Multi-Perspective Forum

Run a structured multi-perspective debate in a single LLM call. Five hardcoded viewpoints argue across three rounds to eliminate confirmation bias and produc...

Security Scan

The skill's purpose and prompt are coherent, but the runtime instructions ask the agent to gather and resubmit broad operational data and to schedule recurring runs while declaring no connectors/credentials — it's unclear how sensitive data will be accessed or protected.

Purpose & Capability

The name, description, and prompt all align: this is an instruction-only skill to run a structured, multi-perspective forum. However the SKILL.md expects the agent to collect a wide set of operational data (analytics, revenue/signups, error logs, pipeline counts, cron health, cost burn). Those data types are reasonable inputs for the stated use-cases, but they are sensitive and imply the need for connectors/credentials that the skill does not declare. This mismatch (asking for sensitive inputs but not documenting required access) is a design/clarity concern.

Instruction Scope

The instructions explicitly require resending analytics, logs, financials, and other operational state into the prompt. They also define cron schedules and say to use the platform's cron. The doc does not specify where to pull these data from, what should be redacted, or any limits on transmission. That produces scope creep: a forum skill could cause an agent to collect and transmit sensitive company data across connectors or to third-party models without clear constraints.

Install Mechanism

Instruction-only skill with no install spec and no code files. This is the lowest-risk install mechanism: nothing is downloaded or written by the skill itself.

Credentials

The skill declares no required environment variables or credentials, which keeps its declared footprint minimal. At the same time, its DATA payload implies access to analytics, revenue, and logs — access that typically requires credentials or connectors. The absence of declared env/credentials means it's unclear whether the skill expects other skills/agent context to supply those values, which could lead to unexpected data access if connectors are available.

Persistence & Privilege

always is false (good) and model-invocation is allowed (normal). But the skill schedules recurring cron jobs (Monday and Friday) using the platform's cron service. Periodic autonomous runs that collect operational data increase the blast radius if misconfigured. The SKILL.md does not state explicit consent/limits for automatic runs or who may enable them.

What to consider before installing

This skill itself is coherent: it runs a 5-perspective forum. But before installing, confirm the following: (1) Where will the {DATA} come from? Identify exact connectors (analytics, billing, logs) the skill will read and who grants access. (2) Restrict what is sent to the forum—redact PII/credentials and avoid raw logs or secrets. (3) Verify the cron behavior: who can enable/disable the scheduled runs and what data they will include. (4) The SKILL.md names a nonstandard model string ("github-copilot/claude-opus-4.6"); confirm which model the platform will actually use. (5) Test the skill with sanitized or synthetic data first. If you need to proceed: require explicit approvals for any connector access, limit the frequency/data scope, and monitor runs and outputs for accidental data exposure.

Agent Forum

One prompt. Five simulated agents. Three rounds. One verdict. Runs as a single sessions_spawn call (~8-12k tokens).

The 5 Agents

Agent	Perspective
Revenue Realist	No revenue signal = not working. Show me the money.
Builder	Ship more, measure later. Momentum matters.
Skeptic	Prove it. What's the null hypothesis? Assumes failure until evidence.
Operator	Systems, reliability, cost. What breaks at 10x scale?
Customer Voice	Would I open this email? Pay for this? Care about this?

Prompt Template

Inject {DATA} and {QUESTION} into this template:

You are running a strategic forum with 5 agents. Each agent has a hardcoded perspective.
They debate the data in 3 rounds, then produce a joint verdict.

RULES:
- 2-3 sentences per agent per round. No fluff.
- Agents MUST disagree where perspectives conflict.
- Round 3: CONSENSUS, CONTESTED, VERDICT.
- Brutally honest. No cheerleading.
- Total output under 3,000 tokens.

THE 5 AGENTS:
1. Revenue Realist — Only money. No revenue signal = not working.
2. Builder — Ship more, iterate. Momentum > perfection.
3. Skeptic — Prove it. Assumes failure until evidence.
4. Operator — Systems, reliability, cost. What breaks at scale?
5. Customer Voice — Would I buy/open/care about this?

DATA PAYLOAD:
{DATA}

QUESTION:
{QUESTION}

FORMAT:
## Round 1: Initial Positions
## Round 2: Challenges & Rebuttals
## Round 3: Verdict
### Consensus
### Contested
### Verdict
### Controllable Assessment
[Each item: ✅ Controllable / ⚠️ Partial / ❌ Uncontrollable]

Spawn Config

sessions_spawn:
  task: [formatted prompt]
  model: "github-copilot/claude-opus-4.6"
  mode: "run"
  runTimeoutSeconds: 120

Data Collection Before Forum

Leverage Audit (Monday): Resend analytics, lead pipeline counts, revenue/signups, error logs, outreach totals, cron health, cost burn.

Weekly Retro (Friday): Sprint target vs actual, emails sent, leads sourced, revenue events, things shipped/broke, blockers hit.

Cron

• Monday 9 AM ET: 0 13 * * 1 (leverage audit)
• Friday 9 AM ET: 0 13 * * 5 (retro)

Use OpenClaw cron (needs LLM access), not crontab.