{
  "markdown": "<p align=\"center\">\n  <a href=\"https://nobrainer.tech\">\n    <img src=\"assets/nobrainer-tech-logo.svg\" width=\"120\" alt=\"NoBrainer.tech logo\">\n  </a>\n</p>\n\n<h1 align=\"center\">NoBrainer Tech Flow</h1>\n\n<p align=\"center\">\n  From task to done.\n</p>\n\n<p align=\"center\">\n  <a href=\"https://github.com/nobrainer-tech/nobrainer-tech-flow/actions/workflows/validate.yml\"><img alt=\"Validation\" src=\"https://github.com/nobrainer-tech/nobrainer-tech-flow/actions/workflows/validate.yml/badge.svg\"></a>\n  <a href=\"LICENSE\"><img alt=\"MIT License\" src=\"https://img.shields.io/badge/license-MIT-27d8cf.svg\"></a>\n</p>\n\n<p align=\"center\">\n  <a href=\"https://nobrainer.tech\">NoBrainer.tech</a>\n  ·\n  <a href=\"https://nobrainer.tech/flow/\">Flow overview</a>\n  ·\n  <a href=\"https://nobrainertech.gumroad.com\">Production-ready agentic workflows</a>\n</p>\n\nTell it what you need. Flow clarifies the goal, does the work, and checks the result.\n\nSay **“Use nb-flow”** or **“Use nb-ultra”** (or invoke `$nobrainer-ultra` in Codex) to fix code, prepare\nan everyday document, or investigate a problem. Clear tasks go straight to execution;\nmeaningful ambiguity gets one focused question round. Done means the agreed\ncriteria are met and the result is checked. A real blocker is reported with the\nnext unblock action.\n\nA quick answer stays a quick answer. The model-neutral workflow uses one plan,\nbounded corrective attempts and a Markdown checkpoint for longer work. Native\ngoals, telemetry, subagents and client-specific tools are optional.\n\nExisting installations keep the `nobrainer-tech-skills` package/plugin ID and all\nskill commands. See the [Flow migration guide](docs/MIGRATION_TO_FLOW.md).\n\n[Source-backed comparison decisions](docs/reviews/2026-09-05-flow-coverage.md)\ncover the current collected Trending snapshot and explain what we adopted or retained.\n\n## Try one real task\n\n[Install safely](#install-safely), start a fresh session, then give the agent a\nsmall task with a checkable result:\n\n> Use nb-ultra. Fix one bug in this project. Reproduce it first, make the\n> smallest correction and run the relevant check. Tell me what changed and\n> what remains unverified. Ask only if a missing decision changes the result.\n\nPrefer a non-coding trial? [Try an invitation, a small code fix or a bounded\ncommand](docs/TRY_IT.md). The guide gives explicit acceptance criteria so you\ncan judge your own result. These are trials, not promised benchmark scores.\n\n## Install safely\n\nClone a reviewed ref, validate it, preview exact targets, then apply:\n\n```bash\n(\n  set -u\n  : \"${NB_REVIEWED_COMMIT:?set a reviewed full 40-character commit SHA}\"\n  test \"${#NB_REVIEWED_COMMIT}\" -eq 40 || exit 2\n  case \"$NB_REVIEWED_COMMIT\" in *[!0-9a-f]*) exit 2 ;; esac\n  git clone --no-checkout https://github.com/nobrainer-tech/nobrainer-tech-flow.git || exit 3\n  cd nobrainer-tech-flow || exit 3\n  git checkout --detach \"$NB_REVIEWED_COMMIT\" || exit 3\n  test \"$(git rev-parse HEAD)\" = \"$NB_REVIEWED_COMMIT\" || exit 3\n  python3 scripts/validate_skills.py --suite || exit 4\n  python3 scripts/install_skills.py --client codex || exit 4\n  python3 scripts/install_skills.py --client codex --apply || exit 4\n)\n```\n\nSet `NB_REVIEWED_COMMIT` to the exact full commit SHA you reviewed. Tags and\nbranches are rejected because they can move; every failed gate stops before the\nnext command.\n\nThe installer defaults to all fifteen canonical skills, supports an exact\nsubset, refuses foreign targets and can use links or copies. Restart the client\nand perform clean-session discovery before claiming runtime installation. Full\nclient-specific steps and rollback are in [Installation](docs/INSTALL.md).\n\n\n![NoBrainer Ultra: direct work for clear small tasks; focused clarification, bounded execution and verification when needed](assets/nobrainer-workflow.svg)\n\n### GitHub flow chart\n\n```mermaid\nflowchart TD\n    A[One outcome] --> B{Material ambiguity?}\n    B -->|yes| C[BUDDY: one focused question round]\n    B -->|no| D{Small and clear?