmark.md).\n\n---\n\n## How It Works\n\nEvery iterative skill follows the same disciplined loop:\n\n```text\n1. REVIEW -- read state, logs, git history\n2. IDEATE -- pick the next change (informed by past failures)\n3. MODIFY -- make ONE atomic change, commit before verify\n4. VERIFY -- run guard (tests + lint + build must all pass)\n5. DECIDE -- improved --> KEEP. Worse --> DISCARD (git reset)\n6. LOG -- append to .godmode/-results.tsv\n7. REPEAT -- until goal met or iteration budget exhausted\n```\n\nNo human approval needed between iterations. Every experiment is committed to git. Every discard is classified and remembered.\n\n### The Full Pipeline\n\n```text\nTHINK --> PLAN --> BUILD --> TEST --> FIX --> OPTIMIZE --> SECURE --> SHIP\n```\n\nGodmode detects which phase you're in and routes to the right skill. Skills chain automatically — optimize finds an issue, triggers fix, re-optimizes, then ships.\n\n### Failure Intelligence\n\nEvery discard is classified into one of 8 failure types and logged to `.godmode/-failures.tsv`. On 3+ consecutive failures, the agent writes a reflective diagnosis analyzing the pattern before trying again. Lessons persist in `.godmode/lessons.md` across sessions.\n\n### Multi-Agent Execution\n\nComplex tasks are decomposed and run in parallel:\n\n```text\nRound 1: Agent 1 [worktree] --\\\n Agent 2 [worktree] ---+-- merge + test\n Agent 3 [worktree] --/\nRound 2: Agent 4 [worktree] --\\\n Agent 5 [worktree] ---+-- merge + test\n```\n\nMax 5 agents per round. Each gets its own git worktree. Merge sequentially, test after each merge.\n\n---\n\n## What Fires by Default\n\nEvery `/godmode:*` invocation — and every natural-language request that routes to a pipeline skill — fires the full learning stack automatically. No flags, no opt-in. See [`SKILL.md §14 Default Activations`](SKILL.md) for the authoritative list.\n\n### Authoring discipline (Karpathy family)\n\n- **Principles prelude** — every agent reads [`skills/principles/SKILL.md`](skills/principles/) before the first Edit. Four rules: Think Before Coding (state assumptions, surface alternatives), Simplicity First (pre-MODIFY strike for single-use helpers and unrequested configurability), Surgical Changes (line-trace rule), Goal-Driven Execution (success criterion is a shell command exiting zero).\n- **Pre-commit discard audit** — builder, tester, and optimizer agents classify every hunk in `git diff --cached` before committing. `line_scope_drift` hunks (formatting churn, \"while we're here\" refactors, adjacent improvements) are surgically dropped via `git restore -p --staged`. Spec: [`docs/discard-audit.md`](docs/discard-audit.md).\n- **DispatchContext schema** — all 7 subagents validate their input at dispatch time. Missing required field → `BLOCKED: invalid_dispatch`. See [`AGENTS.md § DispatchContext Schema`](AGENTS.md).\n- **Discard cost hierarchy** — Cost-0 (pre-MODIFY strike) / Cost-1 (pre-commit audit) / Cost-2 (post-commit revert). Cost-2 discards that should have been caught earlier are logged as `escaped_discard` feedback to `.godmode/lessons.md`.\n\n### Token optimization (four-layer stack)\n\n- **Progressive Disclosure routing** — the orchestrator reads only Tier 1 of each skill file (~20 lines) to match triggers. ~2,700 lines to route vs ~27,000 for full reads. **~90% routing-time context reduction**.\n- **Stdio input-side compression** — [`skills/stdio/SKILL.md`](skills/stdio/) documents 13 canonical command patterns (`git log` → `git log --oneline -20`, `cat` → `wc -l`, `ls -la` → `ls -1`, etc.) every agent prefers. Pairs with [rtk](https://github.com/rtk-ai/rtk) for shell-hook level enforcement.\n- **Terse output-side compression** — [`skills/terse/SKILL.md`](skills/terse/) auto-activates from round 2 onward. Compresses round summaries, status lines, agent reports by 40-60%. TSVs, code, errors, commit messages, and final summary stay verbose. Opt out with `/godmode:terse off` or `GODMODE_TERSE=0`.\n- **Token observability** — [`skills/tokens/SKILL.md`](skills/tokens/) logs per-round input/output token counts to `.godmode/token-log.tsv` using a reproducible `chars/4` heuristic. Answers \"is my loop getting cheaper or more expensive?