[Plugin Generator] Split core into composable plugin sets (#315)

.agents/plugins/marketplace.json

@@ -1,5 +1,6 @@
 {
   "name": "rosetta",
+  "version": "3.1.13",
   "interface": {
     "displayName": "Rosetta"
   },
@@ -8,7 +9,7 @@
       "name": "rosetta",
       "source": {
         "source": "local",
-        "path": "./plugins/core-codex"
+        "path": "./plugins/rosetta-codex"
       },
       "policy": {
         "installation": "AVAILABLE",
@@ -20,7 +21,67 @@
       "name": "rosetta-light",
       "source": {
         "source": "local",
-        "path": "./plugins/core-codex-light"
+        "path": "./plugins/rosetta-codex-light"
+      },
+      "policy": {
+        "installation": "AVAILABLE",
+        "authentication": "ON_INSTALL"
+      },
+      "category": "Productivity"
+    },
+    {
+      "name": "core",
+      "source": {
+        "source": "local",
+        "path": "./plugins/core-codex"
+      },
+      "policy": {
+        "installation": "AVAILABLE",
+        "authentication": "ON_INSTALL"
+      },
+      "category": "Productivity"
+    },
+    {
+      "name": "workflows",
+      "source": {
+        "source": "local",
+        "path": "./plugins/workflows-codex"
+      },
+      "policy": {
+        "installation": "AVAILABLE",
+        "authentication": "ON_INSTALL"
+      },
+      "category": "Productivity"
+    },
+    {
+      "name": "qe",
+      "source": {
+        "source": "local",
+        "path": "./plugins/qe-codex"
+      },
+      "policy": {
+        "installation": "AVAILABLE",
+        "authentication": "ON_INSTALL"
+      },
+      "category": "Productivity"
+    },
+    {
+      "name": "search",
+      "source": {
+        "source": "local",
+        "path": "./plugins/search-codex"
+      },
+      "policy": {
+        "installation": "AVAILABLE",
+        "authentication": "ON_INSTALL"
+      },
+      "category": "Productivity"
+    },
+    {
+      "name": "modernization",
+      "source": {
+        "source": "local",
+        "path": "./plugins/modernization-codex"
       },
       "policy": {
         "installation": "AVAILABLE",

.claude-plugin/marketplace.json

@@ -11,16 +11,38 @@
   "plugins": [
     {
       "name": "rosetta",
-      "source": "./plugins/core-claude",
+      "source": "./plugins/rosetta-claude",
       "description": "Rosetta for Claude - Software Engineering Accelerator instruction set, workflows, and guardrails."
     },
     {
       "name": "rosetta-light",
-      "source": "./plugins/core-claude-light",
+      "source": "./plugins/rosetta-claude-light",
       "description": "Rosetta for Claude - Software Engineering Accelerator instruction set, workflows, and guardrails. (lightweight profile: simpler workflows, smaller models)"
+    },
+    {
+      "name": "core",
+      "source": "./plugins/core-claude",
+      "description": "Rosetta for Claude - core set: composable skills, always-on rules, bootstrap and guardrail hooks; required by every other Rosetta plugin."
+    },
+    {
+      "name": "workflows",
+      "source": "./plugins/workflows-claude",
+      "description": "Rosetta for Claude - workflows set: subagents, workflows, and orchestration automation. Requires the `core` plugin."
+    },
+    {
+      "name": "qe",
+      "source": "./plugins/qe-claude",
+      "description": "Rosetta for Claude - qe set: test automation and test generation. Requires the `core` and `workflows` plugins."
+    },
+    {
+      "name": "search",
+      "source": "./plugins/search-claude",
+      "description": "Rosetta for Claude - search set: Solr and search engineering. Requires the `core` and `workflows` plugins."
+    },
+    {
+      "name": "modernization",
+      "source": "./plugins/modernization-claude",
+      "description": "Rosetta for Claude - modernization set: conversion, modernization, and re-architecture. Requires the `core` and `workflows` plugins."
     }
-  ],
-  "renames": {
-    "core": "rosetta"
-  }
+  ]
 }