}\n    C --> D\n    D -->|yes| E[Direct answer or edit; check the result]\n    D -->|no| F[SCOPE + PLAN: outcome, authority, proof; concise TODO]\n    F --> G[AUTOPILOT: execute the authorized scope]\n    G --> H[Verify; independent REVIEW when useful]\n    H -->|verified defect; attempt budget remains| G\n    H -->|acceptance met| I[Audit delegated artifacts and stop owned workers]\n    H -->|blocked| J[Checkpoint; report one unblock action]\n    E --> K[Deliver evidence and stop]\n    I --> K\n    F -. optional .-> L[Markdown goal for resume]\n    G -. independent work .-> M[Bounded native subagents]\n    M --> H\n    L -. context pressure; transfer supported .-> N[Fresh session: same task + started DD-MM]\n    N --> G\n    L -. unavailable native goals or telemetry .-> G\n```\n\n## Astra Ready Flow, portable by design\n\nVersion **1.6** removes client-specific prerequisites from ordinary work.\nOpenAI [introduced GPT-6 Astra](https://openai.com/index/gpt-6-astra/) on\nSeptember 3, 2026. The suite keeps the host-selected model and uses the same\nplain-text instructions with other models; it does not pin a provider or choose\nan expensive tier automatically.\n\n[Compatibility](docs/COMPATIBILITY.md) separates available models, tested\nbehavior, client loading and source distribution. Local smoke evidence covers\nonly its recorded source and scenarios. It is not a universal compatibility,\nquality or token-savings benchmark.\n\nThe [1.6.1 instruction review](docs/reviews/2026-09-05-astra-instructions.md)\nclarifies user authority, skill-caused pauses and when verification is sufficient.\nThe same rules apply across models; historical runtime proof stays tied to the\ntested release bytes.\n\n```text\nSmall task       -> direct result + relevant check\nLarger task      -> clarify if needed + short TODO + execute + verify\nResumable task   -> same workflow + one Markdown goal/checkpoint\nIndependent work -> optional bounded subagents, audited before integration\n```\n\nLoad specialists only when needed. Reuse the project's instructions, tests,\nspecs and wiki. Choose SDD for durable contracts and TDD when a failing test\nwould expose the behavior; neither requires installing a framework.\nSee the [v1.6 review and research decisions](docs/reviews/v1.6.0-review.md).\n\n## Start with one skill\n\nUse [`nobrainer-ultra`](skills/nobrainer-ultra/) for setup or any non-trivial\noutcome. A small reversible edit without a public contract, routing, workflow or\nportfolio change can use its quick path; other changes use the full path and its\ncoherence gate.\nIn Codex, explicit invocation is `$nobrainer-ultra`; `nb-flow` and `nb-ultra`\nare natural-language trigger phrases and therefore depend on a client's\nimplicit description matching.\n\n```text\nDRIFT_CHECK -> BUDDY -> SCOPE -> AUTOPILOT -> VERIFY -> RECEIVE_AUDIT -> LEARN\n```\n\n- `BUDDY` is the first and only ordinary clarification stage.\n- `SCOPE` freezes outcome, non-goals, expected files, proof, untouched work,\n  minimum solution, test decision and clean completion before a non-trivial write.\n- One canonical plan owns TODO state; the owner sees only a compact Progress\n  checklist and one next action.\n- A durable ledger adds exact identity, dependencies, checkpoints, retries and\n  rollback only for cross-session, dependency-rich or consequential work.\n- `AUTOPILOT` continues through routine approved work without repeated check-ins.\n- `nobrainer-team` selects the minimum useful capabilities;\n  `nobrainer-dispatcher` schedules only approved ready work in bounded batches;\n  `nobrainer-sessions` creates or reuses exact visible sessions only when\n  parallelism, isolation, handoff or independent evidence earns the cost.\n- Failed review returns to `nobrainer-build`; changed code invalidates old proof.\n- Merge, deploy, publishing, spending, credentials, destructive operations and\n  production mutation remain owner gates unless exact authority is already\n  recorded.