\" with one awk one-liner. Opt out with `GODMODE_TOKENS=0`.\n\n### Coordination and research\n\n- **Named coordination patterns** — every plan declares its outermost pattern from [`docs/coordination-patterns.md`](docs/coordination-patterns.md): Pipeline, Fan-out/Fan-in, Expert Pool, Producer-Reviewer, Supervisor, or Hierarchical Delegation. Plans without a declared pattern return `BLOCKED: invalid_plan`.\n- **Research auto-dispatch** — before routing to `think` on any non-trivial task (mentions external lib/framework, >5 file scope, no prior `.godmode/research.md`), the orchestrator auto-dispatches [`skills/research/SKILL.md`](skills/research/) to gather prior art. Skip for trivial fixes or with `--no-research`.\n\n### The 8 pipeline skills inherit by default\n\n```text\nTHINK → PLAN → BUILD → TEST → FIX → OPTIMIZE → SECURE → SHIP\n```\n\nEach of these skills has a `Rule 0` in its Hard Rules section explicitly inheriting all of the above. No per-skill opt-in. One command runs the whole stack.\n\n---\n\n## Subagents (7)\n\n| Agent | Role |\n|-------|------|\n| **planner** | Decomposes goals into parallel tasks |\n| **builder** | Implements tasks with TDD in isolated worktrees |\n| **reviewer** | Code review for correctness, security, performance, style |\n| **optimizer** | Autonomous measure --> modify --> verify loop |\n| **explorer** | Read-only codebase reconnaissance |\n| **security** | STRIDE + OWASP audit with 4 adversarial personas |\n| **tester** | TDD test generation, RED-GREEN-REFACTOR |\n\n---\n\n## Skills (135)\n\nGodmode includes 135 skills across 13 domains. Each skill encodes a real engineering workflow — not just instructions, but a complete protocol with verification steps.\n\n| Domain | Count | Highlights |\n|--------|-------|------------|\n| **Core Workflow** | 16 | `godmode` `think` `plan` `build` `test` `review` `optimize` `debug` `fix` `ship` `verify` `goal-bridge` **(new)** |\n| **Discipline & Context** | 8 | `principles` `terse` `stdio` `tokens` `research` `bench` `team` `tutorial` **(new)** |\n| **Architecture & Design** | 10 | `architect` `rfc` `ddd` `pattern` `schema` `distributed` `scale` `migration` |\n| **API & Backend** | 14 | `api` `graphql` `grpc` `orm` `cache` `queue` `event` `realtime` `webhook` |\n| **Frameworks** | 12 | `react` `nextjs` `vue` `svelte` `node` `fastapi` `django` `rails` `spring` |\n| **Security & Compliance** | 8 | `secure` `auth` `rbac` `pentest` `devsecops` `comply` |\n| **Testing** | 7 | `e2e` `integration` `loadtest` `perf` `webperf` |\n| **DevOps & Infra** | 16 | `k8s` `docker` `cicd` `ghactions` `infra` `observe` `resilience` |\n| **Frontend & UI** | 9 | `ui` `a11y` `seo` `mobile` `designsystem` `responsive` |\n| **Databases** | 3 | `postgres` `redis` `nosql` |\n| **AI & ML** | 5 | `ml` `mlops` `rag` `prompt` `eval` |\n| **Developer Experience** | 13 | `docs` `refactor` `git` `pr` `monorepo` `changelog` `slo` |\n| **Integrations** | 14 | `i18n` `pay` `cli` `agent` `feature` `chaos` `experiment` |\n\n### The 8 Discipline & Context skills (shipped across Phases 0–E)\n\n| Skill | What it does |\n|---|---|\n| [`principles`](skills/principles/) | Karpathy authoring-discipline prelude. Four rules every agent reads before the first Edit. Imported via `@./` so every adapter gets it automatically. |\n| [`terse`](skills/terse/) | Output compression for long autonomous loops. Auto-activates from round 2. 40-60% token reduction on emit side. TSVs, code, errors, commits, final summary stay verbose. |\n| [`stdio`](skills/stdio/) | 13 canonical command patterns for minimizing tool-output token waste. Godmode-native alternative/complement to [rtk](https://github.com/rtk-ai/rtk). |\n| [`tokens`](skills/tokens/) | Per-round token-budget observability. Logs input/output counts to `.godmode/token-log.tsv`. Answers \"is my loop getting cheaper over time?\" |\n| [`research`](skills/research/) | Prior-art gathering phase. Auto-dispatched before `think` on non-trivial tasks. Uses the `explorer` subagent. Writes `.godmode/research.md`. |\n| [`bench`](skills/bench/) | Formal benchmark harness. 