Team selection

Choose roles from the risk and the handoff—not the org chart.

More agents create more context, review, and coordination cost. Start with one owner and add a second perspective only when it changes the evidence or catches a concrete failure mode.

Last reviewed: 2026-07-21 · Decision rule: smallest sufficient team

Prerequisites

Before selecting a kit, write the intended artifact, acceptance checks, sensitive inputs, allowed tools, and external actions. Name the human who can approve the final result. If you cannot define the output or stop condition, team selection is premature.

Start with the work shape

Work shape	Starting kit	Useful separation	Do not delegate
Product slice	Solo Founder	Scope, build, claim review	Release approval and credentials
Editorial campaign	Content Creator	Research, drafting, measurement	Publishing and community identity
Historical strategy study	Quant Trader	Data quality, backtest, risk	Capital allocation and order placement
Single specialist task	One agent ID	Focused execution	Independent acceptance decision

Use five selection steps

Define the artifactChoose one output that can be inspected: a memo, patch, content brief, dataset profile, or backtest report.
Map failure modesList what could make the artifact wrong, unsafe, misleading, or impossible to roll back.
Assign distinct rolesGive each role a different responsibility and evidence requirement. Avoid duplicate generalists debating without new data.
Set authorityState which actions are read-only, which require human approval, and which are forbidden for the evaluation.
Time-box the trialRun one fixture, score the artifact, and keep or remove the team based on observed value.

Preview a kit or one role

# Preview a small starter kit
./install.sh --source "$PWD" --destination /path/to/review-workspace solo-founder

# Preview one specialist instead
./install.sh --source "$PWD" --destination /path/to/review-workspace pe

The default is preview. Compare the proposed footprint with the role you actually need. The full-team option is appropriate only when the evaluation genuinely needs all listed domains and you are prepared to review the larger install plan.

Score observed value

After one fixture, ask whether the team improved correctness, surfaced a risk, reduced rework, or produced evidence that a single-agent baseline missed. Also record coordination cost: repeated context, contradictory advice, latency, and review time. Keep a role only when its distinct contribution exceeds that cost.

Practical signal

If two roles consistently produce interchangeable output, merge their responsibilities for the next trial rather than inventing more handoff process.

Limitations

Role names do not create organizational authority or professional accountability. The repository does not guarantee that a larger team is more accurate, and it does not remove the need for domain experts in legal, financial, security, medical, or operational decisions. Harness support and tool access also vary by client.

Rollback and uninstall

Keep the preview list and remove only newly created workspace-<agent> directories after the trial. If you installed a native harness plugin, uninstall it through that client's scoped package controls. Retain the fixture and scorecard so the next team configuration can be compared without repeating the same assumptions.