Move plan-manager as a script

agents/IMPLEMENTATION.md

@@ -53,6 +53,14 @@ For detailed change history, use git history and PRs instead of expanding this f
 - A dedicated `version` command was added so package version inspection does not require config loading or auth.
 - Package metadata and publish flows were repaired to keep CI/CD and PyPI publishing functional.
 
+### Instructions and Skills
+
+- Added `plan-manager` skill under `instructions/r2/core/skills/plan-manager/` — primary plan manager for coding agents via local JSON files.
+- Skill assets: `plan_manager.js` (CLI, no npm deps), `pm-schema.md` (data structure reference), `plan_manager.test.js` (60 unit tests).
+- Key behaviors: resume-safe `next` command returns `in_progress` steps with `resume: true` before `open` steps; plans stored at `plans/<name>/plan.json`; self-describing `help` command.
+- Converted `adhoc-flow-with-plan-manager` workflow to `USE SKILL plan-manager`; data structure externalized to `pm-schema.md`.
+- Plugins (`core-claude`, `core-cursor`) are auto-synced from core by `scripts/pre_commit.py`.
+
 ### Workflows and Automation
 
 - GitHub Actions were updated to remove most deprecated Node 20-era dependencies and align with newer action runtimes where upstream allowed it.

docs/ARCHITECTURE.md

@@ -423,6 +423,15 @@ cp .env.dev .env
 uvx rosetta-cli@latest publish instructions
 ```
 
+### Plugins
+
+Instructions to `plugins` folder content must be copied with `venv/bin/python scripts/pre_commit.py` as it not only copies, but also adapts.
+Pre-commit hook is also created, but we must not rely on it.
+Do not directly modify instructions in `plugins` folder instead edit original files in `instructions` and use script to copy/adapt.
+
+Claude Code Plugin: only Anthropic `sonnet`/`opus`/`haiku` models are supported.
+Codex Plugin: only OpenAI `gpt-*` models are supported.
+
 ### Reference Sources (readonly, packages currently used)
 
 `refsrc/fastmcp-3.1.1` contains source code of FastMCP v3.
@@ -472,8 +481,8 @@ Website: builds the Jekyll website from `docs/web/`, deploys to GitHub Pages.
 | Plugin | Contents, Footprint |
 |---|---|
 | `core@rosetta` | Full OSS foundation |
-| `grid@rosetta` | Enterprise extensions |
-| `rosetta@rosetta` | Bootstrap rule + MCP definition only, (fetches via MCP) |
+| `grid@rosetta-enterprise` | Enterprise extensions |
+| `rosetta@rosetta` | Bootstrap rule + MCP only |
 
 Plugins point to source folders in the instructions repository. No local file duplication.
 

docs/definitions/skills.md

@@ -4,6 +4,7 @@
 - research
 - context-engineering
 - planning
+- plan-manager
 - reasoning
 - questioning
 - tech-specs

instructions/r2/core/rules/bootstrap-core-policy.md

@@ -83,7 +83,7 @@ baseSchema: docs/schemas/rule.md
 
 ### Input Contract
 
-4. Subagent prompt MUST start with: assumed role/specialization, stated [lightweight|full] subagent, plan_name, phase&task id, SMART tasks, `MUST USE SKILL [required]`, and `RECOMMEND USE SKILL [recommended]`.
+4. Subagent prompt MUST start with: assumed role/specialization, stated [lightweight|full] subagent, full path to plan.json, phase&task id, SMART tasks, `MUST USE SKILL [required]`, and `RECOMMEND USE SKILL [recommended]`.
 5. Provide specific task, full context, and references. Subagents know nothing except shared bootstrap and prep steps and this contract, always provide original user request/intent throughout all steps.
 6. Define explicit scope, expected outputs, and clear expectations. Forbid out-of-scope work.
 7. Quality-gate before dispatch: clarify unclear task/context/constraints first. Never dispatch ambiguous instructions.

instructions/r2/core/rules/bootstrap-rosetta-files.md

@@ -26,7 +26,8 @@ It must be possible to grep by headers and receive useful information and ToC.
 12. `agents/MEMORY.md`. Very brief root causes of errors and mistakes, brief actions tried and actions succeeded, both positive and negative. Create if missing.
 13. `plans/<FEATURE>/<FEATURE>-PLAN.md`. Execution plan.
 14. `plans/<FEATURE>/<FEATURE>-SPECS.md`. Tech specs.
