Fountain Coach Gitowner-controlled · read only

Fountain-Coach.github.io.git · index.html

Fountain-Coach.github.io.git / index.html

revision 637d1931af76032f131a1ac2d3589afbae6af20a · complete file

<!doctype html>
<html lang="en">
  <head>
    <meta charset="utf-8" />
    <meta name="viewport" content="width=device-width, initial-scale=1" />
    <title>Fountain Coach — Extending Presence</title>
    <meta name="description" content="Fountain Coach builds technologies that extend presence across time." />
    <meta property="og:type" content="website">
    <meta property="og:site_name" content="Fountain Coach">
    <meta property="og:url" content="https://fountain-coach.github.io/">
    <meta property="og:image" content="/assets/img/logo.svg">
    <meta name="twitter:card" content="summary_large_image">
    <link rel="stylesheet" href="/assets/css/style.css?v=1" />
    <link rel="icon" href="/assets/img/logo.svg" type="image/svg+xml" />
  </head>
  <body>
    <main class="container">
      <nav class="nav">
        <a href="/" class="nav__home" aria-label="Home"></a>
        <a href="https://github.com/Fountain-Coach" class="nav__link" target="_blank" rel="noopener">GitHub</a>
        <a href="https://github.com/Fountain-Coach/midi2" class="nav__link" target="_blank" rel="noopener">MIDI 2</a>
        <a href="https://github.com/Fountain-Coach/FountainKit" class="nav__link" target="_blank" rel="noopener">FountainKit</a>
      </nav>

      <header class="brand">
        <div class="brand__image" role="img" aria-label="Fountain Coach branding"></div>
        <h1 class="brand__title">Fountain Coach</h1>
        <p class="brand__subtitle">Extending presence — technologies that hold space rather than fill it.</p>
      </header>