\n\nThe two execution axes stay independent:\n\n```text\nCONTROL_MODE: BUDDY -> AUTOPILOT\nSESSION_MODE: MAIN | MULTI_SESSION\n```\n\nFor resumable or delegated work, `SESSION_HEALTH_GATE` uses available,\nconfigured limits. Missing optional telemetry lowers the health claim and still\nallows bounded safe work. An explicitly required hard budget must be enforceable.\n`RUNTIME_RELEASE` concerns actual task-owned workers; a completed result does not\nprove they stopped. `GOAL_FILE` is optional Markdown recovery state. A native goal\nis used only when available and authorized; the file alone suffices.\n\nAutopilot works in one MAIN session. Native subagents need a scoped assignment,\nobservable completion and reviewed output. Use persistent sessions and a\nDispatcher only when a real handoff or dependent queue needs them.\n\n## Fifteen skills, distinct ownership\n\nAliases are trigger phrases, not duplicate directories. Each skill owns one\nrecurring boundary:\n\n| Skill | Alias | Responsibility |\n|---|---|---|\n| [`nobrainer-ultra`](skills/nobrainer-ultra/) | `nb-flow`, `nb-ultra` | End-to-end setup and delivery: one requirements gate, concise progress, bounded execution, recovery, audit and learning |\n| [`nobrainer-team`](skills/nobrainer-team/) | `nb-team` | Minimal capability roster, installed-skill inventory and safe temporary specialist discovery |\n| [`nobrainer-dispatcher`](skills/nobrainer-dispatcher/) | `nb-dispatcher` | Dependency-aware ready-set scheduling, bounded dispatch, backpressure and audited result routing |\n| [`nobrainer-research`](skills/nobrainer-research/) | `nb-research` | Bounded current research from primary sources with facts separated from inference |\n| [`nobrainer-writing`](skills/nobrainer-writing/) | `nb-write`, `nb-brief` | High-signal drafting, compression and short human-sounding comments, issues and stories that preserve meaning, evidence, voice and action |\n| [`nobrainer-build`](skills/nobrainer-build/) | `nb-build` | Smallest verified implementation using calibrated KISS, DRY, SOLID, YAGNI and anti-slop gates |\n| [`nobrainer-security`](skills/nobrainer-security/) | `nb-security` | Threat models, security review, supply-chain audit and high-risk release evidence |\n| [`nobrainer-sessions`](skills/nobrainer-sessions/) | `nb-sessions` | Named visible sessions, exact identity, isolated writers, audited handoff, lease and recovery |\n| [`nobrainer-spec-driven-development`](skills/nobrainer-spec-driven-development/) | `nb-sdd` | Durable specification and acceptance ledger when contracts, risk or resumability justify it |\n| [`nobrainer-wiki`](skills/nobrainer-wiki/) | `nb-wiki` | Targeted retrieval and sourced durable knowledge without hidden memory or live task state |\n| [`nobrainer-browser`](skills/nobrainer-browser/) | `nb-browser` | Playwright-first rendered inspection, bounded CDP profile restart, approved session attach, browser tests and trace evidence |\n| [`nobrainer-autoimprove`](skills/nobrainer-autoimprove/) | `nb-autoimprove` | Measured baseline/variant/eval/holdout improvement with keep-or-revert |\n| [`nobrainer-decide`](skills/nobrainer-decide/) | `nb-decide` | Consequential decisions with different-shaped options, scoring, attack and one commitment |\n| [`nobrainer-rca`](skills/nobrainer-rca/) | `nb-rca` | Read-only causal diagnosis with a continuous evidence chain and explicit uncertainty |\n| [`nobrainer-review`](skills/nobrainer-review/) | `nb-review` | Acceptance trace, adversarial bug hunt and release close gate without speculative findings |\n\nThe [curation audit](docs/SKILL_CURATION.md) records why each skill exists and\nwhat belongs in another skill instead of becoming a sixteenth trigger.\n\n## Correct once, improve permanently\n\nUltra contains portable semantic hooks for four events:\n\n- `OWNER_DECISION_CHANGED` updates the canonical decision, marks the old value\n  superseded and invalidates dependent TODO items and evidence;\n- `AGENT_ERROR_CORRECTED` fixes the active result and classifies one minimal\n  prevention candidate: `AUTO_SCOPED` may persist it to one governed canonical\n  project-local store, `ASK` prepares one exact diff, and `OFF` keeps it\n  task-local without a durable diff;\n- `REVIEW_FAILED` creates a bounded Build correction and sends fresh evidence\n  back to Review;\n- `REPEATED_DEFECT` stops blind retries and invokes RCA with the prior failure\n  fingerprint.