Plain files
Read them, grep them, diff them, review them and carry them between compatible clients. No vector store is needed for this instruction layer.
For work that needs to resume, one Markdown goal or TODO can preserve the next step.Portable agent workflow / v1.14.0
From task
to done.
Tell it what you need. Flow clarifies the goal, does the work, and checks the result. Investigate a problem, build something, prepare a document or improve what you already have.
17focused skills
1entry point: nb-flow / nb-ultra
MITinspectable source
ONE TASK
EXAMPLE WORKFLOW
OWNER REQUEST
โFix this bug. Show me how you checked it.โ
01Understandreproduce the failure
02Do the workmake the smallest correction
03Checkrun the relevant check
Your first useful result
Give it a task.
Wait until itโs done.
Install Flow, start a new conversation and choose an example. Add your details and send the task to your AI assistant.
Read the trial guide โSay โUse nb-flowโ or โUse nb-ultraโ with your task. Both name the same workflow. In Codex, select $nobrainer-ultra explicitly if automatic selection misses it.
01 Compare three offers Research
Use nb-flow. Compare these three offers against my requirements. Check the current sources, show the total costs and tradeoffs, and recommend one. Ask if a missing preference would change the recommendation.
Accept when: the comparison cites its sources, separates facts from assumptions and explains the recommendation.
02 Improve this text Writing
Use nb-ultra. Rewrite this text so my audience understands it on the first read. Keep my meaning and facts, remove repetition and give me a version ready to use. Ask who the audience is if that is unclear.
Accept when: the revised text preserves the facts, fits the audience and contains no invented claims.
03 Turn an idea into a plan Everyday work
Use nb-flow. Turn these notes into a practical plan for our community event. List the tasks, dependencies and decisions still needed. Prepare an invitation using only confirmed details.
Accept when: the plan is actionable, missing decisions are visible and the invitation uses confirmed facts.
04 Implement a feature Development
Use nb-ultra. Implement this feature in the existing project. Clarify any missing requirement that changes the result, reuse what fits and verify the agreed behavior. Show the change and remaining limitations.
Accept when: the agreed behavior works, relevant checks pass and the change stays within scope.
05 Fix a problem Problem solving
Use nb-flow. Investigate this problem, reproduce it where possible and fix its cause. Verify the result and explain what changed. If you cannot verify something, say exactly what remains unchecked.
Accept when: the cause is supported by evidence, the correction is verified and unresolved gaps are explicit.
06 Test before sharing Quality
Use nb-ultra. Check this document, spreadsheet or feature against its intended purpose. Prioritize errors that could mislead users, verify the important cases and give me a short findings list with evidence.
Accept when: findings point to specific evidence, important cases were checked and untested areas are named.
The useful middle layer
Start with the task.
A structure that fits.
When the task is clear, the agent starts. When a missing decision changes the work, it asks a focused question up front. Skills supply the guidance that task needs, from a quick answer or small edit to a longer delivery with review and recovery.
One honest path
Simple coding, research, writing and planning stay direct. Add a plan or durable record when dependencies, risk or resuming the task make it useful.
SDD, wiki setup and diagrams are optional tools for a real need.Evidence over theatre
A clean validator proves structure. A runtime readback proves behavior. A public URL proves publication. Each claim keeps its own proof level.
Unknown stays visible instead of becoming confident copy.One flow, two speeds
Right-size the ceremony.
Clear task: start. Missing decision: ask up front. Add coordination only when it helps produce a result we trust.
QUICK PATHsmall / reversible
Do the clear task.
For simple coding or everyday work with a clear outcome and a useful check.
understandโdoโcheckโdeliver
RESULT The requested result and the check that matters.
FULL PATHnon-trivial / recoverable
Keep complex work manageable.
For dependent steps, several proof layers or recovery needs. Keep a short plan and one durable Markdown goal or TODO if useful.
scopeโplanโdoโreviewโcheck
RESULT A verified delivery or a concrete owner gate.
WHEN USEFUL
OUTCOMENON-GOALSPROOFUNTOUCHEDDONE CLEAN
Open the optional detailed workflow map
The portfolio
Fifteen boundaries.
No orphan skills.
Each skill owns a recurring cross-project boundary. Aliases are triggers, not duplicate directories. The set is intentionally flat enough to inspect in one sitting.
