Introduce plan manager to bootstrap files#78
Conversation
YevheniiaLementova
requested review from
ElizaVetaFomka and
isolomatov-gd
as code owners
Rosetta Triage ReviewSummary: This PR refactors the Rosetta bootstrap rules across all four plugin IDEs (Claude, Codex, Copilot, Cursor) and Findings:
Suggestions:
Automated triage by Rosetta agent |
isolomatov-gd
left a comment
isolomatov-gd
left a comment
There was a problem hiding this comment.
I have commented on one file, but it applies to everything.
We must properly review original content vs the result.
Moreover, We are repeating the same thing, I suggest to fix only one file, once it is done, reuse it for all other cases.
| { | ||
| "id": "s-subagent-contract", | ||
| "name": "Execute subagent contract", | ||
| "prompt": "USE SKILL `subagent-contract` as first action before reading inputs or starting execution. Orchestrator request → read get_context_instructions schema (if needed) → get_context_instructions (if not done yet) → execute all prep steps for subagent → proceed with remaining actions. Perform execution todo tasks level planning. MUST execute todo tasks and adopt changes. Proceed with request and activities.", |
There was a problem hiding this comment.
This prompt is partially correct. Check it out. Please spend time on actually reading and understanding what it says. This prompt changed overall the process. Don't use AI, do it yourself.
Some things should be in this plan, but some things should NOT.
Example: request → read get_context_instructions schema (if needed) → get_context_instructions (if not done yet) → execute all prep steps for subagent → proceed with remaining actions. this is overall flow, why is it in the prompt of the specific step?
How to make AI to properly address that:
- Make AI read original prompt, ask it to identify the purpose of EACH line.
- Ask it to classify what is the step and what is explanation, or any other category
- Ask it to summarize steps as one JSON and keep wording as-is. And to keep the rest AS-IS.
- Ask it now to add all optional steps
- Ask AI to provide rules which will say which optional steps is for which condition.
- Update, and review what actually the purpose of each step.
- Ask it to review if anything is lost.
Please do not directly ask AI to do something, as it will provide a bad results.
| "steps": [ | ||
| { | ||
| "id": "s-subagent-contract", | ||
| "name": "Execute subagent contract", |
There was a problem hiding this comment.
LOAD instead of EXECUTE?
There was a problem hiding this comment.
Thank you for pointing, I'll replace it to 'load'
| "steps": [ | ||
| { | ||
| "id": "s-orchestrator-contract", | ||
| "name": "Execute orchestrator contract", |
There was a problem hiding this comment.
Thank you! I will replace
| { | ||
| "id": "s-select-workflow", | ||
| "name": "Select and load workflow", | ||
| "prompt": "ACQUIRE matching workflow from available workflows list and FULLY EXECUTE its definition for ALL request sizes. Workflow upserts its own phases/steps into the plan. In planning mode, persist workflow tag for execution continuation. In planning mode, store planning and tech-specs results according to system prompt (NOT in plans folder as it is read-only).", |
There was a problem hiding this comment.
I see workflow twice.
There was a problem hiding this comment.
Yes, you're right. Thank you for highlighting! This step should be removed at all as we have selection workflow in s-orchestrator-contract step
…ow selection into orchestrator contract, and clarify agent/subagent self-initialization.
|
@isolomatov-gd I’ve introduced changes only to the bootstrap files in instructions/r3/rules. If this version is approved, I’ll apply the same changes to the plugin folder. |
|
/rosetta Please check consistency |
Rosetta Consistency CheckResponding to ✅ Plugin-to-Plugin ParityAll four plugin directories (
|
| # | Issue | Severity | Location |
|---|---|---|---|
| 1 | Duplicate "Trust Rosetta" points 17 & 18 | Medium | r3/rules/bootstrap.md:57-58, r3/rules/local-files-mode.md:50-51 |
| 2 | s-select-workflow step missing in r3 canonical |
High | r3/rules/bootstrap-core-policy.md |
| 3 | s-subagent-contract depends_on mismatch |
High | r3/rules/bootstrap-core-policy.md vs plugins |
| 4 | Missing space too.Reevaluate in s-request-size prompt |
Low | All 4 plugin bootstrap-core-policy.md |
| 5 | Step name "Load" vs "Execute" inconsistency | Low | r3/rules/bootstrap-core-policy.md |
| 6 | Unclosed ] in s-orchestrator-contract prompt |
Low | All 4 plugin bootstrap-core-policy.md |
| 7 | Single-quote + extra space in s-context prompt |
Low | r3/rules/local-files-mode.md:28 |
| 8 | Missing trailing newline | Trivial | CLAUDE.md |
Items #2 and #3 are the most impactful — they indicate the r3 canonical files and the plugin files have diverged in bootstrap-core-policy.md, with the plugin version appearing more complete/correct. The r3 canonical may need to be updated to match the plugin version for these two points.
Automated consistency check by Rosetta agent
|
|
||
| 1. USE SKILL `plan-manager` — CLI: `npx rosettify plan <subcommand> <plan_file> [args...]` | ||
| 2. Create plan file at `plans/<FEATURE>/plan.json` (derive FEATURE from user request; use `session` if unclear) | ||
| 3. Create plan with the starter template below — fill `[PLACEHOLDERS]` from user request context: |
There was a problem hiding this comment.
PLAN_MANAGER create plans/<FEATURE>/plan.json .....
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
| - CLI: `npx rosettify plan <subcommand> <plan_file> [args...]` | ||
| - Try `rosettify` MCP first (if already available), fallback to CLI: `npx rosettify@latest plan <subcommand> <plan_file> [args...]`, fallback to todo tasks (if none available) | ||
| - Always use full absolute paths for the plan file | ||
| - Seven subcommands: `help`, `create`, `next`, `update_status`, `show_status`, `query`, `upsert` |
There was a problem hiding this comment.
There's a list of valid command in rosettify/dist/commands/plan/index.js:
const VALID_SUBCOMMANDS = [
"create",
"next",
"update_status",
"show_status",
"query",
"upsert",
];
There's no 'help' command. I think we should delete it as it was done in instructions/r2/core/skills/plan-manager/SKILL.md.
| <validation_checklist> | ||
|
|
||
| - `npx rosettify plan help` exits without error and returns structured help JSON | ||
| - `npx rosettify@latest plan help` exits without error and returns structured help JSON |
There was a problem hiding this comment.
help command isn't valid and was removed from core_concepts. It should be removed here as well
| <validation_checklist> | ||
|
|
||
| - `npx rosettify plan help` exits without error and returns structured help JSON | ||
| - `npx rosettify@latest plan help` exits without error and returns structured help JSON |
There was a problem hiding this comment.
help command isn't valid and should be removed
| @@ -0,0 +1,25 @@ | |||
| { | |||
There was a problem hiding this comment.
I assume that we don't need to commit this file. Should we put plans/session/plan.json to .gitignore?
Signed-off-by: isolomatov-gd <isolomatov@griddynamics.com>
…om the plan manager
…otstrap-and-MCP-instructions-v3
2 participants
No description provided.