agentleFS
AtmoskunAtmoskun/claude-skills/skills/gan/SKILL.md

GAN

Skill0 starsChanged 9 months agoMITView on GitHub

Adversarial analysis using Gemini and Claude. Two models independently analyze a query, cross-critique each other, then an arbitrator synthesizes the final report. Use when the user invokes /GAN followed by optional model flags and a quoted query.

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

13 rules. About 0 are common, 10 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈10

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Run both operations in parallel (two tool calls in the same message):

    line 68
  2. Distinctive:

    Write the analysis prompt to a temp file first, then call the script:

    line 72
  3. Very specific:

    Use the Write tool to create C:\Users\10444\Desktop\Claude WorkSpace\gemini-prompt-temp.md with this content:

    line 74
  4. Distinctive:

    Use the Task tool with:

    line 88
  5. Very specific:

    After both complete: Use Read tool to verify both gemini-reply-temp.md and claude-reply-temp.md exist and are non-empty.

    line 103
  6. Distinctive:

    Run both operations in parallel:

    line 109
  7. Very specific:

    Write critique prompt to C:\Users\10444\Desktop\Claude WorkSpace\gemini-critique-prompt-temp.md:

    line 113
  8. Distinctive:

    Run via Bash (timeout 300000):

    line 123
  9. Distinctive:

    Delete all temp files via Bash:

    line 184
  10. Distinctive:

    If any Gemini API call fails, show the error message (model name + HTTP status) and ask the user if they want to retry or abort

    line 193
  11. Distinctive:

    If Claude Task tool fails, report which step failed

    line 194
  12. Distinctive:

    If a temp file is missing when expected, report which step likely failed

    line 196
  13. Distinctive:

    All prompts follow the query language — if the query is in Chinese, responses will be in Chinese

    line 204

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

GAN - Adversarial Analysis Skill

Overview

Two AI models (Gemini + Claude) independently analyze a query, cross-critique each other's work, then an arbitrator synthesizes everything into a final comprehensive report.

Command Format

/GAN "Query"
/GAN -gemini MODEL -claude MODEL "Query"
/GAN -gemini MODEL -claude MODEL -arbitrator MODEL "Query"

Default Models

RoleDefault ModelNotes
Geminigemini-3-flash-previewFast analysis (v1beta)
ClaudesonnetClaude Sonnet 4.5 via Task
Arbitratorgemini-3-pro-previewSenior synthesis (v1beta)

Model Overrides

  • -gemini MODEL — Override Gemini analysis model (e.g., -gemini gemini-2.0-flash)
  • -claude MODEL — Override Claude model. Accepts sonnet, opus, haiku, or full names (mapped: contains "opus" → opus, "haiku" → haiku, else → sonnet)
  • -arbitrator MODEL — Override arbitrator Gemini model (e.g., -arbitrator gemini-2.5-pro)

Argument Parsing

  1. Extract flags: scan for -gemini, -claude, -arbitrator — the token immediately after each flag is its model value
  2. Extract query: everything remaining after removing flags and their values, stripped of outer quotes (", ", ")
  3. Apply defaults for any unspecified flags
  4. Map Claude model: if value contains "opus" → opus; "haiku" → haiku; otherwise → sonnet

Paths

  • Skill directory: C:\Users\10444\.claude\skills\gan\
  • Gemini script: C:\Users\10444\.claude\skills\gan\scripts\gemini_call.py
  • Temp directory: C:\Users\10444\Desktop\Claude WorkSpace\
  • Python: python (available in PATH)

Temp files created during execution:

  • gemini-reply-temp.md — Gemini analysis + Claude peer review
  • claude-reply-temp.md — Claude analysis + Gemini peer review
  • arbitrator-prompt-temp.md — Combined prompt for arbitrator

Execution

Announce Start

Print to user:

GAN Adversarial Analysis
Gemini: [GEMINI_MODEL] | Claude: [CLAUDE_MODEL] | Arbitrator: [ARBITRATOR_MODEL]
Query: [QUERY_TEXT]

Step 1 & 2: Initial Analysis (PARALLEL)

Run both operations in parallel (two tool calls in the same message):

1a. Gemini Analysis → Bash tool

Write the analysis prompt to a temp file first, then call the script:

  1. Use the Write tool to create C:\Users\10444\Desktop\Claude WorkSpace\gemini-prompt-temp.md with this content:
    You are the most diligent investigator and critical journalist. Below is the mission:
    
    [QUERY_TEXT]
  2. Then run via Bash (timeout 300000):
    python "C:/Users/10444/.claude/skills/gan/scripts/gemini_call.py" --model GEMINI_MODEL --prompt-file "C:/Users/10444/Desktop/Claude WorkSpace/gemini-prompt-temp.md" --output "C:/Users/10444/Desktop/Claude WorkSpace/gemini-reply-temp.md"

1b. Claude Analysis → Task tool

Use the Task tool with:

  • subagent_type: general-purpose
  • model: resolved Claude model (sonnet/opus/haiku)
  • run_in_background: true
  • prompt:
    You are the most diligent investigator and critical journalist. Below is the mission:
    
    [QUERY_TEXT]
    
    Write your complete, thorough analysis addressing the mission above.
    When finished, save your FULL analysis to `C:\Users\10444\Desktop\Claude WorkSpace\claude-reply-temp.md` using the Write tool.
    Output only the analysis — no meta-commentary about the task itself.

