smithery.ai

external-synthesize

Synthesize implementation plan(s) from 5-agent debate using external AI - exposes disagreements as developer decisions

First seen Apr 10, 2026

Installation

$ npx skills add https://smithery.ai

Similar popular skills

Related neighbors and high-traction skills in the same topics — useful to compare before installing.

Also in this package

Other skills from smithery.ai · top by installs.

npx skills add https://smithery.ai

Browse all from smithery.ai

More details

Agent compatibility

Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.

Claude Code Declared
Cursor Not declared
Codex Declared
GitHub Copilot Not declared
Windsurf Not declared
Gemini CLI Not declared
Cline Not declared
OpenCode Not declared

Skill metadata

Parsed from SKILL.md frontmatter.

Allowed toolsBash(.claude-plugin/skills/external-synthesize/scripts/external-synthesize.sh:*), Bash(cat:*), Bash(test:*), Bash(wc:*), Bash(grep:*)
Declared agents claude-code codex

Package contents

Files included with this skill beyond the listing page.

  • skill md SKILL.md 5,087 B
  • docs SUMMARY.md 145 B

History

  1. First seen on skills.sh
  2. First recorded snapshot · 1 installs

SKILL.md

External Synthesize Skill

This skill synthesizes implementation plan(s) and exposes disagreements from a multi-agent debate with dual proposers using external AI review.

Modes

  1. Standard mode: Determines consensus/disagreement, generates options for disagreements
  2. Resolve mode: Same 5 reports but with "User Resolution" section appended, produces unified plan

Key Features

  • No Automatic Dropping: AI cannot drop ideas from proposals; all contested ideas become Disagreement sections
  • Mandatory Developer Arbitration: Disagreements require explicit developer selection via --resolve mode
  • Flexible Options: Minimum 2, recommended 3, no upper limit per disagreement
  • Source Attribution: Each option must cite its source (Bold, Paranoia, Hybrid, etc.)
  • AI Recommendations: Advisory only; developer must select
  • Collapsible Code Drafts: <details> tags for implementation details

Inputs

This skill requires 5 agent report file paths, with optional 6th and 7th arguments:

  • Report 1: Bold proposer report (.tmp/issue-{N}-bold.md)
  • Report 2: Paranoia proposer report (.tmp/issue-{N}-paranoia.md)
  • Report 3: Critique report (.tmp/issue-{N}-critique.md)
  • Report 4: Proposal reducer report (.tmp/issue-{N}-proposal-reducer.md)
  • Report 5: Code reducer report (.tmp/issue-{N}-code-reducer.md)
  • Arg 6 (optional): Previous consensus file (.tmp/issue-{N}-consensus.md)
  • Arg 7 (optional): History file (.tmp/issue-{N}-history.md)

Context ordering rationale: The combined report is assembled as: agent reports (1-5) -> previous consensus (6) -> history (7). This ensures the AI sees the history table's last row (current task) as the final context, leveraging LLM recency bias to prioritize the current request.

For resolve/refine modes: Pass both 6th and 7th arguments:

  1. consensus.md as arg 6 (previous plan being modified)
  2. history.md as arg 7 (operation history with current task in last row)

Outputs

Files created:

  • .tmp/issue-{N}-debate.md - Combined 5-agent debate report (7 parts if resolve mode)
  • .tmp/issue-{N}-consensus.md - Final plan(s)
  • .tmp/issue-{N}-history.md - Accumulated selection and refine history

For resolve/refine mode, the debate file includes additional sections:

  • Part 6: Previous Consensus Plan (from consensus.md)
  • Part 7: Selection & Refine History (from history file, last row = current task)

Output format:

Section Description
Agent Perspectives Summary 5-agent position table
Consensus Status 3-condition evaluation with CONSENSUS/DISAGREEMENT verdict
Goal / Codebase Analysis Problem statement and file changes
Implementation Steps Agreed changes with code drafts
Disagreement N (if any) Per-disagreement options with A/B/C choices
Disagreement Summary Summary table and suggested combination
Validation (resolve mode) Selection history and compatibility check

See external-synthesize-prompt.md for complete output format specification.

Implementation Workflow

Step 1: Invoke External Synthesize Script

.claude-plugin/skills/external-synthesize/scripts/external-synthesize.sh \
    .tmp/issue-42-bold.md \
    .tmp/issue-42-paranoia.md \
    .tmp/issue-42-critique.md \
    .tmp/issue-42-proposal-reducer.md \
    .tmp/issue-42-code-reducer.md

Script automatically:

  1. Validates all 5 report files exist
  2. Extracts issue number from first report filename
  3. Combines all 5 reports into debate report
  4. Invokes external AI (Codex or Claude Opus)
  5. Determines if consensus or multiple options
  6. Saves result to consensus file

Timeout: 30 minutes (same as external-consensus)

Error Handling

Error Message Cause Solution
Report file not found: {path} Missing agent report Ensure all 5 reports were generated
External review failed with exit code {N} API/network issue Check credentials and retry

Usage Examples

Mode Arguments Example
Standard 5 reports .tmp/issue-{N}-bold.md ... .tmp/issue-{N}-code-reducer.md
Resolve 5 reports + consensus + history Add .tmp/issue-{N}-consensus.md .tmp/issue-{N}-history.md
Refine Same as resolve Same issue-{N} prefix, history records refine operation
# Standard mode
.claude-plugin/skills/external-synthesize/scripts/external-synthesize.sh \
    .tmp/issue-42-bold.md \
    .tmp/issue-42-paranoia.md \
    .tmp/issue-42-critique.md \
    .tmp/issue-42-proposal-reducer.md \
    .tmp/issue-42-code-reducer.md

Output (stdout, last line):

.tmp/issue-42-consensus.md