.claude/skills/sync-web-site/SKILL.md

@@ -134,7 +134,7 @@ If you learned something new which is reusable, there are process efficiency imp
 - **Root docs sometimes carry trailing AI-agent-imperative sections** (`MUST`, `DO NOT FILTER`, validation scripts) mixed into otherwise human-facing prose. Treat these as intentionally contributor/agent-only — never sync them to the public web page unless explicitly told to.
 - **Verify surprising claims against source code, don't just trust the doc.** Cross-checking `submit_feedback` against `server.py` found the tool's `@mcp.tool` decorator commented out (permanently disabled) while both root and web docs still list it as live. Flag findings like this as asides in the report; don't silently "fix" facts that are outside the sync's approved scope.
 - **r2 = shipped/current, r3 = in-dev.** Plugins under `plugins/core-*` ship the r2 model (five `bootstrap-*` files, classify-and-route). Root `docs/ARCHITECTURE.md` was rewritten to the r3 model (`bootstrap-alwayson.md` + one mode file, typed load aliases, three delivery modes MCP/plugin/local) while its tree/datasets still say r2. This mixed r2-tree / r3-content state is intentional transitional narrative — mirror root faithfully; do NOT "fix" the r2/r3 split in the web doc (that's a root-internal decision).
-- **Verify a flagged aside is actually a defect before "fixing" it.** In one run, 3 of 4 flagged asides were NOT broken: `ACQUIRE <path> FROM KB` is retained in the new alias table as the MCP-only raw shell form (not stale); installation's "read all five `bootstrap-*` files" matched the actually-shipped copilot plugin (`ls plugins/core-copilot*/…/*bootstrap*`); the r2/r3 tree faithfully mirrored root. Only the FAQ prep-step wording was real. Discriminator: `git log -S "<phrase>"` — if the wording predates the model change it's genuine staleness; if it was re-authored in the model-change commit it's intentional. Fix real drift in BOTH root + web (release-agnostic phrasing) so the next sync doesn't re-flag it.
+- **Verify a flagged aside is actually a defect before "fixing" it.** In one run, 3 of 4 flagged asides were NOT broken: `ACQUIRE <path> FROM KB` is retained in the new alias table as the MCP-only raw shell form (not stale); installation's "read all five `bootstrap-*` files" matched the actually-shipped copilot plugin (`ls plugins/rosetta-copilot*/…/*bootstrap*`); the r2/r3 tree faithfully mirrored root. Only the FAQ prep-step wording was real. Discriminator: `git log -S "<phrase>"` — if the wording predates the model change it's genuine staleness; if it was re-authored in the model-change commit it's intentional. Fix real drift in BOTH root + web (release-agnostic phrasing) so the next sync doesn't re-flag it.
 - **New typed load aliases (r3):** `USE/READ SKILL`, `READ/APPLY SKILL FILE`, `USE/READ FLOW`, `APPLY PHASE`, `INVOKE/READ SUBAGENT`, `READ/APPLY RULE`, `READ TEMPLATE`, `READ CONFIGURE`, `LIST <path>`. Verbs: `READ`=load, `APPLY`=load+execute, `USE`/`INVOKE`=activate. Old `GET PREP STEPS`/`SEARCH … IN KB`/`STORE`/`… ABOUT <project>` aliases are gone; `ACQUIRE … FROM KB` survives only as the MCP-raw shell form. The web doc is MCP-centric, so when porting r3 content add a one-line framing of the three delivery modes in place (plugin/local are new vocabulary for a web reader).
 - **Local Jekyll preview:** first run needs `bundle install` (gems aren't vendored in the repo). Then `cd docs/web && bundle exec jekyll serve --detach --port <port>`. Fetch pages with `curl -s http://127.0.0.1:<port>/rosetta/docs/<page>/`. Stop with `pkill -f "jekyll serve"`. Sanity-check rendered HTML for the new section names and for `<table>` counts (a missed blank line before a markdown table renders as a literal `|`-delimited paragraph, not a table).
 - **Vendored bundle can be stale even when `docs/web/vendor/bundle/` exists.** Seen: `bundle exec jekyll serve` fails with `Bundler::GemNotFound` — `Gemfile.lock` pins versions (addressable, json, public_suffix, concurrent-ruby) not present in the vendored dir. Fix is `bundle install`, which rewrites vendored gems + the lockfile — a dependency change, out of scope for a pure doc sync and not to be run autonomously. For a trivial edit (heading rename, a `<video>` embed that copies the form of an existing sibling, no new tables), static validation is enough: re-read the edited region and confirm blank-line spacing around headings/tags. Reserve the server run for changes that touch tables or new page structure.

.cursor-plugin/marketplace.json

@@ -11,13 +11,38 @@
   "plugins": [
     {
       "name": "rosetta",
-      "source": "plugins/core-cursor",
+      "source": "plugins/rosetta-cursor",
       "description": "Rosetta for Cursor - Software Engineering Accelerator instruction set, workflows, and guardrails."
     },
     {
       "name": "rosetta-light",
-      "source": "plugins/core-cursor-light",
+      "source": "plugins/rosetta-cursor-light",
       "description": "Rosetta for Cursor - Software Engineering Accelerator instruction set, workflows, and guardrails. (lightweight profile: simpler workflows, smaller models)"
+    },
+    {
+      "name": "core",
+      "source": "plugins/core-cursor",
+      "description": "Rosetta for Cursor - core set: composable skills, always-on rules, bootstrap and guardrail hooks; required by every other Rosetta plugin."
+    },
+    {
+      "name": "workflows",
+      "source": "plugins/workflows-cursor",
+      "description": "Rosetta for Cursor - workflows set: subagents, workflows, and orchestration automation. Requires the `core` plugin."
+    },
+    {
+      "name": "qe",
+      "source": "plugins/qe-cursor",
+      "description": "Rosetta for Cursor - qe set: test automation and test generation. Requires the `core` and `workflows` plugins."
+    },
+    {
+      "name": "search",
+      "source": "plugins/search-cursor",
+      "description": "Rosetta for Cursor - search set: Solr and search engineering. Requires the `core` and `workflows` plugins."
+    },
+    {
+      "name": "modernization",
+      "source": "plugins/modernization-cursor",
+      "description": "Rosetta for Cursor - modernization set: conversion, modernization, and re-architecture. Requires the `core` and `workflows` plugins."
     }
   ]
 }

.cursorignore

@@ -1,4 +1,5 @@
-/plugins/core-cursor/
-!/plugins/core-cursor/.cursor-plugin/plugin.json
-/plugins/core-claude/
-!/plugins/core-claude/.claude-plugin/plugin.json
+# Generated plugin output: <set>-<ide>[-light] for every set (rosetta, core, workflows,
+# qe, search, modernization) — ~49 trees. Ignore them all, keep the manifests indexable.
+/plugins/*/
+!/plugins/*/.cursor-plugin/plugin.json
+!/plugins/*/.claude-plugin/plugin.json

.github/actions/codex-setup/action.yml

@@ -1,7 +1,7 @@
 name: Rosetta Codex setup
 description: >-
   Prepares a Codex run for the Rosetta pipelines: a TRUSTED Codex home (permission
-  profiles + the hooks feature) and, optionally, the Rosetta `core-codex` plugin
+  profiles + the hooks feature) and, optionally, the Rosetta `rosetta-codex` plugin
   extracted into the workspace. Output `codex-home` feeds `openai/codex-action`.
 
 # Why this action exists
@@ -44,11 +44,11 @@ description: >-
 
 inputs:
   install-plugin:
-    description: "Extract the Rosetta core-codex plugin into the workspace so the agent boots with the Rosetta instruction set."
+    description: "Extract the Rosetta rosetta-codex plugin into the workspace so the agent boots with the Rosetta instruction set."
     required: false
     default: "true"
   release-tag:
-    description: "Rosetta release to take the core-codex asset from. Empty means the latest release."
+    description: "Rosetta release to take the rosetta-codex asset from. Empty means the latest release."
     required: false
     default: ""
   github-token:
@@ -82,7 +82,7 @@ runs:
         echo "Trusted Codex home: $home"
         cat "$home/config.toml"
 
-    - name: Install Rosetta core-codex plugin
+    - name: Install Rosetta rosetta-codex plugin
       if: inputs.install-plugin == 'true'
       shell: bash
       env:
@@ -98,10 +98,10 @@ runs:
         # variant, which is a different (reduced) instruction set.
         if [ -n "$RELEASE_TAG" ]; then
           gh release download "$RELEASE_TAG" --repo griddynamics/rosetta \
-            --pattern 'core-codex-[0-9]*.zip' --dir "$dl" --clobber
+            --pattern 'rosetta-codex-[0-9]*.zip' --dir "$dl" --clobber
         else
           gh release download --repo griddynamics/rosetta \
-            --pattern 'core-codex-[0-9]*.zip' --dir "$dl" --clobber
+            --pattern 'rosetta-codex-[0-9]*.zip' --dir "$dl" --clobber
         fi
 
