Skip to content

docs(README): sharpen positioning, add architecture diagram, concrete quickstart#123

Open
esteinberg wants to merge 3 commits into
mainfrom
claude/distracted-chatelet-b4a1db
Open

docs(README): sharpen positioning, add architecture diagram, concrete quickstart#123
esteinberg wants to merge 3 commits into
mainfrom
claude/distracted-chatelet-b4a1db

Conversation

@esteinberg

Copy link
Copy Markdown
Member

What

A focused editorial pass on README.md to improve time-to-understanding for first-time GitHub visitors. No code or behavior changes — content only.

Positioning & vocabulary

  • Anchored on a consistent category: context engineering + governed instructions (hero + intro), retiring the ambiguous compiled / "governance and context layer" phrasings.
  • Standardized "instructions" as the umbrella term; kept "meta-prompting" out of the user-facing hero (reserved for About/metadata).

New fast-comprehension pieces

  • Without Rosetta / With Rosetta table near the top.
  • Concrete "Add rate limiting to the checkout API" side-by-side example.
  • How it works Mermaid diagram — horizontal, color-coded by ownership (🟦 shipped with Rosetta · 🟩 your org & project · ⬜ third-party tools), showing inputs → Rosetta Runtime → Agent Adapters → Guided Execution → agents.
  • "Why not just use IDE rules?" objection handler.

Quick Start made real

  • Install options table: Plugins (recommended) / Hosted MCP (evaluation) / Self-hosted MCP (enterprise & air-gapped).
  • The actual conversational init prompt and the workspace files it generates (TECHSTACK.md, CODEMAP.md, DEPENDENCIES.md, ARCHITECTURE.md, CONTEXT.md).
  • Removed fabricated pip install rosetta-cli / rosetta init guidance (verified rosetta-cli is a RAGFlow publishing/admin tool, not the entry point).
  • Elevated llms-full.txt as a machine-readable entry point.

Structure & tone

  • Reordered: Quick Start now precedes How it works; "Why use it" precedes "For organizations" (developer-first, then enterprise).
  • Fixed manual validation by AIindependent AI validation backed by execution evidence.
  • Softened a few over-absolute claims in the deep-dive for enterprise credibility; made the daily-work bullets consistent and outcome-led.

Why

The README's content was strong but front-loaded with explanation; this trades some prose for visuals, a concrete example, and a faster trial path.

Reviewer notes

  • Please preview the Mermaid diagram on GitHub — confirm the color classes and legend render legibly in both light and dark themes (third-party gray especially).
  • Follow-ups intentionally left out of this PR: syncing the same "validate" phase wording in docs/web/docs/introduction.md, and extending the canonical category line to GitHub About / pyproject descriptions / site metadata.

🤖 Generated with Claude Code

…nd example

Improve time-to-understanding without changing the underlying content:

- Reframe positioning on "context engineering + governed instructions"
  consistently (hero, intro), retiring the ambiguous "compiled"/"governance
  and context layer" phrasings
- Add "Without Rosetta / With Rosetta" table and a concrete
  "Add rate limiting" side-by-side example near the top
- Add a "How it works" Mermaid diagram (horizontal, color-coded by
  ownership: shipped-with-Rosetta / yours / third-party) showing
  inputs -> Rosetta Runtime -> Agent Adapters -> Guided Execution -> agents
- Make Quick Start concrete: real install options table (Plugins /
  Hosted MCP / Self-hosted MCP), the actual conversational init prompt and
  the workspace files it generates; drop fabricated CLI guidance
- Add "Why not just use IDE rules?" objection handler
- Reorder so Quick Start precedes How it works; "Why use it" precedes
  "For organizations" (developer-first, then enterprise)
- Fix "manual validation by AI" -> "independent AI validation backed by
  execution evidence"; soften over-absolute claims; make daily-work
  bullets consistent and outcome-led; elevate llms-full.txt

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@esteinberg

Copy link
Copy Markdown
Member Author

Follow-up work (if this PR is accepted)

This PR was deliberately scoped to README.md only. To finish the job, these follow-ups should be done in separate changes:

1. Verify the Mermaid diagram renders

  • Open the Files changed tab and confirm the color-coded "How it works" diagram renders legibly in both light and dark GitHub themes (the third-party gray especially). Adjust classDef hues if washed out.

