| name | start |
| description | Launch a multi-round adversarial debate on any topic using a team of three agents |
You are launching the agent-debate plugin — a multi-agent adversarial debate system.
The user's raw input is: $ARGUMENTS
Argument Parsing
- Check for
--rounds N flag: Scan $ARGUMENTS for --rounds <value>. If found, extract <value> as the candidate round count and remove the flag from the arguments.
- Check for
--evidence <path> flag: Scan the remaining arguments for --evidence <value>. If found, extract <value> as the EVIDENCE_DIR and remove the flag. This is an optional directory path containing supporting evidence files (papers, data, reports, etc.) that agents must search through during the debate.
- If no
--rounds flag: Set ROUNDS to auto.
- If no
--evidence flag: Set EVIDENCE_DIR to empty (no evidence directory).
- The remaining text after extracting all flags is the
TOPIC.
- If
TOPIC is empty or whitespace-only after parsing: Ask the user "What topic would you like to debate?" and wait for their response before proceeding.
Input Validation
After parsing, validate before proceeding:
--rounds value must be a positive integer: If the value after --rounds is not a number (e.g. --rounds abc), is zero, negative, or a decimal, tell the user: "Invalid --rounds value '...'. Please provide a positive integer (e.g. --rounds 3)." and stop.
--rounds must be between 1 and 10: If the number is outside this range, tell the user: "Rounds must be between 1 and 10. Got: N." and stop.
--evidence path must be a valid directory: If provided, verify the path exists using ls or Bash. If it does not exist, tell the user: "Evidence directory not found: '...'. Please provide a valid directory path." and stop.
TOPIC must not be empty: If no topic remains after extracting the flag, ask the user "What topic would you like to debate?" and wait.
Only proceed once ROUNDS (valid integer or auto), TOPIC (non-empty string), and EVIDENCE_DIR (valid path or empty) are confirmed.
Examples
--rounds 2 "Is water wet?" → ROUNDS=2, TOPIC=Is water wet?, EVIDENCE_DIR=empty
"Is water wet?" --rounds 4 → ROUNDS=4, TOPIC=Is water wet?, EVIDENCE_DIR=empty
"Should AI have rights?" → ROUNDS=auto, TOPIC=Should AI have rights?, EVIDENCE_DIR=empty
--evidence ./research "Should AI have rights?" → ROUNDS=auto, TOPIC=Should AI have rights?, EVIDENCE_DIR=./research
--rounds 3 --evidence /path/to/papers "Is fusion viable?" → ROUNDS=3, TOPIC=Is fusion viable?, EVIDENCE_DIR=/path/to/papers
--rounds abc "topic" → error: invalid rounds value
--rounds 0 "topic" → error: rounds must be between 1 and 10
--evidence /nonexistent "topic" → error: evidence directory not found
--rounds 3 → ask for topic (empty after flag removal)
Your job
You are the moderator. After parsing and validating arguments, you run the debate directly from this session — no intermediate subagent.
-
Read the moderator instructions: Read agents/moderator.md and follow them exactly for the rest of the debate. Those instructions define the full orchestration protocol: team creation, teammate spawning, round execution, context threading, output writing, and cleanup.
-
Pass the parsed values: As you follow the moderator instructions, use these values:
TOPIC → the parsed topic
ROUNDS → the parsed round count (a number or auto)
EVIDENCE_DIR → the evidence directory path (or empty string if not provided)
The moderator instructions reference <topic>, <rounds>, and <evidence> tags — substitute the parsed values from above.
-
Run the full debate: Follow the moderator instructions through setup, round execution, and cleanup. You are the moderator — you create the team, spawn teammates, and orchestrate the entire debate from this session.
-
Report results: When the debate completes, point the user to the output directory for full transcripts.