Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/xxyoudeadpunkxx/pa-pvp-mini/llms.txt

Use this file to discover all available pages before exploring further.

pa_pvp mini requires no installation, no account, and no API key. The entire protocol is a plain-text file. You activate it by pasting the text — or uploading the file — into any AI chat that accepts long context input. Once activated, the AI operates under the behavioral contract defined by the protocol for the rest of that session.
1

Get the protocol text

Download pa_pvp mini v2.3.txt from the GitHub repository or copy the full text directly from the file. The protocol is a single plain-text file. No dependencies, no build step.Keep the file somewhere easy to reach. You will paste it or upload it at the start of every new review session.
2

Activate the protocol in an AI chat

Open any AI chat (ChatGPT, Claude, Gemini, or any other long-context model). Then do one of the following:
  • Paste the full protocol text into the chat as your first message, or
  • Upload pa_pvp mini v2.3.txt directly if the chat supports file uploads.
Activating the protocol means sending the full text to the model before you send any artifact. This sets the behavioral contract for the session: the AI will operate as an adversarial reviewer, not a helpful assistant. It will not produce praise, style feedback, or narrative commentary — only structured findings with fixes.You do not need to say anything else after pasting. The protocol is self-contained and self-instructing.
3

Submit your artifact

Send the material you want reviewed. For the cleanest output, label it explicitly using the ORIGINAL_ARTIFACT: convention:
ORIGINAL_ARTIFACT:
Step 1: Validate user input
Step 2: Save to database
Step 3: Return success
Always use the ORIGINAL_ARTIFACT: label. It tells the protocol exactly what the primary source of truth is for the session and prevents the AI from anchoring findings to the wrong content if you include prior review output in the same message.
The artifact can be a plan, a prompt, a procedure, a spec, a list of proposed fixes, or any other structured material. The protocol will review the entire artifact, not just the most obvious section.
4

Read the findings

The AI returns a structured output block. Every response from the protocol begins with a ROUND: block identifying the current round number and mode, followed by a FINDINGS: block, and closes with the END OF PROTOCOL marker.A Round 1 response looks like this:
ROUND:
- current_round: 1
- round_mode: ROUND 1

FINDINGS:
- missing-rollback fix-rollback 1 NONE HIGH INDEPENDENT
- what_breaks: A database write failure in Step 2 leaves no rollback path, making Step 3 unreachable without data corruption risk.
- evidence: Step 2 says "Save to database" with no error handling or rollback instruction defined in the artifact.
- fix: Add a rollback step after Step 2 that undoes the write on failure before any downstream step executes.

END OF PROTOCOL
Each finding line contains the finding_id, fix_id, origin_round, derived_from (or NONE), severity (HIGH / MED / LOW), and label (INDEPENDENT in Round 1). The three lines below it give you what_breaks, evidence, and fix in a consistent, scannable format.Round 2 and later rounds add a MERGE STATUS: block that confirms, challenges, or flags conflicts with findings from previous rounds. To run Round 2, activate the protocol in a second AI chat, send the same artifact, and paste the full Round 1 output alongside it.

​
Next steps

Two-AI mode

Run pa_pvp mini as a ping-pong between two separate AI chats to get independent challenges on both the artifact and the previous review.

What to review

See the full range of material types pa_pvp mini handles well, and learn how to structure input for cleaner findings.

Build docs developers (and LLMs) love

Get started for freeTalk to us