         # One release can carry several dated builds of the same version and `gh`
@@ -112,9 +112,9 @@ runs:
         # `sort -r` would rank 3.1.9 above 3.1.13.
         # `|| true` so an empty match reports the message below instead of `set -e`
         # aborting on the glob.
-        zip=$(ls -1 "$dl"/core-codex-[0-9]*.zip 2>/dev/null | sort -Vr | head -n1 || true)
-        [ -n "$zip" ] || { echo "::error::no core-codex asset found in the release"; exit 1; }
-        echo "Rosetta core-codex asset: $(basename "$zip")"
+        zip=$(ls -1 "$dl"/rosetta-codex-[0-9]*.zip 2>/dev/null | sort -Vr | head -n1 || true)
+        [ -n "$zip" ] || { echo "::error::no rosetta-codex asset found in the release"; exit 1; }
+        echo "Rosetta rosetta-codex asset: $(basename "$zip")"
 
         unzip -oq "$zip" -d "$GITHUB_WORKSPACE"
 
@@ -125,7 +125,7 @@ runs:
         if [ -f "$exclude" ]; then
           {
             echo ""
-            echo "# Rosetta core-codex plugin, extracted by .github/actions/codex-setup"
+            echo "# Rosetta rosetta-codex plugin, extracted by .github/actions/codex-setup"
             echo "/.agents/"
             echo "/.codex/"
             echo "/.codex-plugin/"
@@ -136,7 +136,7 @@ runs:
           || { echo "::error::plugin extracted but .codex/hooks.json is missing"; exit 1; }
         test -f "$GITHUB_WORKSPACE/.agents/rules/bootstrap-alwayson.md" \
           || { echo "::error::plugin extracted but .agents/rules/bootstrap-alwayson.md is missing"; exit 1; }
-        echo "Rosetta core-codex plugin installed at the workspace root."
+        echo "Rosetta rosetta-codex plugin installed at the workspace root."
 
     - name: Install enforced policy
       shell: bash

.github/actions/codex-setup/files/config.toml

@@ -8,7 +8,7 @@
 # experimental features are not enabled in this pipeline. Same egress posture as the
 # Claude branch, which restricts the `gh` command surface rather than the network.
 
-# On-disk effect of `codex features enable hooks`; the Rosetta core-codex plugin
+# On-disk effect of `codex features enable hooks`; the Rosetta rosetta-codex plugin
 # ships .codex/hooks.json and is inert without it.
 [features]
 hooks = true

.github/plugin/marketplace.json

@@ -12,13 +12,38 @@
   "plugins": [
     {
       "name": "rosetta",
-      "source": "core-copilot",
+      "source": "rosetta-copilot",
       "description": "Rosetta for Copilot - Software Engineering Accelerator instruction set, workflows, and guardrails. Includes Rosetta MCP for knowledge base access."
     },
     {
       "name": "rosetta-light",
-      "source": "core-copilot-light",
+      "source": "rosetta-copilot-light",
       "description": "Rosetta for Copilot - Software Engineering Accelerator instruction set, workflows, and guardrails. Includes Rosetta MCP for knowledge base access. (lightweight profile: simpler workflows, smaller models)"
+    },
+    {
+      "name": "core",
+      "source": "core-copilot",
+      "description": "Rosetta for Copilot - core set: composable skills, always-on rules, bootstrap and guardrail hooks; required by every other Rosetta plugin."
+    },
+    {
+      "name": "workflows",
+      "source": "workflows-copilot",
+      "description": "Rosetta for Copilot - workflows set: subagents, workflows, and orchestration automation. Requires the `core` plugin."
+    },
+    {
+      "name": "qe",
+      "source": "qe-copilot",
+      "description": "Rosetta for Copilot - qe set: test automation and test generation. Requires the `core` and `workflows` plugins."
+    },
+    {
+      "name": "search",
+      "source": "search-copilot",
+      "description": "Rosetta for Copilot - search set: Solr and search engineering. Requires the `core` and `workflows` plugins."
+    },
+    {
+      "name": "modernization",
+      "source": "modernization-copilot",
+      "description": "Rosetta for Copilot - modernization set: conversion, modernization, and re-architecture. Requires the `core` and `workflows` plugins."
     }
   ]
 }

