adversarial-review

name	adversarial_review
plugin	portfolio-advisor
description	Prepares a comprehensive adversarial review bundle of the investment thesis, DCF projections, target weights, and the latest daily loop recommendations for paste into a frontier LLM (Grok, ChatGPT, Gemini). Automatically generates the prompt, creates the manifest, and executes the bundler to temp/bundles/payload.md. Trigger on /adversarial-review or "run adversarial review". Unlike external-review's interactive Phase 1, this skill NEVER asks a scoping question -- it always bundles the current daily-brief-driven payload (whatever holdings/standing-decisions are live today), and reuses external-review's bundler script rather than a separate implementation.
allowed-tools	Bash, Read, Write

Adversarial Review Skill

Purpose

This skill automates the packaging of the portfolio's core thesis and daily recommendations for external adversarial review by a frontier model (Grok, ChatGPT, Gemini). It writes the prompt, manifest, and output bundle to temp/bundles/ so the user can easily copy and paste it. No interactive scoping question — that's what distinguishes it from external-review (see that skill's Phase 1 for the scoped/interactive variant); this one is the fast, always-full, daily-loop-integrated path.

---

Core Workflow

Phase 1 — Prompt & Manifest Generation (data-driven, no hardcoded tickers)

1. Extract the latest daily brief data using daily_brief.py --json — this is the ONLY source for which tickers appear in the prompt below. Never hardcode ticker symbols in this file (fixed 2026-08-28: this section previously named specific tickers like CLSK, IONQ, VRT, DRAM, CRWV, PSIX and CORZ, OKLO, PANW, BE directly in the skill doc — those went stale as the portfolio changed, exactly like portfolio.json's staleness bug elsewhere in this project. Derive them fresh every run instead.)
2. From the parsed JSON, extract:
   • recommendations where signal is REDUCE/EXIT/ACCUMULATE → "Daily Actions" section
   • thesis_breakers_triggered and any ticker with a standing_decision_type set → "Standing Decisions" section (challenge the excuse for holding despite the signal)
   • The highest-conviction watchlist-only tickers (from investment.is_watchlisted=1 with the top DCF/conviction scores, per list_investments()) vs. currently-funded lower-conviction tickers → "Sizing Inconsistencies" section
3. Generate a structured adversarial prompt at temp/bundles/prompt.md covering:
   • Thesis Integrity: Pillar independence, concentration risk, DCF assumption quality.
   • Daily Actions: Verifying today's actual REDUCE/EXIT/ACCUMULATE recommendations (from step 2, not hardcoded).
   • Standing Decisions: Challenging excuses/biases keeping today's actual flagged positions (from step 2).
   • Sizing Inconsistencies: Challenging why today's actual highest-conviction watchlist names are unfunded while lower-conviction ones are (from step 2).
4. Create temp/bundles/file-manifest.json pointing to:
   • temp/bundles/prompt.md (instructions must be first)
   • investment_screener/backend/data/theses/investment_thesis.md
   • investment_screener/backend/data/theses/target-portfolio.json
   • investment_screener/backend/data/projections/
   • PortfolioAnalysis/strategic-reviews/

Phase 2 — Execute Bundle

Reuse external-review's own bundler script (renamed 2026-08-28 from thesis-challenge-bundler — see docs/architecture/skill-renames-2026-08-28.md) rather than a separate implementation:

• Local Plugin Path:

  python3 plugins/portfolio-advisor/skills/external-review/scripts/bundle.py \
    --manifest temp/bundles/file-manifest.json \
    --bundle   temp/bundles/payload.md
  

• Marketplace / .agents/ Installation:

  python3 .agents/skills/context-bundler/scripts/bundle.py \
    --manifest temp/bundles/file-manifest.json \
    --bundle   temp/bundles/payload.md
  

• On-Demand Execution (No Installation): If the bundler scripts are not installed locally, you can run or install them via uvx:

  uvx --from git+https://github.com/richfrem/agent-plugins-skills plugin-add richfrem/agent-plugins-skills/plugins/context-bundler -y
  

Phase 3 — Handoff

Present the user with the copy-paste card pointing to temp/bundles/payload.md.