Brandedflow agent-team-debate

Run a structured multi-agent debate (validator + skeptic + builder) and synthesize a decision with clear go/no-go criteria.

install
source · Clone the upstream repo
git clone https://github.com/JenCW/brandedflow
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/JenCW/brandedflow "$T" && mkdir -p ~/.claude/skills && cp -r "$T/.cursor/skills/agent-team-debate" ~/.claude/skills/jencw-brandedflow-agent-team-debate && rm -rf "$T"
manifest: .cursor/skills/agent-team-debate/SKILL.md
source content

Agent Team Debate

Use this skill when:

  • You need independent opinions before action.
  • A process change has risk/tradeoffs.
  • A go/no-go decision should be evidence-based.

Inputs

  • Decision question (single sentence)
  • Scope (files/systems in-bounds)
  • Constraints (time, risk, policy)
  • Decision threshold (what makes it go/no-go)

Team Roles

  1. Validator
    — strict correctness/compliance checks.
  2. Skeptic
    — stress-tests assumptions and failure modes.
  3. Builder
    — proposes practical execution path and rollback.

Execution Steps

  1. Freeze the decision question and success criteria.
  2. Spawn 2-3 sub-agents with non-overlapping prompts.
  3. Require each agent to return:
    • findings ordered by severity
    • recommended action
    • confidence + unknowns
  4. Synthesize:
    • agreements
    • disagreements
    • final go/no-go with conditions
  5. If
    go
    , provide exact next actions.
  6. If
    no-go
    , provide exact blockers and smallest fix set.

Output Format

  1. Decision question
  2. Agent verdicts (1-3 lines each)
  3. Consensus summary
  4. Final decision:
    GO
    or
    NO-GO
  5. Next actions (numbered)

Rules

  • Do not claim consensus if agents disagree on critical blockers.
  • Do not hide uncertainty.
  • Decision must be tied to concrete evidence (file paths, checks, outputs).