From kimchi
Synthesizes multiple cross-model plan revisions into a superior hybrid final plan. Eighth stage of Kimchi pipeline; reads PLAN-REVISED-*.md files and outputs PLAN-SYNTHESIZED.md.
How this skill is triggered — by the user, by Claude, or both
Slash command
/kimchi:plan-synthesizeThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
<command_purpose>
<command_purpose> Analyze all model-specific plan revisions, compare their findings and fixes, and create a superior hybrid plan that artfully blends the best of all worlds. This is the TRUE final plan that feeds into bead conversion. </command_purpose>
Glob for .kimchi/PLAN-REVISED-*.md files. If none exist, tell the user: "No revised plans found. Run /kimchi:plan-revise in at least one model first."
Also read:
.kimchi/PLAN-DRAFT.md — the baseline draft plan.kimchi/REQUIREMENTS.md — v1 boundary referenceList all .kimchi/PLAN-REVISED-*.md files found. Extract model names from filenames. Report:
Found revisions from: claude, codex, gemini
Baseline: PLAN-DRAFT.md
For each revision, extract:
Build a comparison matrix:
Agreement: Issues found by multiple models (high confidence — fix these) Unique finds: Issues found by only one model (evaluate carefully — could be insight or noise) Contradictions: Where models disagree on approach (resolve with reasoning)
Apply ACFS-inspired synthesis logic:
"3 competing LLMs reviewed the same plan. Analyze with an open mind. Be intellectually honest about what each model did better. Create a superior hybrid that artfully blends the best of all worlds."
Rules:
Write .kimchi/PLAN-SYNTHESIZED.md:
# Synthesized Plan: [Feature Name]
**Synthesized:** [today's date]
**Sources:** PLAN-DRAFT.md, [list of PLAN-REVISED-*.md files]
**Models consulted:** [list of models]
## Cross-Model Comparison
| Issue | Claude | Codex | Gemini | Resolution |
|-------|--------|-------|--------|------------|
| [Issue description] | Found & fixed | Found & fixed | Not flagged | **Accepted** — unanimous/majority |
| [Issue description] | Not flagged | Found & fixed | Not flagged | **Accepted** — valid unique find |
| [Issue description] | Fix A | Fix B | Fix A | **Fix A** — [reasoning] |
## Synthesis Rationale
### What each model did well
- **Claude:** [specific strengths]
- **Codex:** [specific strengths]
- **Gemini:** [specific strengths]
### Key decisions
1. [Decision and reasoning]
2. [Decision and reasoning]
## Synthesized Plan
[Complete synthesized plan — full standalone plan ready for bead conversion]
Report: "Synthesis complete. Blended [N] model revisions into .kimchi/PLAN-SYNTHESIZED.md"
Suggest: "Run /kimchi:beads to convert the synthesized plan into bead specifications."
npx claudepluginhub tromml/kimchi --plugin kimchiCompares, cross-reviews, and merges competing plans from multiple agents (e.g., Codex, Claude Code) into one recommended execution plan.
Cross-references structured review findings against an implementation plan, classifies each finding into an action category, applies concrete edits, and produces a traceability summary.
Runs a Claude↔Codex convergence loop: Codex reviews a plan, Claude applies fixes, repeats until ALLOW or max-iter. For auto-reviewed plan refinement without manual copypaste.