2. Propagate the canonical positioning line everywhere

The README hero now stays outcome-driven, but the canonical category line — "Meta-prompting, context engineering, and centralized instruction management for AI coding agents" — should be made consistent across the other surfaces, which currently differ:

  • GitHub repo About / description
  • src/*/pyproject.toml package descriptions (ims-mcp, rosetta-cli)
  • docs/web site <meta> description and docs intro

3. Sync duplicated phase wording

  • docs/web/docs/introduction.md:31 still reads Validate (manual validation by AI). Update it to match the README's independent AI validation backed by execution evidence so the phase description is consistent site-wide.

4. Consider a dedicated comparison page

We deliberately kept the README lean and used only a short "Why not just use IDE rules?" section. A fuller comparison vs .cursorrules / CLAUDE.md / Copilot custom instructions / internal wiki could live on a separate docs/ page rather than the top-level README.

🤖 Generated with Claude Code

@github-actions github-actions Bot added the documentation Improvements or additions to documentation label Jun 29, 2026
@github-actions

Copy link
Copy Markdown
Contributor

Rosetta Triage Review

Summary: This PR is a focused editorial pass on README.md to improve time-to-understanding for first-time GitHub visitors — adding a Without/With comparison table, a concrete side-by-side example, a Mermaid architecture diagram, and a cleaner Quick Start section. No code or behavioral changes.

Findings:

  • Documentation-only change (single file: README.md); no code paths affected, no tests needed.
  • Positioning reframed from "governance and context layer" to "context engineering and governed instructions" — more accurate and consistent with project vocabulary.
  • Mermaid diagram uses explicit fill, stroke, and color classDef properties, aligned with the project guideline for readable colors in both light and dark themes. Reviewer attention flagged by the author — verify the light gray (#d0d7de) third-party node contrast in light mode.
  • Hardcoded hosted MCP URL (https://mcp.rosetta.griddynamics.net/mcp) added to Quick Start — intentional for public documentation; confirm this is the stable production endpoint.
  • The concrete "rate limiting" example and Without/With table improve scanability and reduce time-to-first-value for new visitors.
  • PR description is detailed and well-structured (What / Why / Reviewer notes).

Suggestions:

  • Consider whether the follow-up items noted in the PR body (syncing docs/web/docs/introduction.md and updating GitHub About / pyproject descriptions) should be filed as separate issues to avoid losing track.
  • Verify the Mermaid diagram renders correctly in GitHub's light and dark themes before merging, as the author requested.

Automated triage by Rosetta agent

@omaiesh omaiesh requested a review from isolomatov-gd July 1, 2026 15:27
omaiesh
omaiesh previously approved these changes Jul 1, 2026

@isolomatov-gd isolomatov-gd left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We need more work on this one, we added more text, I'd like to think how we can either reduce it or move it.
I like idea and changes to subheader a lot.

Some things needs to be reverted: **Teach agents how to think, not what to do.** section lost its value and just a duplicate of the subheader.

I am not sure about having init workflow inside of the readme, but I really like the diagram. Maybe we should show like that our quickstart process and we can use similar to show few workflows.

Address review feedback on length and duplication:

- Revert hero tagline to "Teach agents how to think, not what to do"
  (the "Encode your engineering discipline" variant duplicated the subheader)
- Remove the redundant "daily work" bullets and collapse the 5-line phases
  list into a single line — the Without/With + Example tables and the
  How-it-works diagram already convey this
- Replace the verbose Quick Start init instructions with a compact process
  diagram (ask → analyze → workspace files → build); full steps stay in
  QUICKSTART.md

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@esteinberg

Copy link
Copy Markdown
Member Author

Update: trimmed length + init-as-diagram

Pushed a follow-up commit addressing length/duplication feedback:

  • Reverted the hero tagline to "Teach agents how to think, not what to do" — the "Encode your engineering discipline" variant duplicated the subheader.
  • Cut redundancy in the intro: removed the 7 "daily work" bullets and collapsed the 5-line phases list into one line. The Without/With + Example tables and the How-it-works diagram already carry this, so the run-up to Quick Start is ~18 lines shorter.
  • Quick Start init is now a process diagram (ask → Rosetta analyzes → workspace files → build), with the full step-by-step kept in QUICKSTART.md rather than inline.

Added to follow-up backlog

  • Workflow diagrams: extend the same visual style to show a few key workflows (e.g. coding, AQA) — deferred to a separate PR to keep this one focused.

…lumns

Merge the separate "Why use it" and "For organizations" lists into one
two-column table (For builders | For organizations). Re-buckets bullets by
audience, dedupes the guardrails/governance overlap, and reclaims the
plain-language-tasks and less-babysitting points cut from the intro.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@esteinberg esteinberg self-assigned this Jul 2, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

documentation Improvements or additions to documentation needs more work

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants