@human-avatar/skills-for-humanity 1.0.0

package/skills/ethics-impact-scan/SKILL.md

@@ -0,0 +1,90 @@
+---
+name: ethics-impact-scan
+description: "Run a quick ethical impact assessment on a proposed feature, change, or decision before it ships. Use when the user is about to build or ship something and wants to surface who benefits, who's harmed, and at what scale — before commitments are made. TRIGGERS: 'ethics scan', 'impact check', 'who does this affect', 'is this safe to ship', any new feature proposal where stakeholder impact is unclear. Lightweight — takes minutes, not hours."
+---
+
+# Ethics Impact Scan
+
+A pre-ship ethical scan. Not a deep council — a structured sweep that forces you to see who's in the blast radius before you commit.
+
+It runs two lenses: **utilitarian** (net effect on aggregate wellbeing) and **justice/fairness** (whether benefits and burdens are distributed equitably). These two together catch the most common pre-ship blind spots: harm that's small-per-person but large-in-aggregate, and harm that falls disproportionately on people with the least power.
+
+---
+
+## Your Process
+
+**Step 1: Clarify the subject**
+State what is being scanned — a feature, change, product decision, or policy. If the subject is vague, ask one clarifying question before proceeding.
+
+**Step 2: Map the stakeholder field**
+Before applying any lens, identify everyone affected:
+- Direct users (who uses this feature and how?)
+- Indirect parties (who is affected by users' use of this feature?)
+- Third parties (suppliers, partners, communities)
+- Non-users (people who don't opt in but are still affected)
+- Future parties (people who will be affected by the precedent this sets)
+
+Don't skip non-users and future parties. They are the most commonly missed.
+
+**Step 3: Apply the Utilitarian Lens**
+For each stakeholder group:
+- What is the likely benefit?
+- What is the likely harm?
+- What is the scale (how many people, how significantly)?
+
+Then: Is the net effect positive? Who bears disproportionate cost to generate that net positive?
+
+**Step 4: Apply the Justice Lens**
+- Are benefits distributed fairly, or do they flow primarily to users who already have more power, resources, or access?
+- Are burdens distributed fairly, or do they fall primarily on those with the least power to resist them?
+- Would the decision-makers accept this outcome if they didn't know which role they'd occupy?
+
+**Step 5: Surface the flags**
+Produce a short list of ethical flags — things that warrant attention before shipping. A flag is not a veto; it is a signal that deserves a conscious decision. Distinguish:
+- 🔴 **Block** — this is a significant harm that should be resolved before shipping
+- 🟡 **Watch** — this is a risk worth monitoring or mitigating, but not a blocker
+- 🟢 **Note** — this is worth being aware of, but is low-risk
+
+---
+
+## Human Check-in
+
+Before proceeding, ask the user:
+
+**How do you want to run this?**
+
+- **A) Full analysis** — complete all steps, reasoning shown throughout
+- **B) Key findings only** — bottom-line output, skip step-by-step detail
+- **C) Go/no-go verdict only** — recommendation and top conditions, skip the benefit/harm map
+- **D) Refine the framing** — adjust what we're analyzing before starting
+
+Proceed based on their choice.
+
+## Output Format
+
+**Subject:** [what is being scanned]
+
+**Stakeholder Map**
+| Stakeholder | Affected How | Scale |
+|---|---|---|
+| [group] | [benefit or harm] | [rough scale] |
+
+**Utilitarian Assessment**
+[2–3 sentences on net effect: who benefits, who's harmed, is the aggregate positive]
+
+**Justice Assessment**
+[2–3 sentences on distribution: are benefits and burdens equitably spread, who bears disproportionate cost]
+
+**Flags**
+- 🔴 / 🟡 / 🟢 [flag + one sentence explanation]
+
+**Bottom Line**
+[One sentence: is this clear to ship, ship with mitigations, or needs more work]
+
+---
+
+## Notes
+
+The scan is designed to be fast. It is not a substitute for the ethics-council on high-stakes decisions — it is the filter that tells you whether you need one. A clean scan means you've thought clearly about impact. A flagged scan means you have a specific thing to address or escalate.
+
+Do not use the scan to *justify* a decision you've already made. Run it before you've committed.

package/skills/ethics-vendor-review/SKILL.md

@@ -0,0 +1,97 @@
+---
+name: ethics-vendor-review
+description: "Evaluate a third-party vendor, supplier, partner, or integration against ethical standards before signing a contract or shipping their code or service. Use when considering a new supplier, SaaS tool, data provider, API integration, outsourcing arrangement, or partnership. TRIGGERS: 'vet this vendor', 'is this supplier ok', 'ethics check on this partner', 'should we use this third party', any third-party evaluation where labour practices, data handling, business model, or alignment with values is a consideration."
+---
+
+# Ethics Vendor Vetting
+
+Every supplier relationship is a values choice. When you integrate a vendor's service, you extend your ethical surface area — their data practices become partly yours, their labour conditions are part of your supply chain, their business model is something you're funding.
+
+Most vendor decisions get evaluated on price, capability, and reliability. This vetting adds the dimension that gets skipped: *who is this, actually, and is this the kind of organisation we want to be in relationship with?*
+
+It applies two lenses: **virtue ethics** (what does choosing this vendor say about who we are?) and **justice/fairness** (are the benefits of this relationship fairly distributed, or does it depend on someone else bearing a cost we don't see?).
+
+---
+
+## Your Process
+
+**Step 1: Define the relationship**
+What is the vendor providing? What is the nature and depth of the dependency? How much leverage would they have once integrated? What data or access do they require?
+
+**Step 2: Virtue Ethics Assessment — Character and Alignment**
+
+Virtue ethics asks: what does this choice say about us? Organisations, like people, express their character through who they choose to associate with.
+
+Examine:
+- **Track record**: Has this vendor been involved in significant ethical violations — data breaches, regulatory penalties, labour disputes, discriminatory practices, deceptive advertising?
+- **Business model**: How do they actually make money? Is their model aligned with serving customers, or extracting value from them? Is there a misalignment between their incentives and ours?
+- **Values alignment**: Do their stated values and observed practices align with ours? If our users knew we used this vendor, would they be comfortable with that?
+- **Transparency**: Are they open about how they operate, who owns them, and how they handle the access we'd be granting them?
+
+**Step 3: Justice Assessment — Supply Chain and Structural Fairness**
+
+Justice asks: are the benefits of this arrangement fairly distributed, or is the pricing attractive because someone else is bearing an unacknowledged cost?
+
+Examine:
+- **Labour practices**: Where are people employed in connection with this vendor? Are those workers fairly compensated and treated? Is the pricing low because labour costs are being externalised to lower-wage jurisdictions or gig workers without protections?
+- **Data practices**: Who benefits from the data generated by your use of their service? Are users whose data flows through this vendor aware of and consenting to that flow?
+- **Market power**: Does integrating this vendor entrench monopolistic practices? Does it give disproportionate power to a player whose dominance is itself ethically questionable?
+- **Displacement**: Does this integration displace workers or practices that served people fairly?
+
+**Step 4: Dependency and Exit Assessment**
+- How difficult is it to exit this relationship once established?
+- What leverage do they gain over us once we're integrated?
+- What happens to our users if this vendor changes terms, is acquired, or shuts down?
+
+---
+
+## Human Check-in
+
+Before proceeding, ask the user:
+
+**How do you want to run this?**
+
+- **A) Full analysis** — complete all steps, reasoning shown throughout
+- **B) Key findings only** — bottom-line output, skip step-by-step detail
+- **C) Red flags only** — specific concerns, skip domains that look clean
+- **D) Refine the framing** — adjust what we're analyzing before starting
+
+Proceed based on their choice.
+
+## Output Format
+
+**Vendor Being Vetted:**
+[Name, what they provide, depth of integration]
+
+**Virtue Ethics Assessment**
+| Dimension | Finding | Concern Level |
+|---|---|---|
+| Track record | [finding] | 🔴 / 🟡 / 🟢 |
+| Business model alignment | [finding] | 🔴 / 🟡 / 🟢 |
+| Values alignment | [finding] | 🔴 / 🟡 / 🟢 |
+| Transparency | [finding] | 🔴 / 🟡 / 🟢 |
+
+**Justice Assessment**
+| Dimension | Finding | Concern Level |
+|---|---|---|
+| Labour practices | [finding] | 🔴 / 🟡 / 🟢 |
+| Data practices | [finding] | 🔴 / 🟡 / 🟢 |
+| Market power | [finding] | 🔴 / 🟡 / 🟢 |
+| Displacement | [finding] | 🔴 / 🟡 / 🟢 |
+
+**Dependency Risk**
+[1–2 sentences: lock-in risk, exit difficulty, leverage they gain]
+
+**Verdict**
+[Proceed / Proceed with conditions / Do not proceed — and why]
+
+**Conditions or Mitigations**
+- [Specific contractual, technical, or procedural safeguard if proceeding with concerns]
+
+---
+
+## Notes
+
+This vetting is based on available information — it is not a guarantee. For high-value or high-trust integrations (e.g. a vendor receiving sensitive user data), request their SOC 2 report, data processing agreement, and subprocessor list before proceeding.
+
+Where concerns are found but the business case is strong, name the trade-off explicitly rather than explaining it away. A conscious decision to proceed despite an ethical concern is better than a decision where the concern was never surfaced.

package/skills/game-theory/SKILL.md

@@ -0,0 +1,59 @@
+---
+name: game-theory
+description: "Routes to the right game-theory skill for your strategic situation. Triggers: 'game theory', 'strategic interaction', 'what will they do', 'payoff analysis', 'how do I think about this strategically', 'incentive design', 'cooperation problem', 'bidding strategy', any situation where the best choice depends on what others choose."
+---
+
+# Game Theory
+
+Your best move depends on what they'll do — and their best move depends on what you'll do. This interdependence is the defining feature of strategic interaction. Game theory provides formal tools for reasoning through it: mapping payoffs, finding stable outcomes, designing incentives, and analysing how cooperation forms and breaks down.
+
+---
+
+## Your Process
+
+**Step 1: Diagnose the interaction type**
+
+Read the situation and identify which of the following patterns applies:
+
+- **One-shot strategic choice** — players make a single decision simultaneously or sequentially, and payoffs depend on the combination of choices. You need to identify the stable outcome (equilibrium) and whether it's efficient.
+  → Use `/game-theory-equilibrium`
+
+- **Cooperation vs. defection** — each player is individually tempted to defect even though mutual cooperation would be better for everyone. You're facing a race to the bottom, collective action failure, or asking whether to cooperate or hold back.
+  → Use `/game-theory-prisoners-dilemma`
+
+- **Credibility problem** — you have private information and need others to believe your claim, or you're being told something and aren't sure whether to believe it. Cheap talk, costly signalling, commitment devices.
+  → Use `/game-theory-signaling`
+
+- **Rule design** — you're not playing the game, you're designing it. You want to create rules, incentives, or mechanisms that make players behave in a desired way — especially to elicit honest information or align self-interest with collective benefit.
+  → Use `/game-theory-mechanism-design`
+
+- **Long-run repeated relationship** — the same two or more parties will interact repeatedly over time. Reputation, trust, retaliation, and the shadow of the future are all active.
+  → Use `/game-theory-iterated`
+
+- **Coalition and fair division** — multiple players could form alliances and share gains. Which coalition will form? How should value be divided fairly? Who holds power?
+  → Use `/game-theory-coalition`
+
+- **Competitive bidding** — a structured auction or procurement process where you're either bidding or designing the process. How much to bid? How to avoid the winner's curse? How to design a revenue-maximising or efficient auction?
+  → Use `/game-theory-auction`
+
+**Step 2: Confirm and route**
+
+Present the diagnosis clearly — what kind of interaction this is and which skill fits — then ask:
+
+> *I've read this as [interaction type]. Does that match your situation, or is there a different aspect you'd like to focus on?*
+
+If confirmed, invoke the appropriate skill. If the situation spans multiple types (e.g., a cooperation problem inside a long-run relationship), note both and ask which dimension is most pressing.
+
+---
+
+## Important distinction
+
+Game theory provides formal payoff structure: it tells you what a rational player will do given the rules, payoffs, and other players. Strategy (see `/strategy`) provides contextual wisdom: how to position, when to act, how to use terrain and timing. They are complementary — use game theory to understand the structure of the interaction, and strategy to act effectively within it. When both apply, use game theory first to clarify what the incentives actually are, then strategy to decide how to play.
+
+---
+
+## Notes
+
+The category skills are: `/game-theory-equilibrium`, `/game-theory-prisoners-dilemma`, `/game-theory-signaling`, `/game-theory-mechanism-design`, `/game-theory-iterated`, `/game-theory-coalition`, `/game-theory-auction`.
+
+Related categories: `/strategy` (contextual wisdom for acting within games), `/decision` (single-player choice without strategic interaction), `/social` (power dynamics and coalition politics).

package/skills/game-theory-auction/SKILL.md

@@ -0,0 +1,96 @@
+---
+name: game-theory-auction
+description: "Analyses bidding strategy and auction design — how much to bid, how to avoid the winner's curse, and how to design revenue-maximising or efficient auctions. Triggers: 'bidding strategy', 'auction analysis', 'how much should I bid', 'auction design', 'winner's curse', 'sealed bid', 'how do I avoid overbidding', 'procurement auction', 'design an auction', 'competitive offer'."
+---
+
+# Game Theory: Auction Analysis
+
+William Vickrey's 1961 discovery is one of the cleanest results in economics: in a second-price sealed-bid auction, bidding your true value is a *dominant strategy* — the best move regardless of what others bid. The mechanism works because you pay the second-highest bid, not your own. Overbidding your true value doesn't help you (you might win but pay more than the item is worth); underbidding doesn't help you either (you might lose an item worth more than you'd have paid). So you bid your true value and let the second-highest bid determine the price. Vickrey received the Nobel Prize in 1996 for this result and related work.
+
+First-price auctions are strategically different: you pay what you bid, so optimal play requires *shading* your bid below your true value. The optimal shade depends on the number of competitors (shade more with more competitors) and the distribution of their valuations (shade more when competition is intense). In equilibrium, first-price and second-price auctions generate the same expected revenue — the revenue equivalence theorem — under standard conditions.
+
+The winner's curse is the most common failure mode in *common-value* auctions (where the item has an underlying objective value everyone is trying to estimate, rather than a private personal value). Winning means you bid highest, which means your estimate was the most optimistic among all bidders. In expectation, if you bid your unconditional estimate and win, you've overpaid — because winning reveals that you were the most optimistic, not the most accurate. The correct bid is your estimate *conditional on winning*, which is lower than your unconditional estimate.
+
+Paul Milgrom and Robert Wilson (Nobel 2020) developed the modern theory of auction design, including the simultaneous ascending auction used in FCC spectrum allocation — showing how auction design directly affects both revenue and efficient allocation.
+
+---
+
+## Your Process
+
+**Step 1: Auction type identification**
+Identify the auction format:
+- *First-price sealed bid*: all bidders submit one bid simultaneously; highest bid wins and pays their own bid
+- *Second-price sealed bid (Vickrey)*: highest bid wins but pays the second-highest bid
+- *Ascending (English)*: price rises until only one bidder remains; winner pays the final price
+- *Descending (Dutch)*: price falls from a high start until the first bidder claims the item at the current price
+- *Other*: procurement reverse auctions, combinatorial auctions, multi-round formats
+
+**Step 2: Private vs. common value**
+Determine the value structure:
+- *Private value*: each bidder has their own subjective valuation, independent of others. What the item is worth to you doesn't depend on what it's worth to others. Most art auctions, personal property sales.
+- *Common value*: the item has an underlying value that is the same for all bidders, but each has an imperfect estimate. Mineral rights, spectrum licenses, antique coins (where the value is objective but uncertain). *Winner's curse applies here.*
+- *Affiliated values*: intermediate case — your valuation is positively correlated with others'. Most real situations fall here.
+
+**Step 3: Optimal bidding strategy by type**
+
+*Second-price (Vickrey):*
+Bid your true value. This is a dominant strategy — it is best regardless of what others bid. No adjustment needed.
+
+*First-price sealed bid:*
+Shade your bid below your true value. As a rough rule with *n* symmetric bidders: bid approximately (n−1)/n × your true value. With 2 bidders, bid 50% of your value; with 4 bidders, 75%; with 10 bidders, 90%. In practice: bid higher when competition is intense (many bidders, strong demand) because the shading needs to be small to remain competitive.
+
+*Ascending (English):*
+Stay in the auction until the price exceeds your true value, then drop out. Never bid beyond your valuation. The private-value dominant strategy is identical in structure to the Vickrey auction.
+
+*Descending (Dutch):*
+Accept at the price that equals your true value. No advantage to waiting longer (you risk losing), and accepting earlier costs you money.
+
+**Step 4: Winner's curse adjustment** *(common value only)*
+In common-value settings, adjust your bid downward to correct for the selection bias of winning. Procedure:
+1. Estimate the item's true value using your available information
+2. Ask: conditional on winning (i.e., conditional on having submitted the highest bid), what does that tell me about the true value? Winning means everyone else estimated lower — your estimate is the most optimistic
+3. Revise your estimate downward by an amount that increases with the number of bidders and the uncertainty in your estimate
+4. Bid based on this revised, downward-adjusted estimate, not your initial estimate
+
+**Step 5: Auction design** *(for designers)*
+Apply the following principles:
+- *Reserve price*: set a floor below which you won't sell. Even in revenue-maximising design, a properly set reserve price increases expected revenue by eliminating sales at too-low prices
+- *Revenue equivalence*: under standard conditions, first-price and second-price formats generate equal expected revenue. Choose based on other considerations: second-price is simpler to reason about (dominant strategy bidding); first-price gives more price certainty upfront
+- *Efficiency vs. revenue*: second-price with no reserve is most efficient (item goes to highest-value bidder); adding a reserve or using a first-price format trades some efficiency for revenue
+- *Multi-unit and combinatorial*: when multiple items are sold and bidders value combinations, use a format that handles complementarities — the Vickrey-Clarke-Groves mechanism generalises the Vickrey auction to multi-item settings
+
+---
+
+## Output Format
+
+### Auction Analysis
+
+**Auction Type**
+[Format identified: first-price sealed bid / second-price / ascending / descending / other]
+
+**Value Structure**
+[Private value / common value / affiliated values — and the implication for strategy]
+
+**Optimal Bidding Strategy**
+[Specific recommended strategy for this auction type — dominant strategy or optimal shade with reasoning]
+
+**Winner's Curse Adjustment** *(common value only)*
+[Revised estimate after conditioning on winning, and the magnitude of the adjustment]
+
+**Specific Bid Recommendation**
+[If a bidder: the recommended bid with precise reasoning. If asked to evaluate a strategy: assessment of whether it is optimal]
+
+**Designer Recommendations** *(if applicable)*
+[Reserve price, format choice, revenue vs. efficiency trade-offs, multi-unit design considerations]
+
+---
+
+## Notes
+
+The revenue equivalence theorem holds under strong assumptions: symmetric bidders, private values, independent valuations, risk-neutral bidders. When these fail — bidders are asymmetric, values are affiliated, bidders are risk-averse — the revenue equivalence breaks down and format choice matters for revenue.
+
+The winner's curse is not irrational behaviour corrected by experience alone. It is a structural consequence of the selection process: winning an auction carries information about the item's value, and that information updates your estimate downward. Even sophisticated bidders overbid in novel common-value contexts.
+
+Auction analysis is a specialised application of mechanism design. For the general framework of designing rules to produce desired behaviour from self-interested players, use `/game-theory-mechanism-design`. For the equilibrium analysis of the auction (confirming which strategy is actually optimal for each player), use `/game-theory-equilibrium`.
+
+Pairs with: `/game-theory-mechanism-design` (the general framework for auction design), `/game-theory-equilibrium` (equilibrium analysis of specific auction formats), `/decision-expected-value` (when the bidding decision is primarily about expected value under uncertainty, not strategic interaction).

package/skills/game-theory-coalition/SKILL.md

@@ -0,0 +1,84 @@
+---
+name: game-theory-coalition
+description: "Analyses which coalitions form and how to divide gains fairly using cooperative game theory and the Shapley value. Triggers: 'coalition formation', 'who should we partner with', 'cooperative game theory', 'Shapley value', 'fair division', 'stable coalition', 'which coalition will form', 'equity split', 'power in a coalition', 'how should we divide this up fairly'."
+---
+
+# Game Theory: Coalition Analysis
+
+Cooperative game theory asks a different question from strategic (non-cooperative) game theory. Rather than asking what rational self-interested players will do when they can't coordinate, it asks: when players *can* form binding agreements and share gains, which coalitions will form, and how should the value be divided?
+
+Lloyd Shapley's answer to the division question — the Shapley value (1953, Nobel Prize 2012) — is remarkable for its mathematical precision and moral intuition. Each player's fair share is their average marginal contribution across all possible orderings of coalition formation. Formally: for each permutation of all players, calculate how much value player *i* adds when they join the coalition that has formed before them. Average this marginal contribution across all permutations. The result is the Shapley value — the uniquely fair allocation given four axioms: efficiency (the grand coalition's total value is fully distributed), symmetry (identical players receive equal shares), dummy (players who contribute nothing receive nothing), and additivity (allocations across independent games add correctly).
+
+The core captures coalition stability: an allocation is in the core if no subset of players can collectively do better by breaking away and forming their own coalition. If an allocation is in the core, no group has an incentive to defect — the grand coalition is stable. If the core is empty, no allocation is fully stable and some defection pressure is unavoidable.
+
+These two concepts are complementary but distinct. The Shapley value is always unique and always exists — it answers "what is fair?" The core may be empty — it answers "what is stable?"
+
+---
+
+## Your Process
+
+**Step 1: Player-value map**
+List all players and, for each possible coalition (every subset), specify the value that coalition can generate on its own. This is the characteristic function of the game — v(S) for every subset S. For small groups (3–4 players), enumerate all subsets. For larger groups, focus on the most relevant coalitions: the grand coalition, each individual player alone, and the likely competing sub-coalitions.
+
+**Step 2: Grand coalition assessment**
+Is the grand coalition (all players together) the most efficient arrangement? Check whether v(everyone) ≥ v(any subgroup) + v(remaining players). If yes, the grand coalition maximises total value and the question is only how to divide it. If no, some smaller coalition creates more value, and the question is which one forms.
+
+**Step 3: Shapley value calculation**
+For each player, calculate their average marginal contribution:
+- List all permutations of player ordering (for n players, there are n! permutations — for 3 players: 6; for 4 players: 24)
+- For each permutation, identify what coalition exists just before player i is added, and calculate v(coalition + i) − v(coalition)
+- Average this marginal contribution across all permutations
+- The result is player i's Shapley value
+
+For practical purposes with more than 4 players, compute the Shapley value for the players of primary interest using a representative subset of permutations, or use the formula: φᵢ(v) = Σ [|S|!(n−|S|−1)!/n!] × [v(S∪{i}) − v(S)] summed over all subsets S not containing i.
+
+**Step 4: Core analysis**
+An allocation x = (x₁, x₂, ..., xₙ) is in the core if:
+- It is efficient: Σxᵢ = v(all players)
+- No coalition can do better: for every subset S, Σᵢ∈S xᵢ ≥ v(S)
+
+Check whether the proposed allocation satisfies the blocking constraint for every relevant coalition. If no such allocation exists, the core is empty — identify which coalitions have the most credible defection threat.
+
+**Step 5: Stability analysis**
+Even with a core allocation, identify threats:
+- Which players are most tempted to defect to a subcoalition?
+- What external conditions (new opportunities, changing valuations, information revelations) could shift the characteristic function?
+- What governance or enforcement mechanisms could reinforce stability?
+
+---
+
+## Output Format
+
+### Coalition Analysis
+
+**Player-Value Map**
+[All relevant coalitions and the value each generates — v(S) for each subset S]
+
+**Grand Coalition Assessment**
+[Is the grand coalition efficient? Is total value maximised by full cooperation? Y/N and why]
+
+**Shapley Values**
+[Each player's Shapley value — their average marginal contribution and what this means for the fair allocation]
+
+**Core**
+[The set of stable allocations — or "core is empty" with identification of the most credible defection threat]
+
+**Stability Threats**
+[Which subcoalitions pose the greatest defection risk, under what conditions, and what would trigger instability]
+
+**Recommended Structure and Allocation**
+[The specific coalition and allocation that balances fairness (Shapley) with stability (core), with practical implementation notes]
+
+---
+
+## Notes
+
+The Shapley value answers "what is fair" — it does not predict what will happen. What actually happens depends on bargaining power, outside options, timing, and negotiating skill. The Shapley value is most useful as a reference point: a principled allocation that no player can argue violates fairness criteria.
+
+When the core is empty, no fully stable allocation exists. In practice this means: allocations will always face some coalition's objection. The goal shifts to finding the allocation with the *smallest* objection — the one that minimises the maximum advantage any blocking coalition could gain.
+
+For the rules governing how a coalition is formed and how players reveal their contributions — the design of the process rather than the analysis of the outcome — use `/game-theory-mechanism-design`.
+
+For analysis of the strategic (non-cooperative) interactions happening inside the coalition — after it forms — use `/game-theory-equilibrium` or `/game-theory-iterated`.
+
+Pairs with: `/strategy-alliance` (strategic and contextual dimension of partnership decisions), `/social-coalition-mapping` (the social and power dynamics of alliance building), `/game-theory-mechanism-design` (designing the process for fair coalition formation).

package/skills/game-theory-equilibrium/SKILL.md

@@ -0,0 +1,73 @@
+---
+name: game-theory-equilibrium
+description: "Finds the stable outcome of a strategic interaction — the point where no player can improve their result by changing their strategy alone. Triggers: 'Nash equilibrium', 'dominant strategy', 'what's the stable outcome', 'equilibrium analysis', 'payoff matrix', 'what will rational players do', 'where will this land'."
+---
+
+# Game Theory: Equilibrium Analysis
+
+John Nash's central insight: in any finite game, there exists at least one outcome — the Nash equilibrium — where no player can improve their payoff by unilaterally switching strategy, assuming all other players hold theirs. This is the point of stability; it's where rational play converges.
+
+The power of equilibrium analysis is not that it finds the best outcome. It finds the *actual* outcome — where unconstrained, rational, self-interested players end up. Many Nash equilibria are collectively inefficient: the classic prisoners' dilemma equilibrium is both stable and bad for everyone. Knowing where the game ends up is the prerequisite for deciding whether to play, to change the rules, or to engineer a better outcome.
+
+Thomas Schelling added a critical extension: when multiple equilibria exist, players coordinate on *focal points* — outcomes that feel natural or salient without explicit communication. The focal point is often obvious in context (the prominent location, the round number, the culturally expected choice) and determines which of several possible equilibria is reached.
+
+---
+
+## Your Process
+
+**Step 1: Map the players and strategies**
+Identify every player in the interaction. For each player, list their available strategies — the distinct actions they can choose. Keep the strategy set realistic: exhaustive but not so granular it becomes unmanageable.
+
+**Step 2: Build the payoff matrix**
+Construct a matrix showing each player's payoff for every combination of strategies. Fill in all cells. If payoffs are uncertain, use expected values. If precise payoffs aren't available, use ordinal rankings (best, good, neutral, bad, worst) — the analysis still holds.
+
+**Step 3: Find dominant strategies**
+A dominant strategy is one that is better for a player regardless of what others do. Check each player: is there a strategy that beats or ties all alternatives across every possible opponent choice? If a dominant strategy exists, rational players will always choose it — it simplifies the analysis substantially. Iterated elimination of dominated strategies can further reduce the game.
+
+**Step 4: Identify Nash equilibria**
+For each cell in the matrix, ask: given what the other player(s) are doing, would this player want to switch? If no player wants to switch — the combination is a Nash equilibrium. Mark all such outcomes. (A game may have one, several, or in mixed-strategy form, infinitely many equilibria.)
+
+**Step 5: Efficiency assessment**
+Evaluate the equilibrium outcome(s): is this collectively good, or is there a better outcome that rational play fails to reach? An outcome is *Pareto-inefficient* if there exists an alternative where everyone would be at least as well off and at least one player would be strictly better off. Name exactly why the efficient outcome is unreachable without external intervention.
+
+**Step 6: Multiple equilibria and coordination**
+If more than one Nash equilibrium exists, analyse: what determines which one is reached? Consider focal points (salience, cultural convention, historical precedent), communication (can players talk before choosing?), and coordination mechanisms (common knowledge, public commitments, third-party arbitration).
+
+---
+
+## Output Format
+
+### Equilibrium Analysis
+
+**Players and Strategies**
+[Each player and their available strategies]
+
+**Payoff Matrix**
+[Full matrix — all cells filled, with payoffs for each player at each strategy combination]
+
+**Dominant Strategies**
+[Which players have dominant strategies, and what they are — or "none" if absent]
+
+**Nash Equilibria**
+[All equilibria identified, with the strategy combination and payoffs at each]
+
+**Efficiency Assessment**
+[Efficient or Pareto-inefficient? If inefficient: which outcome would be better for all, and why rational play doesn't reach it]
+
+**Coordination Mechanism** *(if multiple equilibria)*
+[What determines which equilibrium is reached — focal points, communication, history]
+
+**Strategic Implication**
+[What this analysis means for the player asking — what to expect, what to watch, what leverage exists]
+
+---
+
+## Notes
+
+The Nash equilibrium describes where rational play leads — it does not prescribe what to do. If the equilibrium is bad (as in the prisoners' dilemma), the question becomes how to change the game. See `/game-theory-mechanism-design` for how to redesign rules so the equilibrium is efficient.
+
+The most common equilibrium failure is the prisoners' dilemma structure: dominant strategies lead both players to an outcome worse than the alternative. See `/game-theory-prisoners-dilemma` for dedicated analysis of cooperation failures.
+
+When players interact repeatedly, the equilibrium changes: future consequences make cooperation rational even when it isn't in the one-shot game. See `/game-theory-iterated` for repeated game analysis.
+
+Pairs with: `/strategy-positioning` (acting effectively given the equilibrium you've identified), `/game-theory-mechanism-design` (changing the game so the equilibrium is efficient), `/decision-criteria-weighting` (when the strategic dimension is secondary and this is primarily a personal choice).

package/skills/game-theory-iterated/SKILL.md

@@ -0,0 +1,83 @@
+---
+name: game-theory-iterated
+description: "Analyses long-run repeated interactions — how cooperation forms, how trust is built, how defection spirals start, and which strategies sustain cooperation. Triggers: 'repeated game', 'iterated strategy', 'long-run relationship', 'reputation effects', 'how do I sustain cooperation', 'tit for tat', 'shadow of the future', 'will they defect', 'ongoing relationship after betrayal', 'how do we recover from this'."
+---
+
+# Game Theory: Iterated Games
+
+Robert Axelrod's 1984 computer tournament is one of the most important results in social science. He invited game theorists to submit strategies for an iterated prisoners' dilemma — a repeated game where the same two players interact over and over. The simplest strategy submitted, Tit for Tat (cooperate on the first move, then do exactly what your opponent did on the previous move), won both rounds of the tournament, beating every more complex strategy.
+
+Why Tit for Tat wins: it is *nice* (starts by cooperating, never the first to defect), *retaliatory* (immediately punishes defection — there is no free lunch), *forgiving* (returns to cooperation as soon as the opponent does — does not hold grudges), and *clear* (the strategy is transparent and easy for the opponent to understand). Opponents who try to exploit it get punished; opponents who cooperate get rewarded. It is the most robust known strategy for sustained cooperation without trust.
+
+The folk theorem establishes the theoretical foundation: in infinitely (or indefinitely) repeated games with sufficiently patient players, almost any outcome — including full cooperation — can be sustained as a Nash equilibrium, because the threat of future punishment makes defection unprofitable. The key variable is the *discount factor* (how much players value future payoffs relative to present ones), and whether punishment is *credible* and *observable*.
