Power your agent
การติดตั้งจากผู้เผยแพร่ ไม่ใช่เอนด์พอยต์ของ PlugMCP
URL แบบรีโมตหรือคำสั่งบนเครื่องของคุณ? การติดตั้งสองแบบต่างกันอย่างไร
รันบน PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
เปิดจากผู้เผยแพร่
https://github.com/user-attachments/assets/6df6e217-3e5c-4691-84ed-7440701a87de
AI coding agents are great — until you try to use them across a real team. Everyone builds their own prompts and instructions, knowledge stays in silos, and the agent — not knowing your architecture or constraints — guesses from a few open files and confidently does the wrong thing.
An instruction layer for AI coding agent
That's why we built Rosetta — open-source engineering governance and context for AI coding agents. It's not another proprietary agent; it works with the tools you already use (Claude Code, Cursor, Copilot, Codex, and other MCP-compatible agents) and loads your team's shared engineering instructions into every session. Everything is versioned in Git and can run inside your perimeter.
Teach agents how to think, not what to do. The model already knows Python and React; what it lacks is your engineering discipline. That's what Rosetta encodes.
[!NOTE] If you are effectively using your current setup, writing your own skills, and managing AI using your own processes, you probably don't need Rosetta.
Rosetta-guided work follows five phases — Prepare → Research → Plan → Act → Validate — with approval gates at the key decision points.
Read more in the User Guide.
Overall Process: Install (plugin preferred) → Initialize workspace → Configure workspace ecosystem (tools, MCPs) → Proceed with Coding, Modernization, Analysis, etc.
Please follow Quick Start Guide which will walk you through the process.
Very detailed information on installation and configuration in all available modes is in Installation Guide.
Reusable units of work the AI loads on demand. Workflows pick them for you, and most can also be called by name.
Let Rosetta decide
rosetta: You want Rosetta to handle it itself. Classifies your request, picks the matching workflow, and runs it end to end.Understand what exists
load-project-context: Gives the AI your project's business context, architecture, and past decisions before any work starts.codemap: Gives the AI a structural map of the repo to navigate by. Uses your LSP or code-graph tooling when available.discovery: Evidence-backed picture of what exists today, including whether this was already attempted and done wrong.reverse-engineering: Recovers domain rules and intent from code into business specs.research: Researches any question grounded in your project, with real sources. Compares options and validates its own conclusions.Decide what to build
requirements-authoring: Precise and atomic requirement units: EARS acceptance criteria, rationale, alternatives, dependencies, traceability. You combine them into stories or features.requirements-use: Plans and builds from approved requirements. Every task, test, and result traces back to a requirement ID.design: Deep architecture decisions with defended rejections, plus domain playbooks: distributed systems, multi-tenancy, payments, regulated data, migrations.tech-specs: Produces target-state interfaces, API contracts, data models, and security considerations. Forces the AI to read the real code, grounding its assumptions.planning: An ordered graph of AI sessions for automated execution, each with its own scope, checklist, and handoff.backlog: Turns a thin ticket into a buildable one: gaps found, questions posed, facts written back. Also produces a WBS for people.reasoning: Breaks a hard problem down, identifies roles, boundaries, events, models. Works via Tree-of-Thoughts, expanding every alternative to its conclusion before committing.Build it
coding: Simple, minimal, in-scope changes following KISS/SOLID/DRY. Systematic dependency-ordered validation: database, API, web, mobile. Covers IaC.testing: Isolated and idempotent at 80%+ coverage, external calls mocked only. Scenario tests start from a sequence diagram: what depends on what, what is shared setup, then the test.debugging: OODA. Root cause with evidence before any fix, one hypothesis at a time, prevention recommended. Three failed fixes means a design problem.security: Threat modeling plus code, dependency, IaC, container, cloud, and API review. Secrets gated before ingestion, active testing only where authorized. Output is remediation tasks.natural-writing: Strips AI giveaways: hype words, em-dashes, rhetorical questions, fake engagement. For docs, emails, posts, release notes.Test automation and QA
qa-knowledge: QA engineering end to end: sources into requirements, gap analysis, Given-When-Then specs, TMS export, runnable API and UI tests, failure triage. Every case is implemented or recorded as a gap, never dropped.qa-structure: Canonical paths, slugs, and state files for QA runs. Each session in its own folder, so parallel runs never collide and a long flow knows which phase is done.data-collection: Read-only pull from issue tracker, test management, wiki, and the codebase. A ticket gets recorded, not acted on. Gaps and permission walls recorded, never guessed.Delegate the work