.github/prompts/repo-analysis.md

@@ -46,7 +46,7 @@ MUST read `instructions/r3/core/skills/coding-agents-prompt-authoring/references
   - Terse/compressed phrasing, abbreviations, and shortcuts are **intentional** (every loaded token is paid for on every turn) — do not flag them as unclear writing.
   - File paths referenced inside `instructions/**` (e.g. `docs/CONTEXT.md`, `agents/MEMORY.md`) describe the **target repo's** structure, not this repo's — they are not broken links here (except the init-workspace workflow, which creates/upgrades that structure).
   - Releases: `r3` is active (incremental updates), `r2` is backport-only, earlier releases get no new work. Default any new instruction work to `r3` unless told otherwise.
-  - Edits under `instructions/r3/**` ripple into the generated plugin directories (`plugins/core-claude/`, `plugins/core-cursor/`, etc. via the plugin generator) — flag this as a follow-up dependency in the issue body, don't treat it as an isolated file edit.
+  - Edits under any of the five instruction roots (`instructions/r3/{core,workflows,qe,search,modernization}/**`) ripple into the generated plugin directories (`plugins/rosetta-claude/`, `plugins/core-claude/`, `plugins/qe-cursor/`, etc. — one `<set>-<ide>` folder per set x IDE, via the plugin generator) — flag this as a follow-up dependency in the issue body, don't treat it as an isolated file edit.
 
 AI Coding Agents use MCP to load bootstrap instructions `instructions/r3/core/rules/bootstrap-*.md` as the first thing (exactly the same you have loaded too).
 After that AI Coding Agent is instructed to follow one workflow and to load skills/agents/rules when needed.

.github/prompts/repo-implement-code.md

@@ -74,7 +74,7 @@ MUST read docs/CONTEXT.md and docs/ARCHITECTURE.md.
 
 **Two different mental models in this repo — check which one the issue is in before implementing:**
 - `src/` (rosettify, rosetta-mcp-server, rosetta-cli, ims-mcp-server, hooks, helm-charts) is a **normal software project**. Ordinary engineering judgment applies.
-- `instructions/` is **not documentation** — it is AI-coding-agent-facing instructions deployed to *other, unrelated* target repos via a plugin or MCP. Terse/compressed phrasing is intentional (token cost), not a defect — do not "clean up" it toward human-readable prose. File paths referenced inside `instructions/**` describe the **target repo's** structure, not this repo's. `r3` is active, `r2` is backport-only. Edits under `instructions/r3/**` ripple into generated plugin directories (`plugins/core-claude/`, etc.) — if the plan didn't already call this out, flag it explicitly in the PR description as a manual follow-up.
+- `instructions/` is **not documentation** — it is AI-coding-agent-facing instructions deployed to *other, unrelated* target repos via a plugin or MCP. Terse/compressed phrasing is intentional (token cost), not a defect — do not "clean up" it toward human-readable prose. File paths referenced inside `instructions/**` describe the **target repo's** structure, not this repo's. `r3` is active, `r2` is backport-only. Edits under any of the five instruction roots (`instructions/r3/{core,workflows,qe,search,modernization}/**`) ripple into generated plugin directories (one `plugins/<set>-<ide>/` folder per set x IDE, e.g. `plugins/rosetta-claude/`, `plugins/core-claude/`, `plugins/qe-cursor/`) — if the plan didn't already call this out, flag it explicitly in the PR description as a manual follow-up.
 - **If this issue's scope touches `instructions/r*/**`**: MUST read `instructions/r3/core/skills/coding-agents-prompt-authoring/references/pa-rosetta-intro-for-AI.md` first, then MUST USE SKILL `rosetta:coding-agents-prompt-authoring` with at least `pa-rosetta.md`, `pa-patterns.md`, `pa-hardening.md`, `pa-schemas.md` before making any edit.
 
 AI Coding Agents use MCP to load bootstrap instructions `instructions/r3/core/rules/bootstrap-*.md` as the first thing (exactly the same you have loaded too).