+
+---
+
+## Your Process
+
+**Step 1: Stage game**
+Describe the single-period interaction — what are the two players' choices in any given round, and what are the payoffs? Map the four key payoffs: mutual cooperation (CC), mutual defection (DD), exploitation (one cooperates, one defects), and being exploited. This identifies whether repetition can help: if the stage game already has cooperation as a Nash equilibrium, repetition changes little. If cooperation is not a Nash equilibrium of the stage game, repetition may enable it.
+
+**Step 2: Is cooperation a stage-game equilibrium?**
+Check whether cooperation would be chosen in a one-shot interaction. If yes, the repeated game is not necessary to explain or enable it. If no (cooperation requires an ongoing relationship to be rational), proceed with the shadow-of-the-future analysis.
+
+**Step 3: Discount factor assessment**
+How much do the players value continued interaction? Assess:
+- *Time horizon*: is the relationship expected to continue indefinitely, or does it have a known end-point? (Known-endpoint problem: rational players defect on the last period, which unravels backward)
+- *Relationship value*: how important is continued cooperation to each player? How much would they lose if the relationship ended?
+- *Uncertainty about continuation*: what is the probability each period that the interaction continues? Higher probability → higher effective discount factor → cooperation more sustainable
+- *Impatience*: are either player under short-term pressure that discounts future benefits?
+
+**Step 4: Folk theorem conditions**
+Assess whether the conditions for sustained cooperation are met:
+- Discount factor is sufficiently high (both players value future interactions enough)
+- Defection is observable (players can detect when cooperation breaks down)
+- Punishment is credible (the punishing party would actually carry it out — it must be in their interest to do so)
+
+**Step 5: Strategy recommendation**
+Based on the discount factor and relationship context, recommend the best strategy from the following:
+
+- **Tit for Tat**: cooperate first, then mirror the opponent's last move. Best for stable, ongoing relationships where misunderstandings are rare.
+- **Generous Tit for Tat**: like Tit for Tat but occasionally cooperates even after a defection (with low probability). Better when there is noise — accidental defections or miscommunications — that could trigger unnecessary retaliation spirals.
+- **Grim Trigger**: cooperate until the opponent defects once, then defect forever. Maximum punishment credibility; best when the relationship is asymmetric and one defection is catastrophic. Risk: one mistake ends everything.
+- **Win-Stay, Lose-Shift**: if last round's outcome was good (for you), repeat your choice; if it was bad, switch. Simpler to execute, surprisingly robust in noisy environments.
+- **Unconditional cooperation**: only rational if you have strong external enforcement, the discount factor is extremely high, or you are trying to unilaterally rebuild a relationship.
+
+---
+
+## Output Format
+
+### Iterated Game Analysis
+
+**Stage Game**
+[One-shot interaction: choices, payoffs at each combination — CC / DD / CD / DC]
+
+**Cooperation in Stage Game**
+[Is cooperation a Nash equilibrium of the one-shot game? Yes / No — and why this matters for the iterated analysis]
+
+**Discount Factor Assessment**
+[Time horizon, relationship value, continuation probability, and impatience — overall rating: high (cooperation sustainable) / moderate (cooperation fragile) / low (cooperation unlikely)]
+
+**Folk Theorem Conditions**
+- Discount factor sufficient: [Yes / No / Marginal]
+- Defection observable: [Yes / No / Delayed — and by how much]
+- Punishment credible: [Yes / No — and what makes it credible or not]
+
+**Recommended Strategy**
+[Specific strategy from the set above, with the reasoning for this context]
+
+**Defection Spiral Warning Signs**
+[Specific indicators that cooperation is breaking down — what to watch for and how to respond before full defection occurs]
+
+**Recovery Path** *(if trust has already broken down)*
+[How to re-establish cooperation after defection — the sequence of signals, concessions, and credible commitments required]
+
+---
+
+## Notes
+
+The iterated analysis is the temporal complement to the one-shot prisoners' dilemma analysis. If you need to understand why the one-shot game produces defection in the first place, use `/game-theory-prisoners-dilemma`. The iterated skill focuses on how to sustain cooperation in ongoing relationships.
+
+The shadow of the future is the mechanism, not the effect. Future cooperation is only valuable if players believe the relationship will continue. Actions that reduce confidence in continuation — signalling you might exit, visibly shortening your time horizon, threatening to end the relationship — also reduce the incentive for today's cooperation. Handle with care.
+
+Pairs with: `/game-theory-prisoners-dilemma` (the one-shot structure this analysis builds on), `/game-theory-signaling` (in long-run relationships, reputation is a signal — how to maintain and repair it), `/strategy-timing` (when to cooperate, when to test, and when to act on defection).