orchestration: Sizes the work, builds and briefs a subagent team, and gates every result through fresh eyes instead of self-review. Large plans run with phases, steps, and tasks.subagent-directives: Requested automatically by the orchestrator for every subagent: stay in scope, stop and report when blocked rather than improvise, prove each claim with links and line ranges.large-workspace-handling: Use when a repository is too big to handle at once. Splits into non-overlapping scopes, one subagent each, then a second wave verifies the work.coding-agents-farm: Runs Claude, Codex, Copilot, Gemini and others in parallel on isolated git worktrees, for throughput or cross-model validation. Burns money fast, so it is gated behind explicit consent.Keep the session on track
hitl: Makes the AI work together with you. Approval gates at specs, implementation, and closing, with review in small batches. A clear affirmative counts.questioning: A few high-impact questions before work starts, each one decision, with why it matters and a recommended answer. Anything researchable it answers itself.self-organization: Keeps work organized as it grows: splits oversized tasks and files, clears stale content, flags when a session should restart. Announces before it reorganizes.Stay safe
sensitive-data: Secrets, keys, PII, PHI, and payment data never get read, printed, logged, or committed. Masked on sight, and the same care applies to the code it writes.dangerous-actions: Before anything irreversible: blast radius first, safer alternatives offered, explicit approval required. Higher environments are off limits.risk-assessment: Checks what the AI can actually reach, such as databases, cloud accounts, and production servers. Rates the risk and blocks outright when it is critical.Recover when it goes wrong
deviation: Stops the moment work no longer matches what you asked, or the AI cannot stand behind its own solution. Escalates instead of pressing on.self-learning: After a failure or a wrong result, stops and finds the cause, then records a reusable rule in the repo's memory so it does not repeat.post-mortem: Diagnoses why a session went wrong across your prompt, workspace docs, local config, Rosetta instructions, and tooling. Can file a sanitized issue when Rosetta is at fault.Extend
harness: Builds what the AI needs to run and check your system: command-line and MCP actions, dev containers, skills, subagents, workflows, hooks, and unattended automation. Encoded once, proved on every run.coding-agents-prompt-authoring: Write, review, harden, and port your own skills, subagents, workflows, and rules, including between different IDEs and agents.coding-agents-hooks-authoring: Author, register, and debug hooks that fire on agent actions across every supported IDE.Domain packs
solr-query: Apache Solr queries that return what you meant: eDisMax, block join, JSON facets, kNN, explain output, relevancy tuning.solr-schema: Apache Solr schema and solrconfig, audit or design: field types, analyzer symmetry, docValues, synonyms, commit strategy.solr-extending: Custom Apache Solr plugins: SearchComponent, DocTransformer, QParser, update processors, function queries, and their solrconfig wiring.solr-semantic-search: Phrase-tagging semantic search on Solr: concept tagging, taxonomy, graph paths, ambiguity resolution. Lexical, not vector.specflow-use: Connects your workspace to Grid Dynamics SpecFlow through its MCP.coding-flow: AI creates features, fixes defects, and performs refactoring, everything end-to-end. AI performs discovery, design, specs and a plan, user review, then AI implements and runs separate review and validation passes (including running application). Most useful for medium to large coding tasks, and for controlled component-by-component migration/modernization work.requirements-authoring-flow: AI works with user and raw artifacts to define entire-application requirements. AI discovers context and existing constraints, captures intent, drafts atomic requirement units, validates them, and finalizes traceability artifacts. This is the most efficient use of coding agents. Requirements then Coding.security-flow: AI runs an authorized, evidence-preserving security review through mandatory specialist subagents. It gates secret-bearing files before source ingestion, bounds active testing to approved pre-production targets, independently reviews evidence, and prepares concise inputs for a later coding flow without starting remediation.testgen-flow, api-aqa-flow, ui-aqa-flow: AI handles QA-related work such as generating test cases and creating API or UI automation tests. AI first collects project context, requirements, and existing QA assets, clarifies gaps, and only after that generates test cases or automation tests.code-analysis-flow: AI creates grounded analysis documents based on the codebase. AI first loads project context, asks clarification questions, then produces either one focused analysis document or parallel module analyses plus a summary.help-flow: AI explains available Rosetta workflows, skills, and agents. Most useful when the user is unsure which Rosetta capability to use.init-workspace-flow: AI sets up a repository for AI use in both brownfield and greenfield projects. AI first analyzes the workspace, builds baseline docs, asks gap-filling questions, and verifies the result. Use it once per repository as its purpose is to build context for subsequent sessions.If you prefer to follow your own workflows, check the skills above.