ORIENT
nobrainer-ultraEnd-to-end setup and delivery nobrainer-decideConsequential choices nobrainer-spec-driven-developmentDurable contractsEXECUTE
nobrainer-buildSmallest verified implementation nobrainer-browserRendered behavior and traces nobrainer-researchBounded current research nobrainer-writingHigh-signal human-facing proseGUARD
nobrainer-reviewAcceptance and release close nobrainer-securityThreat and supply-chain review nobrainer-rcaEvidence-backed causal diagnosis nobrainer-autoimproveBaseline, variant, holdoutCOORDINATE
nobrainer-teamMinimum useful roster nobrainer-dispatcherBounded ready-set scheduling nobrainer-sessionsNamed, visible handoffs nobrainer-wikiSourced durable knowledgeThe open-source landscape
Twenty-nine projects. Twenty-nine useful signals.
Snapshot read on 03 Sep 2026. Counts drift; the point is fit, not a leaderboard.
ECC โ
ECC combines a broad workflow pack with hooks, learning and optional services. NoBrainer keeps fifteen skills and adds an explicit bounded command runner. The useful comparison is what each mechanism actually controls, how it is installed and what has been tested.
Read the comparison โAWS AI-DLC โ
AWS AI-DLC publishes a staged development lifecycle from one harness-neutral core. NoBrainer chooses a short or full workflow according to the task. Both make planning and verification explicit; the difference is the amount of lifecycle structure you choose to maintain.
Read the comparison โgstack โ
gstack presents an opinionated virtual engineering team, including browser QA, review and release workflows. NoBrainer organizes work around one task and adds specialists selectively. The first-use experience is worth studying; the author's productivity claims are not our benchmark.
Read the comparison โAgent Skills โ
The open format: SKILL.md, optional scripts, references and assets, with progressive disclosure.
Signal package once, reuse across compatible clients.
Superpowers โ
A composable development method built around brainstorming, plans, TDD, worktrees and review.
Signal put the workflow before the edit.
Caveman โ
Compresses agent prose while keeping code, commands, paths and exact errors intact.
Signal cut ceremony, not diagnostic payload.
Ponytail โ
Uses a first-rung-that-holds ladder: skip, reuse, native capability, then minimum implementation.
Signal YAGNI is a routing decision, not a slogan.
Autoresearch โ
A bounded loop modifies, measures and keeps or discards an experiment against a fixed signal.
Signal freeze the evaluator before self-improvement.
Karpathy Wiki โ
A Markdown-native wiki pattern where agents read, write and lint durable project knowledge.
Signal memory should be inspectable and source-linked.
GitHub Agentic Workflows โ
Compiles Markdown workflow sources into GitHub Actions with read-only jobs and controlled safe outputs.
Signal keep deterministic work deterministic and writes scoped.
Deep Agents โ
An opinionated runtime with planning, sub-agents, persistence, filesystem backends and human gates.
Signal add a harness when long-horizon state earns it.
agnix โ
Validates instruction files, skills, hooks and tool configuration across agent ecosystems.
Signal treat agent instructions as code.
12-Factor Agents โ
Principles for owning prompts, context, control flow, state, pause/resume and focused agents.
Signal make control flow and human contact explicit.
GitHub Spec Kit โ
A formal, extensible specification-driven path from intent through plan, tasks, implementation and convergence.
Signal make the build contract a durable artifact.
Addy Osmani Agent Skills โ
A broad, production-oriented collection covering review, planning, frontend quality, security and more.
Signal reusable specialists can stay composable.
Planning with Files โ
Persistent Markdown plans, progress and recovery hooks for long-running, interrupted agent sessions.
Signal keep the next state outside the context window.
Anthropic Skills โ
Reference and example skills that show how a client can load focused instructions, scripts and resources.
Signal keep the unit self-contained and progressively disclosed.
Matt Pocock Skills โ
Focused engineering skills for research, specification, implementation, TDD and review.
Signal compose sharp moves instead of one oversized prompt.
memsearch โ
Markdown-first semantic memory with a rebuildable index and hybrid retrieval when the corpus earns it.
Signal make memory searchable without confusing recall with proof.
avoid-ai-writing โ
Signal-based editing that removes mannered prose while preserving voice, quotes and code.
Signal sound human; keep the technical payload.
reverse-skill โ
Route intent to a specialist and toolchain, then execute within scope and journal the result.
Signal route first, bind power before execution.
Hermes dev-pair โ
Uses a read-only second model from a different family for deliberate, accountable review.
Signal independent review is an earned layer, not an automatic tax.
Agent Teams AI โ
A coordinator UI/runtime for agent teams across coding clients, with tasks, kanban state, messages and review.
Signal Coordination can be a runtime surface; Flow keeps delegation procedural.
Read the comparison โ
Loop Engineering โ
Patterns and CLI tools for discovering work, handing it to agents, verifying results and persisting state with approval gates.