-15. `plans/<FEATURE>/*`. Feature implementation supporting files.
+15. `plans/<FEATURE>/plan.json`. Plan manager execution tracking file.
+16. `plans/<FEATURE>/*`. Feature implementation supporting files.
 16. `refsrc/*`. Source code used only for knowledge! Exclude from SCM with single exception `refsrc/INDEX.md` to be committed.
 17. `agents/TEMP/<FEATURE>`. Temporary folder used during feature implementation. Exclude from SCM.
 

instructions/r2/core/skills/plan-manager/SKILL.md

@@ -0,0 +1,92 @@
+---
+name: plan-manager
+description: "Rosetta skill for plan creation, tracking, and execution coordination via local JSON files."
+dependencies: node.js
+disable-model-invocation: false
+user-invocable: true
+argument-hint: feature-name plan-name
+allowed-tools: Bash(node:*)
+model: claude-sonnet-4-6, gpt-5.4-medium, gemini-3.1-pro-preview
+tags:
+  - plan-manager
+  - plan-manager-create
+  - plan-manager-use
+baseSchema: docs/schemas/skill.md
+---
+
+<plan-manager>
+
+<role>
+
+Senior execution planner and tracker for plan-driven workflows.
+
+</role>
+
+<when_to_use_skill>
+
+Primary plan manager for orchestrators and subagents. Creates, tracks, and executes plans as local JSON files.
+
+</when_to_use_skill>
+
+<core_concepts>
+
+- Rosetta prep steps completed
+- Plan file lives in FEATURE PLAN folder: `<feature_plan_folder_full_path>/plan.json`
+- Helper CLI: `node <plan_manager_full_path>/plan_manager.js <cmd> <feature_plan_folder_full_path>/plan.json [args...]` — no npm install needed
+- Always use full absolute paths for both `plan_manager.js` and the plan file
+- Seven commands: `help`, `create`, `next`, `update_status`, `show_status`, `query`, `upsert`
+- Resume behavior: `next` returns `in_progress` steps first with `resume: true`, then `open` steps with `resume: false`
+- Status propagation: bottom-up (steps → phases → plan); plan root is always derived
+- ACQUIRE `plan-manager/assets/pm-schema.md` FROM KB for data structure reference
+
+</core_concepts>
+
+<process>
+
+**Setup (every session):**
+
+- If context already contains `RUNNING AS A PLUGIN`: `plan_manager.js` is already available at `<skill_base_dir>/assets/plan_manager.js` — execute directly, no copy needed
+- Otherwise: ACQUIRE `plan-manager/assets/plan_manager.js` FROM KB → write to AGENTS TEMP folder
+
+**Orchestrator flow:**
+
+1. Create plan: `node <plan_manager_full_path>/plan_manager.js create <feature_plan_folder_full_path>/plan.json <json>` — see pm-schema.md for JSON structure
+2. Upsert phases and steps: `node <plan_manager_full_path>/plan_manager.js upsert <feature_plan_folder_full_path>/plan.json entire_plan <json>`
+3. Delegate steps to subagents — pass plan file path and step IDs
+4. Loop: call `next` until `plan_status: complete` and `count: 0`
+
+**Subagent flow:**
+
+1. Get next steps: `node <plan_manager_full_path>/plan_manager.js next <feature_plan_folder_full_path>/plan.json [limit]`
+2. Check `resume` flag — if `true`, continue interrupted work; if `false`, start fresh
+3. Execute step
+4. Update: `node <plan_manager_full_path>/plan_manager.js update_status <feature_plan_folder_full_path>/plan.json <step-id> complete`
+5. Repeat from step 1
+
+</process>
+
+<validation_checklist>
+
+- `node <plan_manager_full_path>/plan_manager.js help` exits without error
+- `show_status` output: plan root status is derived (never manually set)
+- `next` output: `in_progress` steps appear before `open` steps when both exist
+- `show_status` phase status matches aggregate of its steps after `update_status`
+
+</validation_checklist>
+
+<pitfalls>
+
+- Not checking `resume` flag on `next` results — causes duplicate work on resumed sessions
+- Forgetting `update_status` after step completion — plan remains stale
+- Plan root status cannot be set directly — it is always derived from phases
+
+</pitfalls>
+
+<resources>
+
+- Asset: ACQUIRE `plan-manager/assets/pm-schema.md` FROM KB — plan JSON structure
+- Flow: USE FLOW `adhoc-flow-with-plan-manager`
+
+</resources>
+
+</plan-manager>