game-theorist
AGENTS.md16 starsChanged 3 months agoMITView on GitHub
AGENTS.md est la source canonique de ces instructions. CLAUDE.md est un lien relatif vers ce fichier, ne pas l’éditer séparément. @./skills/game-theorist/SKILL.md Public skill repo for AI coding agents (Claude Code, Codex, Gemini CLI, Cursor, Windsurf, Cline). The single source of truth is skills/game-theorist/SKILL.md. Copies under .cursor/ and .windsurf/ are auto-generated by the GitHub Action sync-skill.yml; never edit them directly. Conventions: British English throughout. No em dashes. No "it depends" endings. Six phases, one at a time, dialogue-driven. Phase headers must…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository11 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- .cursor/rules/game-theorist.mdcCursor rule
- animateSkill
- auditSkill
- bolderSkill
- critiqueSkill
- game-theoristSkill
- impeccableSkill
- liveSkill
- polishSkill
- shapeSkill
The file
We split this AGENTS.md into the rules it gives an agent
8 rules. About 0 are common, 7 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 84
No runnable commands. There is no build step.
- Distinctive: line 88
No CLI invocation needed.
- Distinctive: line 93
No em dashes: Use commas, colons, or full stops.
- Very specific: line 103
Read SKILL.md → verify six phase headers exist in order.
- Distinctive: line 104
Run the skill in one agent with a real situation.
- Distinctive: line 129
Escalation lock detection: if both sides' threats mutually destroy value, recommend unilateral conditional de-escalation instead of "hold …
- Distinctive: line 131
Information cost check: before Phase 3, "Would revealing that I need this info cost more than it's worth?" If yes, proceed with incomplete …
- Distinctive: line 133
No forced clean answer: if no dominant strategy and no clean equilibrium, say so.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AGENTS.mdest la source canonique de ces instructions.CLAUDE.mdest un lien relatif vers ce fichier, ne pas l’éditer séparément.
@./skills/game-theorist/SKILL.md
game-theorist
Public skill repo for AI coding agents (Claude Code, Codex, Gemini CLI, Cursor, Windsurf, Cline).
The single source of truth is skills/game-theorist/SKILL.md. Copies under .cursor/ and
.windsurf/ are auto-generated by the GitHub Action sync-skill.yml; never edit them directly.
Conventions: British English throughout. No em dashes. No "it depends" endings. Six phases,
one at a time, dialogue-driven. Phase headers must read Phase N: Name exactly or the skill
breaks. See AGENTS.md for full workspace context and LESSONS.md for known edge cases.
Project context retained from the previous CLAUDE.md
game-theorist: workspace context
Skill overview
game-theorist is a six-phase game theory analysis skill for AI coding agents. It models strategic situations (negotiations, pricing wars, market entry, partnerships, fundraising) and walks through: Deconstruction → Incentive Mapping → Strategy Space → Equilibrium → Recommendation → Dynamic Adaptation. One phase at a time, dialogue-driven.
When to activate / redirect
Use game-theorist when: user asks "what's their incentive?", "how will they respond?",
"should I raise prices?", "how to negotiate?": any situation where another party's strategic
response is part of the problem. Slash: /game-theorist.
Do NOT use for: NPV/IRR calculations, email drafting, generic negotiation tips, pure educational "what does X mean?" questions, descriptive root-cause analysis ("why is churn high?"), PRD writing.
Exception: "Should I take this job offer?" → NOT game-theorist (simple yes/no). But "I'm getting competing offers from two companies" → IS game-theorist (independent actors with their own incentives).
What this project is
A skill-only repo for AI coding agents. Packages game-theorist and ships it to multiple platforms
(Claude Code, Codex, Gemini CLI, Cursor, Windsurf, Cline). No build step, no runtime. The "codebase"
is Markdown only. Public repo: https://github.com/saltandsilicon/game-theorist.
