donnyclaude
Advanced tools
The verified workflow engine for Claude Code. Durable .planning project state, phased roadmaps, specialized subagents, and deterministic verification gates - installed in one command.
The verified workflow engine for Claude Code.
Claude Code is excellent at generating code. It is weaker at long projects:
context gets messy, /clear loses decisions, and the model can call a task
"done" before tests, requirements, or security checks actually pass.
DonnyClaude wraps Claude Code in a durable planning layer and a deterministic verification loop:
.planning/ so work survives /clear and new sessionsnpx donnyclaude
One command installs the Donny engine, the skills and agents, coding rules for 13 languages, project hooks, and a curated MCP setup - then Claude itself walks you through your first project.
LLM coding tools are good at producing work. They are much worse at knowing when work is actually finished. The recurring failure modes:
/clearMost Claude Code setups are prompt packs: they change what the model is told, then still trust it to declare success. DonnyClaude is built around one idea: the model generates the work, but completion is decided by durable state, skeptical review, and deterministic gates.
Everything installs under ~/.claude/ (user scope); nothing touches a project
until you start one there.
~/.claude/skills/, agents/, commands/, hooks/, rules/ - the toolkit~/.claude/donny/ - the workflow engine~/.claude/bin/, cco-memory/, scrapers/, docs/, statusline.py, and a
skill-index registry (.donnyclaude-skill-index.json) - support tools~/.claude/CLAUDE.md - a fresh machine gets the full operating guide; an
existing file gets only a clearly marked, idempotent standards block appended~/.claude/settings.json - merged, with your original backed up to
settings.json.bak firstExisting settings are preserved, not clobbered. And you can inspect or reverse all of it:
npx donnyclaude --dry-run # preview what install would change - no writes
npx donnyclaude doctor # check installation health
npx donnyclaude diff # drift between installed files and this version (exit 1 = drift)
npx donnyclaude uninstall # remove every DonnyClaude-owned file; yours stay
uninstall accepts --dry-run, --yes, and --no-mcp. It removes only files
this package ships, strips just the managed block from CLAUDE.md, leaves
settings.json in place, and never touches anything you created yourself.
npx donnyclaude # install, then the setup wizard
# then, in Claude Code:
/donny-init # research -> requirements -> roadmap
/donny-plan-phase 1 # plan a phase (dependency + requirement gates)
/donny-execute-phase 1 # build it (parallel subagents, atomic commits)
/donny-verify-work 1 # conversational UAT
/donny-ship # open a PR once verification passes
Use /donny-progress any time to see where you are, /donny-autonomous to
advance the whole roadmap hands-off, and /donny-help for every command.
/plugin marketplace add d0nmega/donnyclaude
/plugin install donnyclaude@donnyclaude
The plugin surfaces the skills, commands, and agents through Claude Code's
plugin system. The Donny engine, hooks, rules, and MCP registrations still
come from npx donnyclaude - run that once first. Pick one distribution for
the skill surface: a plugin install on top of a full npx install registers
the slash commands twice.
A project under Donny keeps its state on disk, not in chat history:
.planning/
PROJECT.md # what is being built, and why
REQUIREMENTS.md # scoped requirements with IDs
ROADMAP.md # phases mapped to requirements
STATE.md # current position, decisions, blockers
phases/
01-foundation/
01-CONTEXT.md # locked decisions from the discuss step
01-01-PLAN.md # executable plan with tasks and gates
01-01-SUMMARY.md # what was actually built
01-VERIFICATION.md # goal-backward verification report
Every non-trivial task writes requirements, roadmaps, phase plans, summaries,
and verification reports to .planning/, so long-running projects survive
/clear and new sessions. Subagents get a curated, isolated slice of that
state rather than the whole transcript. The Context7 MCP supplies live library
docs so the model codes against current APIs, and a global operating guide
loads the coding and writing standards every session.