\n\nOnly durable, sourced, authorized and non-secret knowledge is promoted through\n`nobrainer-wiki`. One mutable fact has one canonical owner; the system does not\nappend contradictory copies to the plan, instructions and wiki.\nProject setup records `LEARNING_WRITE_POLICY: AUTO_SCOPED | ASK | OFF`, so an\nowner can enable automatic project-local learning without granting global or\npublishing authority.\n\n## Quality without AI slop\n\nThe shared delivery contract operationalizes:\n\n- `KISS`: the simplest complete design wins;\n- `YAGNI`: no speculative extension points or future options;\n- `DRY`: deduplicate owned knowledge and state, not incidental similarity;\n- `SOLID`: cohesive responsibilities and stable boundaries without class or\n  interface ceremony;\n- evidence: no invented APIs, fake runtime claims, placeholder logic, swallowed\n  errors, generic prose, broad unrelated rewrites or mock-only confidence;\n- content quality: purpose, audience, correctness sources, completeness,\n  coherence and target-workflow usefulness are frozen before execution.\n\nIf a decision-relevant fact may be current, niche, uncertain, high-stakes or\nsource-attributed, Ultra routes the smallest sufficient check through Research.\nA stable local syntax, import, test or configuration error starts from local\nevidence instead of an automatic wiki/web detour. If required primary evidence\nis unavailable, it says `RESEARCH_BLOCKED` instead of guessing.\n\n## Compatibility is a proof ladder\n\nAll clients consume the same `skills/` tree. Thin adapters cover Claude Code,\nCodex, Cursor, OpenCode, Gemini CLI, Kimi Code and Pi; the portable Agent Plugin\nmanifest and project instructions are the fallback for other Agent Skills\nconsumers.\n\nCompatibility claims use five distinct levels:\n\n```text\nSOURCE_VALIDATED -> REPOSITORY_CHECKED -> CLIENT_LOADED\n                 -> RUNTIME_VERIFIED -> DISTRIBUTED\n```\n\nA valid manifest does not prove clean-session routing. A local test does not\nprove production. See [Compatibility](docs/COMPATIBILITY.md) for current proof\nand [Testing](docs/TESTING.md) for acceptance evidence.\n\n\nCurrent source version: **1.8.1**. Check the\n[latest published GitHub release](https://github.com/nobrainer-tech/nobrainer-tech-flow/releases/latest)\nfor distribution and the [v1.8.1 release record](docs/releases/v1.8.1.md) for the\ncurrent source scope and preserved installation identities. The unchanged command runner keeps its\n[v1.7.0 verification scope](docs/releases/v1.7.0.md). Source publication does not\nimply client marketplace discovery or improved model reasoning. The earlier\n[v1.6.1 publication readback](docs/releases/v1.6.1-publication-readback.md)\nremains historical evidence.\n\nVersion [`v1.5.0`](docs/releases/v1.5.0-publication-readback.md) remains an\naccepted rollback source release. Its historical pre-publication checkpoint is\n[`docs/releases/v1.5.0.md`](docs/releases/v1.5.0.md).\n\nVersion [`v1.4.0`](docs/releases/v1.4.0-publication-readback.md) remains the\nprevious accepted source release and rollback option. Client-specific runtime\nrows remain evidence-scoped; source publication does not imply marketplace\ndiscovery.\n\nThe [`v1.3.1` release](docs/releases/v1.3.1-publication-readback.md) remains the\nprevious accepted source release and a rollback option. It adds the English\n`BRIEF` writing mode, concise issue/story templates and surface-specific bug\nevidence.\n\nThe [`v1.3.0` release](docs/releases/v1.3.0-publication-readback.md) remains an\nolder accepted source release and rollback option.