3-arm eval (baseline + variants) with N-run variance recovery. Writes `.godmode/bench-results.tsv`. |\n| [`team`](skills/team/) | Team bundle primitive. Compose existing skills into named sequences via YAML bundles in `.godmode/teams/.yaml`. Uses one of the 6 coordination patterns. |\n| [`tutorial`](skills/tutorial/) | Day-0 walkthrough. 7 steps, ≤5 minutes, first `/godmode:optimize` run. |\n\nFull skill reference: [skills/](skills/)\n\n---\n\n## Platforms\n\n| Harness | Install | Integration | Agents | Models |\n|---|---|---|---|---|\n| **Claude Code** | `claude plugin install godmode` | Plugin (`commands/`, `agents/`, `.claude-plugin/` marketplace) | Parallel (worktrees) | Full |\n| **pi** | `bash adapters/pi/install.sh` | Skills dir — `$HOME/.pi/agent/skills/godmode/` | Skills only | Full |\n| **omp** | `PREFIX=\"$HOME/.omp/agent/skills\" bash adapters/pi/install.sh` | Skills dir — `$HOME/.omp/agent/skills/godmode/` + one-line `customDirectories` registration (omp scans one level per skill; see `adapters/pi/README.md`) | Skills only | Skill-level (same corpus as pi) |\n| **Codex** | `bash adapters/codex/install.sh` | Copies `.codex/` agents into your repo; harness reads root `AGENTS.md` | Native (sequential) | Partial |\n| **OpenCode** | `bash adapters/opencode/install.sh` | Plugin (`plugin.json` + adapter entry) | Sequential | Partial |\n| **Gemini CLI** | `bash adapters/gemini/install.sh` | Generated files in your repo (`GEMINI.md` + config) | Sequential | Session-level |\n| **Cursor** | `bash adapters/cursor/install.sh` | Generated files in your repo (`.cursorrules`) | Background agents | Session-level |\n| **Amp** | `bash adapters/amp/install.sh` | Skills dir — `.agents/skills/` in your repo (Amp-native) + root `AGENTS.md` | Skills only | — |\n\n> **omp:** served by the same pi installer — `~/.omp/agent/skills` is omp's native user skills dir, confirmed from upstream source (can1357/oh-my-pi @ main, 2026-08-27) and verified against omp v18.0.8 (linux-x64). omp scans one level per skill, so the installed `godmode/` wrapper is not auto-discovered — register the collection once in `~/.omp/agent/config.yml`:\n>\n> ```yaml\n> skills:\n> customDirectories:\n> - ~/.omp/agent/skills/godmode\n> ```\n>\n> With an active omp profile, skills load from `~/.omp/profiles//agent/skills` instead. `PREFIX` is retained for pi-style forks — see [adapters/pi/README.md](adapters/pi/README.md) for the full omp walkthrough.\n> **Amp:** reads root `AGENTS.md` and the `.agents/skills/` project skills directory natively ([docs](https://ampcode.com/docs/customize/skills)); the adapter wires godmode's skills there. Subagents and model routing are Amp's own — the adapter wires neither.\n\nAll 135 skills work on every platform. Parallel agent skills automatically degrade to sequential on platforms without native agent dispatch. The authoring-discipline prelude, Progressive Disclosure routing, and pre-commit discard audit all reach every adapter — Claude Code via `SKILL.md`, Gemini and OpenCode via their respective entry files importing `@./skills/principles/SKILL.md`.\n\nVerify your installation: `bash adapters//verify.sh` (pi family: point it at your skills dir, e.g. `PREFIX= bash adapters/pi/verify.sh` for an omp install)\n\n## Multi-model routing\n\nZero config is a valid, complete default: with no configuration at all,\nevery role inherits the session model, and godmode works fully with zero\nsetup. A missing `godmode.models.json` is a valid state, and model tiering is\nopt-in only. When a role *is* pinned, resolution is exactly one chain:\n\n`GODMODE_MODEL_` env -> `godmode.models.json` roles -> session model.\n\nThe repo-root `godmode.models.json` wins per key over\n`~/.config/godmode/models.json` when both exist: a role pinned in the repo\nfile wins, while roles set only in the user file still apply.