48 agents and 94 slash-command skills, each a deliberately engineered prompt with a single responsibility and a minimal tool grant: planners, implementers, reviewers, build-fixers, researchers, verifiers. The right prompt runs for the right job instead of one overloaded system prompt trying to do everything. Model-tiered execution (Opus for planning and verification, lighter tiers for mechanical work) keeps cost proportional to the task.
Under the agents sits a deterministic Node CLI that checks the things a language model should not guess at:
verify phase-verified - a phase ships only when its verification status is
actually passed, parsed from disk, never an LLM's say-soverify milestone-coverage - requirement coverage computed across plans,
summaries, and verifications, not hand-talliedverify threats-clear / security scan-secrets - open threats and leaked
credentials block the commit deterministicallyphase defer - a skipped phase is recorded honestly as deferred, never
silently marked completediscover -> plan -> execute -> verify -> ship -> repeat
The generator never grades its own work: a separate skeptical verifier checks
goal achievement by running code, and beneath that sits the engine gate so the
verdict cannot drift from the truth. Run one phase by hand, or hand the whole
roadmap to /donny-autonomous and let it advance unattended.
CLAUDE.md is the right place for project instructions. DonnyClaude adds the
workflow layer around it:
| Need | CLAUDE.md | DonnyClaude |
|---|---|---|
| Persistent project memory | partial | yes, via .planning/ |
| Phase-based roadmap | no | yes |
| Specialized subagents | manual | included |
| Deterministic verification gates | no | yes |
| Requirement coverage checks | no | yes |
| Autonomous phase loop | no | yes |
| Installable, repeatable setup | no | one command |
DonnyClaude is for developers who use Claude Code on projects large enough that chat history stops being enough. It earns its weight if you:
/clear often but need decisions and progress to survive itIt is probably overkill if you only use Claude Code for one-off edits, small scripts, or quick explanations.
| Component | Count | What it is |
|---|---|---|
| Skills | 94 | Slash commands - the Donny workflow plus utilities and language packs |
| Agents | 48 | Specialized subagents - planners, reviewers, verifiers, build-fixers |
| Commands | 63 | Prompt commands - review, verification, and harness utilities |
| Rules | 13 | Coding and writing standards (common + 12 language packs), auto-loaded |
| Hooks | 24 | Format, guard, and state-integrity scripts (33 registrations across 8 lifecycle events) |
| MCP | 2 | context7 (live docs) and playwright (browser), registered at user scope |
| Engine | 1 | The Donny CLI and workflow library |
| Config | 1 | A global operating guide (~/.claude/CLAUDE.md) that loads the rules; never clobbers an existing one |
uv for the bundled research scrapers (optional; self-bootstraps on first run)~/.claude/.The best way to improve DonnyClaude is to run it on a real repo and report where the workflow breaks. Especially useful:
DonnyClaude is MIT licensed. The Donny engine began as a fork of the GSD workflow engine and has since diverged substantially - rebuilt around deterministic verification gates, model-tiered subagents, and a subagent-safe research path.
The optional browser-harness integration references
browser-use/browser-harness
(MIT, Copyright (c) 2026 Browser Use); it is not bundled, and all credit for
that project belongs to its authors.
If you want Claude Code workflows with durable memory and real verification gates, star the repo so other Claude Code users can find it.
FAQs
The verified workflow engine for Claude Code. Durable .planning project state, phased roadmaps, specialized subagents, and deterministic verification gates - installed in one command.
The npm package donnyclaude receives a total of 9 weekly downloads. As such, donnyclaude popularity was classified as not popular.
We found that donnyclaude demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 1 open source maintainer collaborating on the project.
Security News
GPT-6 Astra hits 100% on ExploitBench and finds zero-days autonomously, while independent tests reveal scope violations and monitoring gaps.
Product
Socket can now send alerts and supply chain attack notifications to Microsoft Teams, with filters that route the right updates to each channel.
Security News
pnpm 12 rewrites the package manager in Rust, cutting install times by up to 90% while preserving pnpm 11 workflows and lockfiles.