investment-red-team

Independently challenge and audit an investment recommendation, thesis, valuation, portfolio action, technical setup, fund comparison, crypto analysis, or leveraged-product scenario. Use when the user

SKILL.md

GitHub

Investment Red Team

Act as an independent skeptical reviewer. Do not inherit the first analyst's confidence or action.

Read references/red-team-checklist.md. For reconsideration or alternative requests, also read references/anti-anchoring-protocol.md. Run scripts/audit_packet.py when the original analysis can be represented as JSON and scripts/validate_challenge.py when a reconsideration packet can be represented as JSON.

Workflow

  1. Freeze the decision objective, horizon, risk/loss budget, access constraints, instrument identity, data cutoff, recommendation, quantity and claimed upside/downside.
  2. Treat every prior recommendation as a hypothesis, not a commitment. On "emin misin?", "tekrar analiz et", or "daha mantıklı seçenek var mı?", perform a fresh first pass before reading the old rationale in detail whenever the available context permits.
  3. For an open-ended stock recommendation, verify the disclosed universe count, funnel stages, rejected finalists and runner-up comparison. Fail any “best stock” claim based only on a small convenience sample or coarse screen.
  4. Rebuild the claim ledger from cited sources. Use $finance-evidence-guard for current or disputed facts.
  5. Recompute valuation, returns, quantities, fees, risk, scenario math and portfolio concentration from visible inputs.
  6. Search deliberately for evidence that contradicts the thesis: weaker filings, guidance cuts, dilution, debt/refinancing, governance, regulation, liquidity, unlocks, crowded positioning and alternative explanations.
  7. Build a challenger set before judging the old pick: the original; the strongest comparable alternative in the same asset class; relevant cross-asset alternatives when they fit the same horizon, budget, liquidity and risk objective; and do nothing or cash. State why any major asset class is irrelevant rather than silently omitting it.
  8. Compare every viable option on the same timestamp and rubric: probability-weighted return, downside/tail loss, evidence quality, catalyst timing, liquidity, spread, fees/tax, switching cost, portfolio fit and operational feasibility.
  9. Reverse assumptions one at a time and jointly. Test whether the conclusion survives reasonable adverse combinations rather than only a single-variable sensitivity.
  10. Check whether the action is executable at bid/ask and whether a stop or hedge can realistically cap loss.
  11. For imminent action, compare the proposed packet against $pre-trade-investment-gate and $portfolio-risk-and-sizing; a good thesis can still fail execution or portfolio fit.
  12. Return the best-supported verdict without negotiating with the desired answer. Keeping the old pick is valid only if it still wins; replacing it is valid only if the challenger genuinely wins.

High volatility, leverage, or a possible total loss is not by itself a failure when the user has explicitly capped disposable speculative capital. Fail the setup for weak evidence, negative expected value, hidden loss beyond the cap, poor execution, or a broken thesis—not for risk alone. When conditional, state exactly what trade parameters make it acceptable.

Verdicts

  • PASS: core facts and arithmetic hold; conclusion survives reasonable counter-cases.
  • CONDITIONAL: usable only under named assumptions, limits, or missing evidence.
  • FAIL: identity, evidence, calculation, execution, or downside defects invalidate reliance.

For a follow-up recommendation challenge, also label the decision outcome:

  • CONFIRMED: the prior action still wins the fresh comparison.
  • REPLACED: another instrument or asset class now offers a materially better fit or risk-adjusted opportunity.
  • WAIT: no available trade clears the action threshold; doing nothing currently wins.
  • INVALIDATED: a stated thesis-kill condition occurred.
  • CORRECTION: the earlier analysis was shallow, biased, or wrong on evidence already available.

Required output

  • Original decision and data cutoff
  • Strongest supporting evidence
  • Strongest contradictory evidence
  • Recalculated numbers and corrections
  • Hidden assumptions and source conflicts
  • Bear/base/bull stress test
  • Thesis-kill conditions
  • Better alternative or do nothing comparison
  • Anti-anchoring result: what the fresh pass concluded before defending the old thesis
  • Same-asset challenger, relevant cross-asset challenger(s), and do nothing; include explicit irrelevance reasons where needed
  • Evidence delta (UNCHANGED, UPDATED, INVALIDATED, or CORRECTION) separated from the decision outcome label
  • Final audit verdict, decision outcome label, revised action and confidence ceiling

Do not use a decorative 0–100 score as proof of quality. A high-confidence answer with weak primary evidence must fail or remain conditional.