\n\nA one-off override is an env var per role — role name uppercased, dashes to\nunderscores (`code-review` -> `GODMODE_MODEL_CODE_REVIEW`):\n\n```bash\nGODMODE_MODEL_EXECUTE=zai/glm-5.3-flash GODMODE_MODEL_PLAN=anthropic/claude-opus-4 godmode optimize\n```\n\nDurable config lives in `godmode.models.json` — at your project root (commit\nit to share team-wide routing) or in `~/.config/godmode/models.json` (personal,\nmachine-wide). Mix providers freely:\n\n```json\n{\n \"roles\": {\n \"plan\": \"anthropic/claude-opus-4\",\n \"review\": \"anthropic/claude-opus-4\",\n \"build\": \"openai/gpt-5.2\",\n \"optimize\": \"zai/glm-5.3-flash\"\n }\n}\n```\n\n> **Illustrative example:** a 20-round optimize loop can run fast/cheap\n> executors under strong reviewer models — spend shifts toward the rounds that\n> benefit instead of paying premium prices for every round. Actual savings\n> depend on your providers and pricing.\n\nCapability tiers — which harnesses accept per-child model params:\n\n| Harness | Model routing |\n|---|---|\n| **pi** | Full — per-child model params |\n| **Claude Code** | Full |\n| **OpenCode** | Partial |\n| **Codex** | Partial |\n| **Cursor** | Session-level |\n| **Gemini CLI** | Session-level |\n\nInspect what will run before you start: `godmode doctor` (`/godmode:doctor`)\nprints the resolved role -> model table with each value's source\n(env / file / session).\n\nRoles are open-ended — there is no fixed enum — and any unknown or unpinned\nrole simply falls back to the session model. Details and the full reference\nresolver: [`adapters/pi/README.md`](adapters/pi/README.md).\n\n---\n\n## Philosophy\n\n**Discipline before speed.** Every change is measured. Bad changes are reverted.\n\n**Evidence before claims.** \"Looks good\" is rejected. Numeric proof is required.\n\n**Git is memory.** Every experiment is committed. Every revert is in the log.\n\n**Keep or discard.** Binary decisions only. No maybes.\n\n**Learn from failure.** Every discard is classified and remembered.\n\n**Simplicity first.** Complex changes that marginally improve metrics are discarded. The pre-MODIFY checklist strikes speculative code before it's written.\n\n**Think before coding.** If two interpretations exist, present them — never silently pick. Emit `NEEDS_CONTEXT` on ambiguity.\n\n**Surgical changes only.** Every semantically changed line must trace directly to the user's request. Adjacent \"improvements\" are `line_scope_drift` — dropped at pre-commit audit.\n\n**Goal-driven execution.** Success is a shell command that exits zero. Subjective criteria (\"works well,\" \"looks good,\" \"is faster\") are vibes — reject them before coding.\n\n**Token budget is a first-class metric.** Input-side (stdio), routing (Progressive Disclosure), output-side (terse), and observability (tokens) stack multiplicatively. Every round logs its context cost.\n\n---\n\n## Contributing\n\nEvery skill is a Markdown file. If you can write clear instructions, you can add a skill.\n\nSee **[CONTRIBUTING.md](CONTRIBUTING.md)** for the complete guide:\n\n- [Adding a New Skill](CONTRIBUTING.md#complete-skill-creation-guide)\n- [Skill Quality Checklist](CONTRIBUTING.md#skill-quality-checklist)\n- [Adding a New Platform Adapter](CONTRIBUTING.md#adding-a-new-platform-adapter)\n- [Testing Your Changes](CONTRIBUTING.md#testing-your-skill)\n- [Style Guide](CONTRIBUTING.md#skill-writing-style-guide)\n\n---\n\n## Frequently Asked Questions\n\n
\nHow does Godmode differ from Copilot, Cursor, or other AI coding tools?\n\nGodmode is not a replacement for these tools — it's a plugin that makes them better. Copilot and Cursor generate code in a single pass. Godmode adds autonomous iteration: it measures results, keeps improvements, reverts failures, and repeats until the goal is met. It works *inside* Cursor and Claude Code, not instead of them.\n
\n\n
\nDoes Godmode work with any programming language?\n\nYes. Godmode skills are language-agnostic — they define engineering workflows (test, measure, verify), not language-specific syntax. If your AI assistant supports a language, Godmode's skills work with it. Framework-specific skills (React, Django, Rails, etc.) provide additional specialized guidance.\n
\n\n
\nCan I use Godmode for free?\n\nGodmode itself is free and open source (MIT license). You need a working installation of one of the supported AI coding tools (Claude Code, Cursor, Codex, Gemini CLI, or OpenCode), which may have their own pricing.\n
\n\n