\n\nVersion [`v1.2.1`](docs/releases/v1.2.1.md) remains the rollback source release\nat full commit `0010140d19a7ff847dff776569772ef04d82c314`, with its own exact\ntag, archive and isolated-install evidence. To reproduce the reviewed v1.3.0\nsource:\n\n```bash\ngit checkout --detach 8ae4a26548ce908fc5f98b22663f52e163541f56\ntest \"$(git rev-parse HEAD)\" = \"8ae4a26548ce908fc5f98b22663f52e163541f56\"\npython3 scripts/validate_skills.py --suite\npython3 -m unittest discover -s tests -q\n```\n\n`v1.2.0` remains published but failed archive acceptance; its exact boundary is\n[recorded separately](docs/releases/v1.2.0.md). GitHub reports the `v1.2.1`\nrelease object as non-immutable and tag protection was not independently\nverified, so security-sensitive consumers should pin the full commit SHA. The\ncurrent tag archive was re-read after the metadata-only history rewrite and\npassed 88/88 tests; historical CI binds the same tree, not the current commit\nidentity.\n[`v1.1.0`](docs/releases/v1.1.0.md) remains the accepted rollback anchor at full\ncommit `711be31d654835a04ef8c70674c3e493aeb2da8a`.\n\n## One source, thin adapters\n\n```text\nskills/               canonical portable behavior\nadapters/bootstrap.md small session-start route to Ultra\nhooks/                tested client lifecycle adapters\nscripts/              validation and conflict-safe installation\ntests/                deterministic and behavior-contract gates\ndocs/                 compatibility, curation, eval and release evidence\nassets/               brand and workflow diagrams\n```\n\nClient-specific forks of a skill are prohibited. External skills discovered by\nTeam are untrusted input: inspect the exact source/ref, instructions, scripts,\npermissions, network/credential behavior, trigger overlap and rollback. Prefer\ntemporary project-scoped use; persistent or global installation is an owner\ngate.\n\n## Attribution\n\n- `nobrainer-autoimprove` is an independent adaptation of Andrej Karpathy's\n  [autoresearch](https://github.com/karpathy/autoresearch) measure-change-keep\n  loop.\n- `nobrainer-wiki` is an independent adaptation of Andrej Karpathy's\n  [LLM wiki concept](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f).\n- Other external design sources remain cited inside the exact skill where they\n  influenced a behavior contract.\n\n## Contributing and security\n\nRead [CONTRIBUTING.md](CONTRIBUTING.md), [SECURITY.md](SECURITY.md) and\n[RELEASE-NOTES.md](RELEASE-NOTES.md). Every change goes through a focused PR,\npressure scenario, validators, diff review and secret scan. Do not report\npublication, distribution, live routing or user-visible success without\nreadback from that layer.\n\nNoBrainer Tech builds practical agentic workflows for teams that want speed\nwithout surrendering control. Learn more at [nobrainer.tech](https://nobrainer.tech)\nor browse ready-to-use workflow products on\n[Gumroad](https://nobrainertech.gumroad.com).\n\n## Adaptive session restart\n\nFor multi-day work, enable `session-restart` once. Flow keeps progress in the\nproject's existing task file and quietly assesses whether a fresh conversation\nwill repay its startup cost. It does not rotate simply because a day passed.\nSupported clients create a fresh continuation, verify its takeover and only then\narchive the old conversation. Other clients receive a compact manual handoff.\nWhen title mutation is supported, every source and successor keeps the stable task\nname and adds its own ` | started DD-MM` suffix. The full timestamp, timezone and\nsession ID remain in the registry; the short title is never used as identity.\n\nThis belongs to [Sessions](skills/nobrainer-sessions/references/session-restart.md),\nnot a sixteenth skill. The optional stdlib [decision helper](skills/nobrainer-sessions/scripts/restart_gate.py)\ncan serve a client hook without requiring one. The protocol is included in the\nv1.8.1 source; native transport and all-client savings are not implied. The\nv1.8.0 archive remains the rollback anchor. See [session restart](docs/SESSION_RESTART.md).\n",
  "bytes": 19817,
  "sha": "5d89c240e4158ea54cbad24a7cc927b9a75b926c8aba0161494ac08cdab87442",
  "repo_slug": "nobrainer-tech/nobrainer-tech-flow",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_nobrainer_tech_nobrainer_tech_flow_51f0edea/readme"
}