Rosetta layers your instructions at runtime — core, then organization, then project, each building on the one above — and adapts the result to whatever agent you use:
flowchart LR
A["<b>Core</b><br/>best practices & workflows"] --> B["<b>Organization</b><br/>standards, policies & guardrails"]
B --> C["<b>Project</b><br/>architecture, context & constraints"]
C --> R["<b>Rosetta Runtime</b><br/>context engineering + instruction orchestration"]
R --> AD["<b>Agent Adapters</b>"]
AD --> G["<b>Guided Execution</b><br/>workflows • guardrails • validation"]
G --> T["Claude Code · Cursor · Copilot · Codex · Antigravity"]
classDef rosetta fill:#1f6feb,stroke:#1b4fb8,color:#ffffff;
classDef yours fill:#2da44e,stroke:#1a7f37,color:#ffffff;
classDef third fill:#d0d7de,stroke:#8c959f,color:#1f2328;
class A,R,AD,G rosetta
class B,C yours
class T third
Legend: 🟦 shipped with Rosetta (Core, Runtime, adapters, execution) · 🟩 authored by your org & project · ⬜ your existing AI tools (third-party).
Higher layers propagate to every project automatically; teams customize without forking. Everything is authored in markdown and versioned in Git.
IDE rules (.cursorrules, CLAUDE.md, Copilot custom instructions) are useful, but they are usually local to one tool, one repo, or one developer. Rosetta makes instructions layered, versioned, reusable, and portable across agents and IDEs — organization standards flow into every project, while project-specific context stays local and customizable. On top of that, Rosetta adds the workflows, guardrails, and approval gates that flat rules files do not provide.
AI coding agents can read code, generate code, and run commands. But that is only part of what makes software engineering reliable — they are missing much of the discipline a professional engineer brings. Each point below addresses a real, repeatedly observed failure mode, not a theoretical concern.
Why these problems exist. LLMs generate tokens sequentially from their current context. If the model passes a point where it should weigh a specific concern — security, an existing convention, an assumption it made three steps ago — it does not reliably circle back. This is not merely a temporary limitation; it is rooted in how autoregressive generation works. A coding agent's system prompt only ensures the model calls the right tools in the right format — it can't carry your project's guardrails, workflows, or quality standards, because it has no idea what you are building: a PoC, a pet project, or enterprise software with regulated data. Rosetta provides that guidance, and tells the agent how and when to load project-specific context so the model acts on it instead of skipping it.
Why this list is long. Ask any coding agent to design a full feature workflow and it will give you two or three steps — "write code," "run tests," maybe "make a plan." It rarely thinks to load context first, classify the request, assess risk, separate specs from plans, get approval before implementing, review with fresh eyes, or record lessons learned. Every point below is something agents consistently skip.
Deep project context instead of blind guessing. Without structured context, coding agents read a few line ranges around the problem and guess the rest. They do not know the architecture, the business rules, the conventions, or the dependencies. They assume. The result is code that appears correct on the surface but violates constraints the agent never knew existed. Imagine hiring a developer from outside your organization, handing them ten lines of code with zero documentation, and asking them to fix the system properly. That is how every coding agent works by default. Planning mode partially addresses this — at much higher token cost — and the agent still has to guess the purpose and target because it has no business context.
Rosetta instructions reverse this. During repository initialization, the agent — guided by Rosetta — reverse-engineers the project's architecture, tech stack, business context, coding patterns, and dependencies into structured workspace files. The agent reads these before every task. Context loads progressively — bootstrap rules first, then project context, then only the skills and workflow the current task needs. Context stays lean. Reasoning stays sharp. Token efficiency is high because the agent is not loading irrelevant material or re-discovering the project from scratch on every request.
Guardrails and enforced safe behavior. Coding agents rarely question their own actions. They do not question their understanding. They do not think about whether something is right or wrong. They just do it. They do not reliably assess what they have access to — databases, cloud services, S3 buckets. They do not handle sensitive data with care. They can copy personal data, credentials, and regulated information into logs, messages, and outputs without a second thought. They rarely evaluate whether an action is dangerous or irreversible.