After both complete: Use Read tool to verify both gemini-reply-temp.md and claude-reply-temp.md exist and are non-empty. Report to user: Step 1&2 done — initial analyses ready.


Step 3 & 4: Cross-Critique (PARALLEL)

Run both operations in parallel:

3a. Gemini Critiques Claude → Bash tool

  1. Write critique prompt to C:\Users\10444\Desktop\Claude WorkSpace\gemini-critique-prompt-temp.md:
    You are the most diligent investigator and critical journalist. Your colleague has written the following report in response to this mission: [QUERY_TEXT]
    
    Review their report below and:
    1. Identify key details that are missing or underexplored
    2. Point out potential logical flaws, unsupported claims, or biases
    3. Provide specific, actionable improvement suggestions
  2. Run via Bash (timeout 300000):
    python "C:/Users/10444/.claude/skills/gan/scripts/gemini_call.py" --model GEMINI_MODEL --prompt-file "C:/Users/10444/Desktop/Claude WorkSpace/gemini-critique-prompt-temp.md" --context-file "C:/Users/10444/Desktop/Claude WorkSpace/claude-reply-temp.md" --output "C:/Users/10444/Desktop/Claude WorkSpace/claude-reply-temp.md" --append

3b. Claude Critiques Gemini → Task tool

Use the Task tool with:

  • subagent_type: general-purpose
  • model: resolved Claude model
  • run_in_background: true
  • prompt:
    You are the most diligent investigator and critical journalist. Your colleague has written the following report in response to this mission: [QUERY_TEXT]
    
    Steps:
    1. Read the colleague's report from `C:\Users\10444\Desktop\Claude WorkSpace\gemini-reply-temp.md`
    2. Write a thorough critique covering:
       - Key details that are missing or underexplored
       - Potential logical flaws, unsupported claims, or biases
       - Specific, actionable improvement suggestions
    3. Read the current full content of `C:\Users\10444\Desktop\Claude WorkSpace\gemini-reply-temp.md` again
    4. Use the Write tool to rewrite that file with: the original content, then "\n\n---\n\n## Peer Review\n\n", then your critique

After both complete: Report to user: Step 3&4 done — cross-critiques complete.


Step 5: Arbitration

  1. Read both files:
    • C:\Users\10444\Desktop\Claude WorkSpace\gemini-reply-temp.md
    • C:\Users\10444\Desktop\Claude WorkSpace\claude-reply-temp.md
  2. Write the arbitrator prompt to C:\Users\10444\Desktop\Claude WorkSpace\arbitrator-prompt-temp.md:
    You are the most senior investigator and journalist. Attached are a set of reports and follow-up questions prepared by your junior colleagues. Based on these preliminary materials, please compile a thorough study on:
    
    [QUERY_TEXT]
    
    === REPORT A (Gemini Analysis + Claude Peer Review) ===
    
    [Full contents of gemini-reply-temp.md]
    
    === REPORT B (Claude Analysis + Gemini Peer Review) ===
    
    [Full contents of claude-reply-temp.md]
  3. Execute via Bash (timeout 300000). IMPORTANT: On Windows, Chinese/Unicode output via stdout is garbled. Always save to file then read:
    python "C:/Users/10444/.claude/skills/gan/scripts/gemini_call.py" --model ARBITRATOR_MODEL --prompt-file "C:/Users/10444/Desktop/Claude WorkSpace/arbitrator-prompt-temp.md" --output "C:/Users/10444/Desktop/Claude WorkSpace/arbitrator-result-temp.md"
  4. Read C:\Users\10444\Desktop\Claude WorkSpace\arbitrator-result-temp.md using the Read tool, then display the contents directly to the user as the final report (no modification, no wrapping).

Step 6: Cleanup

Delete all temp files via Bash:

rm -f "C:/Users/10444/Desktop/Claude WorkSpace/gemini-reply-temp.md" "C:/Users/10444/Desktop/Claude WorkSpace/claude-reply-temp.md" "C:/Users/10444/Desktop/Claude WorkSpace/arbitrator-prompt-temp.md" "C:/Users/10444/Desktop/Claude WorkSpace/gemini-prompt-temp.md" "C:/Users/10444/Desktop/Claude WorkSpace/gemini-critique-prompt-temp.md" "C:/Users/10444/Desktop/Claude WorkSpace/arbitrator-result-temp.md"

Report: Cleanup done.

Error Handling

  • If any Gemini API call fails, show the error message (model name + HTTP status) and ask the user if they want to retry or abort
  • If Claude Task tool fails, report which step failed
  • Always run Step 6 cleanup even if earlier steps fail
  • If a temp file is missing when expected, report which step likely failed

Notes

  • Available Gemini models (tested):
    • v1beta (preview): gemini-3-flash-preview, gemini-3-pro-preview
    • v1 (stable): gemini-2.5-flash, gemini-2.5-pro, gemini-2.0-flash, gemini-2.0-flash-lite, gemini-2.5-flash-lite
  • The script auto-selects API version: v1beta for models containing "preview", v1 for others
  • All prompts follow the query language — if the query is in Chinese, responses will be in Chinese

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.