Code structure
skills/game-theorist/SKILL.md ← SINGLE SOURCE OF TRUTH. Never edit copies.
skills/game-theorist/references/
frameworks.md Eleven game theory frameworks
edge-cases.md Eight failure-mode diagnostics
worked-examples.md Real-world case studies
artefact-template.html HTML artefact template (filled by agent at Phase 6)
skills/game-theorist/tests/
evals.json Evaluation test cases
rubric.md Scoring rubric
docs/
frameworks.md Framework reference
resources.md Books, papers, courses
examples/
worked-examples.md Case studies
edge-cases.md Failure-mode diagnostics
analysis-example.html Example HTML artefact (OpenAI board crisis)
demos/
salary-negotiation-transcript.md Full six-phase conversation demo
wrong-call-correction-transcript.md Demo: correcting a bad initial call
assets/
game-theorist-knight.webp Branding
screenshots/ Screenshot PNGs for README
.claude-plugin/ Claude Code marketplace + plugin manifests
.github/workflows/sync-skill.yml Auto-syncs SKILL.md to .cursor/ and .windsurf/
.cursor/rules/game-theorist.mdc Cursor rule file
.mcp.json / AGENTS.md / GEMINI.md / .clinerules Per-agent reference docsKey convention: docs/ and examples/ mirror the corresponding files in
skills/game-theorist/references/. Edit the skills/ version; push to main and the GitHub Action
syncs to Cursor/Windsurf. Public docs/ and examples/ should be updated when SKILL.md changes.
Common commands
No runnable commands. There is no build step.
The HTML artefact is produced by the agent filling
skills/game-theorist/references/artefact-template.html at the end of a six-phase session.
No CLI invocation needed.
Coding patterns & conventions
- British English:
colour,behaviour,modelling,organisation,program. - No em dashes. Use commas, colons, or full stops.
- No filler: clean of patterns like "actually", "leverage", "Additionally", "Moreover", "As such".
- Concrete > abstract: name the framework, the move, the condition.
- No "it depends": end every analysis with a concrete move + explicit conditions.
- Phase headers must read
Phase N: Nameexactly.
Test strategy
No automated tests. Minimum verification:
- Read SKILL.md → verify six phase headers exist in order.
- Run the skill in one agent with a real situation.
- Phases fire in order, one at a time. If all six dump at once, the SKILL.md procedure is broken.
Active workflows
- Improve skill → edit
skills/game-theorist/SKILL.md, commit, push. Sync handles the rest. - Add framework → edit SKILL.md +
skills/game-theorist/references/frameworks.md+docs/frameworks.md. - Generate HTML artefact → agent fills
skills/game-theorist/references/artefact-template.htmlat Phase 6. No script to run.
Available skills
| Skill | Location | Activation hints |
| game-theorist | skills/game-theorist/SKILL.md | /game-theorist, "game theory", "what's their incentive", "how will they respond" |
| game-theorist (Cursor) | .cursor/skills/game-theorist/SKILL.md (synced) | Same |
| game-theorist (Windsurf) | .windsurf/skills/game-theorist/SKILL.md (synced) | Same |
Recurring patterns
- Phase-by-phase dialogue: one phase at a time, wait for user input. Core contract.
- BATNA-first: map every player's outside option before enumerating strategies.
- Dominance elimination: eliminate dominated strategies before recommending moves.
- Face-saving > face-attacking: diagnose off-ramps for public commitments, never attack positions directly.
- Coalition mapping: for org-level players, map veto holder, champion, opposition.
- Escalation lock detection: if both sides' threats mutually destroy value, recommend unilateral conditional de-escalation instead of "hold firm."
- Information cost check: before Phase 3, "Would revealing that I need this info cost more than it's worth?" If yes, proceed with incomplete data.
- No forced clean answer: if no dominant strategy and no clean equilibrium, say so.
Coding rules (do's and don'ts)
- [Do] Edit
skills/game-theorist/SKILL.md: single source of truth. - [Don't] Edit
.cursor/skills/or.windsurf/skills/: auto-generated by the sync workflow. Redirect edits to SKILL.md. - [Do] Keep the six phase headers in
Phase N: Nameformat. The skill breaks if they drift. - [Don't] Include API keys or secrets in committed files: the repo is public.
- [Do] Test skill end-to-end: phases must fire one at a time before marking changes complete.
Communication style notes
- British English in all text output.
- Action > discussion: recommend, then do. Concrete over abstract.
- No "it depends": end every analysis with a concrete move and explicit conditions.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