Rosetta instructions require the agent to: critically review every user request before execution, assess risk of the current environment and available tools, detect and block dangerous and potentially dangerous actions, mask sensitive data and never log or share it, follow transparency rules and behavior boundaries, respect orchestration contracts between agents, and handle deviations when execution diverges from intent. These guardrails load at bootstrap and are treated as required execution constraints — not suggestions the agent can quietly skip.
Human-in-the-loop at decision points, not after the damage. AI coding agents fully and unconditionally trust user input — even when it is factually incorrect. At the same time, they almost never ask deep questions. When they do ask, the questions are shallow and few. This is the reverse of how collaboration should work. Users are biased, forget to mention critical requirements, provide information without much thought, and rely on common project knowledge that the agent does not have. Once implementation starts, the agent never stops — even when real conflicts or blockers exist in the code. It gets carried away, burns tokens, hallucinates to fill gaps, and delivers the wrong result. There are no checkpoints. There is no pause to verify understanding.
Rosetta workflows define approval gates at critical decision points: after specs, after plans, before risky actions, before test work continues. The agent batches questions (5–10 per round), prioritizes by impact, and targets a single decision per question. When something is unclear, the agent — instructed by Rosetta — stops and asks instead of guessing. It is almost always cheaper to stop and ask one question than to redo hours of wrong implementation.
Source of truth and request classification. AI does not establish or maintain a source of truth. It does the opposite — it mixes everything together, confuses its own outputs with ground truth, leaks abstractions, and blends responsibilities. It does not take time to think about systems, actors, relationships, and actions at a foundational level. On brownfield projects this is catastrophic: the agent cannot tell if the existing code is wrong, if the test is wrong, or if the user's request contradicts the actual system behavior. It just tries to make things fit.
Rosetta instructions require the agent to handle requirements with traceability. Before any work begins, the agent — following Rosetta's bootstrap — auto-classifies every request into one of thirteen workflow types: coding, security review, testing, research, requirements, initialization, modernization, code analysis, QA automation, and others. Each type loads entirely different instructions, subagents, skills, and approval gates. A "fix this bug" request follows a completely different path than "analyze this architecture" or "write requirements for checkout." Classification eliminates the guessing that agents do when they receive an unstructured prompt and try to figure out on the fly what kind of work this is.
Analysis before execution. The majority of AI coding agents are optimized to start implementation as fast as possible. This is the opposite of quality. This is the opposite of enterprise software development, where the cost of an error is extremely high. A bug caught during development costs minutes. The same bug caught after release costs the combined time of the engineer who debugs it, the lead who triages it, QA who verifies the fix, the manager who tracks it, and every person involved in the review, release, and retest cycle. Even a small bug amplifies the total cost by an order of magnitude once it escapes local development.
Rosetta workflows define explicit preparation, research, planning, and approval phases before any code is written. They instruct the agent to apply SMART, MECE, DRY, and SOLID principles during planning. They separate plans from specs — the plan says what to do and in what order; the spec says what the target state looks like and why. The process scales by task size: small tasks get lightweight planning, medium tasks get full planning with subagents, large tasks get extensive planning with heavy delegation. It is much cheaper to burn 2x tokens and spend a few extra minutes on analysis than to pay for the cascade of rework a missed defect triggers.
Review by separate agent with fresh context. AI makes mistakes. Sometimes it makes a lot of mistakes. The majority of those mistakes are trivially caught by review — but only if the reviewer has not been part of the implementation. A model reviewing its own work in the same context window rubber-stamps its own decisions. It cannot see its own blind spots. The accumulated assumptions, false starts, and iterative workarounds all feel correct because the model generated them.
Rosetta workflows instruct the agent to delegate review to a separate subagent with a fresh context window. The reviewer has never seen the debugging session, the failed attempts, or the rationalizations. It inspects the implementation against the original specs and intent. This separation is what makes review actually catch problems instead of confirming the implementer's biases.
Validation with real execution evidence. Without validation requirements, AI changes multiple files, runs nothing, and declares success. Then it spends three times the original effort trying to fix cascading failures it could have caught immediately. It builds dependent artifacts on top of broken foundations.
Rosetta instructions require the agent to build, run, and execute real tests at each foundation level before creating dependent work. The validator subagent runs in a clean context with actual execution evidence. This requirement — prove it works before moving on — is simple, and it transforms AI coding from "generate and hope" into "generate, verify, continue."