\nWhat does \"autonomous\" mean? Does it run without human input?\n\nYes. Once you set a goal and a metric (e.g., \"reduce API latency, measured by `curl -w '%{time_total}'`\"), Godmode runs the optimization loop autonomously — modifying code, verifying results, keeping improvements, reverting failures — without asking for approval between iterations. You can set an iteration limit or let it run until interrupted.\n
\n\n
\nIs it safe? Can it break my code?\n\nEvery change is committed to git *before* verification. If a change makes things worse, it's automatically reverted with `git reset`. Your codebase never stays in a broken state. Guard commands (tests, linting, build) must pass for any change to be kept.\n
\n\n
\nHow do I add my own skills?\n\nEvery skill is a Markdown file. Copy an existing skill, modify it, and drop it in the `skills/` directory. See [CONTRIBUTING.md](CONTRIBUTING.md) for the full guide. New skills should follow the Progressive Disclosure convention: `## Activate When` immediately after the frontmatter (Tier 1), workflow + hard rules next (Tier 2), optional `` marker before examples and error recovery.\n
\n\n
\nWhat is the authoring-discipline prelude?\n\nA set of four behavioral rules based on Andrej Karpathy's observations on common LLM coding mistakes, shipped as [`skills/principles/SKILL.md`](skills/principles/). Every agent reads it before the first Edit on every task: (1) Think Before Coding — state assumptions, never silently pick; (2) Simplicity First — run the pre-MODIFY checklist, strike speculative code; (3) Surgical Changes — line-trace rule, adjacent improvements are `scope_drift`; (4) Goal-Driven Execution — success is a shell command exiting zero. Enforced mechanically by the pre-commit discard audit in `agents/builder.md`, `agents/tester.md`, and `agents/optimizer.md`.\n
\n\n
\nHow does Godmode save tokens on long autonomous loops?\n\nFour layers of compression that stack multiplicatively:\n\n1. **Progressive Disclosure** — the orchestrator reads only Tier 1 (~20 lines) of each skill when routing. ~2,700 lines to route vs ~27,000 for full reads. ~90% routing-time context reduction.\n2. **Stdio input-side compression** — 13 canonical command patterns (`git log --oneline -20` instead of `git log`, `wc -l` instead of `cat`, etc.). Godmode's native convention; pairs with [rtk](https://github.com/rtk-ai/rtk) for shell-hook enforcement.\n3. **Terse output-side compression** — auto-activates from round 2. Round summaries and agent reports compress 40-60%. TSVs, code, errors, commits, final summary stay verbose.\n4. **Token observability** — per-round input/output counts logged to `.godmode/token-log.tsv` so you can actually measure whether the other three are helping.\n\nOpt out selectively: `/godmode:terse off`, `GODMODE_TOKENS=0`. Principles prelude and Progressive Disclosure have no opt-out — they're mechanical gates.\n
\n\n
\nWhat does \"Default Activations\" mean?\n\nEvery improvement shipped in Phases 0–E fires automatically on every `/godmode:*` invocation. No flags. No opt-in. The 8 pipeline skills (`think`, `plan`, `build`, `test`, `fix`, `optimize`, `secure`, `ship`) each inherit the full stack via a `Rule 0` in their Hard Rules section that references [`SKILL.md §14 Default Activations`](SKILL.md). A first-time user running `/godmode make my API faster` on a fresh repo gets: research auto-dispatched (if non-trivial), principles prelude read by every agent, pre-commit audit dropping drift hunks before commit, terse mode activating at round 2, token logging at every round, Progressive Disclosure routing at ~90% context reduction, and a plan with a declared coordination pattern. Every one of those was opt-in before Phase E.\n
\n\n---\n\n## Community\n\nWins, questions, and failure stories are welcome in [Discussions](https://github.com/arbazkhan971/godmode/discussions). Share what godmode kept — or reverted — for you.\n\n

\"Star

\n\n---\n\n## License\n\nMIT -- see [LICENSE](LICENSE).\n\n---\n\n
\n\n**Discipline before speed. Evidence before claims. Git is memory.**\n\n**[Install](#quick-start)** | **[Docs](docs/)** | **[FAQ](docs/FAQ.md)** | **[Troubleshooting](docs/troubleshooting.md)** | **[Discuss](https://github.com/arbazkhan971/godmode/discussions)**\n\n
\n"}