Signal Recurring loops earn their state and gates.
Read the comparison โ
LazyCodex โ
A Codex-centered harness for project memory, planning, execution and verification hooks in complex codebases.
Signal Host-specific continuity is a separate adoption choice.
Read the comparison โ
Context Mode โ
An optional MCP and hook runtime for sandboxed tool output, code analysis, routing and session memory.
Signal Context plumbing should stay distinct from final prose and proof.
Read the comparison โ
last30days โ
A multi-source research workflow for current social, news, video, code and market signals.
Signal Connector breadth still needs dates, provenance and review.
Read the comparison โ
Impeccable โ
Visual commands and deterministic UI diagnostics for AI-slop patterns and broader design quality.
Signal A detector complements brand direction and rendered checks.
Read the comparison โ
Awesome Claude Code Subagents โ
A collection of specialized Claude Code subagents organized through plugin categories.
Signal Select the specialist; keep integration and trust review explicit.
Read the comparison โ
WHAT THIS CLARIFIES
These projects optimize different layers. NoBrainer starts from the task: one canonical skills tree, direct execution for clear work, focused questions when needed and optional structure for coordination and recovery.
Proof ladder
Portable by source.
Honest at runtime.
Version 1.8.1 gives each supported fresh session the stable task name plus its own started DD-MM date. Adaptive handoff still depends on context pressure, remaining work and verified transfer; see the current release evidence. There is no fixed restart interval, universal rename API, model-equivalence or token-savings claim.
Integration / coding host: Codex runs the tools; NoBrainer Skills supply portable policy for scope, pauses and proof. Open the Codex integration โ
01SOURCE VALIDATEDPortable skill format; no client claim
02REPOSITORY CHECKEDDeterministic structure and suite gates
03CLIENT LOADEDClean-session discovery on a named client
04EXPLICIT RUNTIME VERIFIEDExplicit invocation and representative task
05RUNTIME VERIFIEDAutomatic discovery and representative task
06DISTRIBUTEDExact release and install-channel readback
Start with one skill
Give the agent
one outcome.
Use a release you can inspect. Start from release v1.14.0, validate it locally and preview the install. Read the compatibility evidence for your client before relying on it.
Read installation$ git clone --branch v1.14.0 https://github.com/nobrainer-tech/nobrainer-tech-flow.git
$ cd nobrainer-tech-skills
$ python3 scripts/validate_skills.py --suite
$ python3 scripts/install_skills.py --client codex
# Review the plan before using --apply.Clear answers
Less magic.
More signal.
What does โmodel-neutralโ mean here?
The core instructions are provider-neutral. Use the host's available model and tools within the agreed limits. Each client's loading and runtime behavior needs its own clean-session evidence.
Why not put everything in one AGENTS.md?
Keep the root map small and put a recurring boundary in the skill that owns it. Progressive loading protects attention and makes changes easier to review.
When does NoBrainer use more agents?
One primary agent is the default. Optional native subagents get bounded, independent tasks when parallel work or separate evidence helps; their results still need review.
Does every task need a goal tool or detailed tracking?
No. A missing native goal tool, token counter or telemetry does not block safe work. Keep simple tasks direct. For work that needs to resume, one Markdown goal or TODO can hold the outcome, next step and useful evidence.
What is the honest next step?
Open the source, choose one skill and test it on one representative task. If the proof does not pass, keep the gap visible and improve the smallest failing boundary.
Explore 18 more workflow peers and complements
Selected from the collected repository and developer Trending views. Source-backed comparisons, not benchmark rankings.
- AIDevOps ยท workflow / automation framework source โ
- Fernando Skills ยท spec-driven skills source โ
- Archon ยท workflow engine / harness builder source โ
- AI Software Factory ยท autonomous delivery workflow source โ
- LoopX ยท long-horizon control plane source โ
- FirstMate ยท agent team orchestration source โ
- Ruflo ยท agent meta-harness source โ
- Squad ยท agent team framework source โ
- Codex Autorunner ยท long-horizon meta-harness source โ
- beefstack ยท claude code workflow stack source โ
- ai-memory ยท memory / handoff runtime source โ
- LLM Wiki ยท knowledge workflow source โ
- Old Coder ยท verification workflow source โ
- delegate-skills ยท delegation workflow source โ
- Claude Skills ยท broad skill collection source โ
- ai-dotfiles ยท host-specific workflow source โ
- HumanLayer Skills ยท control-loop workflow source โ
- Brad Frost Skills ยท specialist / usage limits source โ