Workflows designed from observed failure modes. Ask any AI to create a complete coding workflow from scratch. It will produce something superficial — a few obvious steps that cover maybe 20% of what actually matters. It will focus on one or two concerns and completely forget about everything else. This is not a failure of intelligence. It is a failure of experience. The model has never watched itself fail across hundreds of real tasks and identified the patterns.
Rosetta contains workflows created by humans who used AI extensively, observed every category of failure, identified root causes, and encoded solutions as structured processes. These workflows cover thirteen SDLC activities. Each defines phases, subagents, skills, HITL gates, and artifact expectations. The agent with Rosetta workflows does not become smarter — it stops skipping the steps that matter. It discovers knowledge, conventions, and dependencies it would otherwise miss entirely. It installs the package that another project in the same solution already uses. It distinguishes planning from specs. It performs reviews and checkpoints at the moments where they catch the most errors.
Self-learning and self-organization. AI coding agents are only now getting basic memory features, but self-learning is not just memory. Self-organization is equally important. AI is fully capable of reorganizing files, restructuring its approach, cleaning up stale information, and adapting based on past mistakes — but it does not do any of this because it was never instructed to. It treats reorganization as deviation from the task. It treats cleanup as out of scope. It treats learning as someone else's job.
Rosetta instructs the agent to maintain agents/MEMORY.md — root causes of errors, actions tried, lessons learned. The agent consults this during planning and records new lessons after failures. It is instructed to reorganize working files when context grows large, and to proactively clean up when work spans many files or sessions.
State persistence turns crashes into checkpoints. AI coding sessions are fragile. Context loss, timeout, or a crash means starting over. For anything beyond a small fix, this wastes significant time and money. The agent has no memory of what it already completed.
Rosetta instructs the agent to write execution state — plans, specs, phase progress, flow status — to disk files. If a session fails, the next session resumes from the last recorded checkpoint. Medium and large tasks become resumable multi-session workflows instead of all-or-nothing gambles.
Security by design — no source code leaves your perimeter. Instruction delivery is deterministic: the agent requests content by tag, not by sending source code for analysis. There is no semantic search over your codebase. No code transfers to Rosetta servers. Write mode is disabled by default and requires explicit deployment configuration to enable. Schema-strict input validation rejects any unexpected payloads. The architecture works with limited internet access and runs entirely inside your organization's perimeter.
One system, every AI tool, customizable at every level. Rosetta works across Cursor, Claude Code, VS Code, JetBrains, Windsurf, Codex, Antigravity, OpenCode, and any MCP-compatible IDE. Instructions are written once and adapt to each environment. Organizations that switch between AI tools or use multiple tools simultaneously keep their entire instruction investment intact. No vendor lock-in. No per-tool maintenance.
Three layers merge at runtime: core (universal best practices shipped with Rosetta), organization (your company's conventions and policies), and project (local constraints and context). Teams customize without forking. Improvements to higher layers propagate to every project automatically. Instructions are versioned in place: R3 is the current release and evolves through incremental, reviewed updates without breaking agents on it. Rollback is immediate. AI behavior is authored in markdown, version-controlled in Git, reviewed in pull requests, and approved before deployment — the same engineering rigor applied to the instructions that control your AI agents.
Use Rosetta plugins to develop Rosetta.
Contributions welcome. See CONTRIBUTING.md for workflow and expectations.
[!WARNING] Rosetta is intended for legitimate software engineering workflows. Users are responsible for ensuring their use complies with applicable laws, regulations, and contractual obligations.
Full information: https://griddynamics.github.io/rosetta/llms-full.txt
See LICENSE for details.
An instruction layer for AI coding agent
เครื่องมือ
ยังไม่มีรายการเครื่องมือ เซิร์ฟเวอร์แบบรีโมตจะได้รายการหลังการตรวจสอบสถานะครั้งแรก
FAQ
คัดลอกคอนฟิก stdio สำหรับ Cursor, Claude Desktop, Claude Code หรือ VS Code ได้ในหน้านี้ คอนฟิกนี้ใช้การติดตั้งของผู้เผยแพร่ ไม่ใช่พร็อกซีของ PlugMCP
ระเบียนนี้ไม่ได้ระบุตัวแปรสภาพแวดล้อมที่จำเป็น
Rosetta ทำงานแบบโลคัลผ่าน stdio
ตรวจสอบล่าสุดเมื่อ 2026-11-13 An instruction layer for AI coding agent