      <div class="page">
        <div class="paper">
          <div class="spread mirrored-readme">
            <!-- BEGIN MIRROR: ORG_PROFILE_README -->
<h1 dir="auto">Fountain Codex Instruction System (FCIS)</h1>
<h2 dir="auto">RFC 0001 — Orthogonal Instruction Architecture for LLM-Governed Repositories</h2>
<p dir="auto"><strong>Status:</strong> Adopted<br>
<strong>Category:</strong> Standards Track<br>
<strong>Organization:</strong> fountain.coach<br>
<strong>Applies to:</strong> All Codex-operated repositories<br>
<strong>Version:</strong> 1.0.0</p>
<hr>
<h2 dir="auto">Abstract</h2>
<p dir="auto">This document defines the <strong>Fountain Codex Instruction System (FCIS)</strong>, a normative, organization-wide standard for governing how OpenAI Codex is instructed, constrained, and enabled across repositories.</p>
<p dir="auto">FCIS establishes <strong>four orthogonal instruction layers</strong>:</p>
<ol dir="auto">
<li><code class="notranslate">AGENTS.md</code> — behavioral law</li>
<li><code class="notranslate">PLANS.md</code> — intent and execution reasoning</li>
<li>Skills (<code class="notranslate">SKILL.md</code>) — reusable execution techniques</li>
<li>MCP servers — external capabilities and data access</li>
</ol>
<p dir="auto">Strict separation of these layers is required to ensure <strong>predictability, auditability, and scalability</strong> of LLM-driven development.</p>
<hr>
<h2 dir="auto">Compliance &amp; Policies</h2>
<ul dir="auto">
<li>FCIS Apple Intelligence Compliance Specification (Preflight): <a href="../docs/FCIS-AIC-Preflight-RFC.md">FCIS-AIC-Preflight-RFC</a></li>
<li>Security policy: <a href="../.github/SECURITY.md">SECURITY</a></li>
<li>Governance: <a href="../.github/GOVERNANCE.md">GOVERNANCE</a></li>
<li>Contributing: <a href="../.github/CONTRIBUTING.md">CONTRIBUTING</a></li>
<li>Code of Conduct: <a href="../.github/CODE_OF_CONDUCT.md">CODE_OF_CONDUCT</a></li>
</ul>
<hr>
<h2 dir="auto">1. Terminology and Conventions</h2>
<p dir="auto">The key words <strong>MUST</strong>, <strong>MUST NOT</strong>, <strong>REQUIRED</strong>, <strong>SHALL</strong>, <strong>SHOULD</strong>, <strong>SHOULD NOT</strong>, <strong>RECOMMENDED</strong>, <strong>MAY</strong>, and <strong>OPTIONAL</strong> in this document are to be interpreted as described in RFC 2119.</p>
<hr>
<h2 dir="auto">2. Design Principle: Orthogonality</h2>
<h3 dir="auto">2.1 Definition</h3>
<p dir="auto">Within FCIS, <em>orthogonality</em> means:</p>
<blockquote>
<p dir="auto">Instruction layers are <strong>independent dimensions of responsibility</strong>, such that<br>
modifying one layer does not redefine, weaken, or invalidate the others.</p>
</blockquote>
<p dir="auto">Orthogonality is a <strong>hard requirement</strong>, not a stylistic preference.</p>
<h3 dir="auto">2.2 Motivation</h3>
<p dir="auto">LLM-based systems degrade rapidly when:</p>
<ul dir="auto">
<li>responsibilities overlap</li>
<li>priorities are implicit</li>
<li>instructions conflict probabilistically</li>
</ul>
<p dir="auto">Orthogonality prevents these failure modes by making responsibility <strong>explicit,<br>
exclusive, and reviewable</strong>.</p>
<hr>
<h2 dir="auto">3. Instruction Layers (Normative)</h2>
<p dir="auto">Each instruction layer answers <strong>exactly one question</strong>.</p>
<markdown-accessiblity-table><table role="table">
<thead>
<tr>
<th>Layer</th>
<th>Question</th>
</tr>
</thead>
<tbody>
<tr>
<td><code class="notranslate">AGENTS.md</code></td>
<td>What is allowed or forbidden?</td>
</tr>
<tr>
<td><code class="notranslate">PLANS.md</code></td>
<td>What is intended and why?</td>
</tr>
<tr>
<td>Skills</td>
<td>How is a task executed?</td>
</tr>
<tr>
<td>MCP</td>
<td>What capabilities are available?</td>
</tr>
</tbody>
</table></markdown-accessiblity-table>
<p dir="auto">No layer MAY answer another layer’s question.</p>
<hr>
<h2 dir="auto">4. AGENTS.md — Behavioral Law</h2>
<h3 dir="auto">4.1 Purpose</h3>
<p dir="auto"><code class="notranslate">AGENTS.md</code> <strong>MUST</strong> define non-negotiable behavioral constraints under which Codex<br>
operates in a repository.</p>
<h3 dir="auto">4.2 Mandatory Content</h3>
<p dir="auto"><code class="notranslate">AGENTS.md</code> <strong>MUST</strong> include:</p>
<ul dir="auto">
<li>invariants and prohibitions</li>
<li>safety and review requirements</li>
<li>routing rules for PLANS.md and skills</li>
</ul>
<h3 dir="auto">4.3 Prohibited Content</h3>
<p dir="auto"><code class="notranslate">AGENTS.md</code> <strong>MUST NOT</strong>:</p>
<ul dir="auto">
<li>contain step-by-step procedures</li>
<li>include tool configuration details</li>
<li>encode execution workflows</li>
</ul>
<p dir="auto">Violation constitutes a <strong>law breach</strong>.</p>
<hr>
<h2 dir="auto">5. PLANS.md — Intent and Reasoning Layer</h2>
<h3 dir="auto">5.1 Purpose</h3>
<p dir="auto"><code class="notranslate">PLANS.md</code> <strong>MUST</strong> capture the intent, structure, and risk management of multi-step work.</p>
<h3 dir="auto">5.2 Invocation Rule</h3>
<p dir="auto"><code class="notranslate">AGENTS.md</code> <strong>MUST</strong> define when PLANS.md is required.<br>
Codex <strong>SHALL NOT</strong> execute qualifying tasks without a plan.</p>
<h3 dir="auto">5.3 Required Structure</h3>
<p dir="auto">A plan <strong>SHOULD</strong> include:</p>
<ul dir="auto">
<li>goal</li>
<li>constraints</li>
<li>phased approach</li>
<li>test plan</li>
<li>rollback plan</li>
<li>definition of done</li>
</ul>
<hr>
<h2 dir="auto">6. Skills — Execution Technique Modules</h2>
<h3 dir="auto">6.1 Definition</h3>
<p dir="auto">A skill <strong>MUST</strong>:</p>
<ul dir="auto">
<li>exist in its own directory</li>
<li>contain exactly one <code class="notranslate">SKILL.md</code></li>
</ul>
<h3 dir="auto">6.2 Responsibilities</h3>
<p dir="auto">A skill <strong>MUST</strong> define:</p>
<ul dir="auto">
<li>when it applies</li>
<li>ordered execution steps</li>
<li>output guarantees</li>
</ul>
<h3 dir="auto">6.3 Constraints</h3>
<p dir="auto">Skills <strong>MUST NOT</strong>:</p>
<ul dir="auto">
<li>redefine behavioral law</li>
<li>replace PLANS.md for complex work</li>
<li>configure MCP servers</li>
</ul>
<hr>
<h2 dir="auto">7. MCP Servers — Capability Layer</h2>
<h3 dir="auto">7.1 Role</h3>
<p dir="auto">MCP servers <strong>MUST</strong> provide access to external tools or data only.</p>
<p dir="auto">They <strong>MUST NOT</strong>:</p>
<ul dir="auto">
<li>encode workflow logic</li>
<li>enforce policy</li>
<li>substitute for skills</li>
</ul>
<h3 dir="auto">7.2 Configuration</h3>
<p dir="auto">MCP configuration <strong>MUST</strong> occur outside repositories.<br>
Repositories <strong>MUST NOT</strong> rely on MCP presence for correctness.</p>
<hr>
<h2 dir="auto">8. Orthogonality Violations (Non-Exhaustive)</h2>
<p dir="auto">The following are <strong>spec violations</strong>:</p>
<ul dir="auto">
<li>procedures in <code class="notranslate">AGENTS.md</code></li>
<li>behavioral rules in skills</li>
<li>workflow logic in MCP</li>
<li>unplanned long-running tasks</li>
<li>duplicated responsibility across layers</li>
</ul>
<p dir="auto">Violations <strong>SHOULD</strong> block merges.</p>
<hr>
<h2 dir="auto">9. Standard Repository Layout</h2>
<pre class="notranslate"><code class="notranslate">repo-root/
  AGENTS.md
  PLANS.md
  .codex/
    skills/
      &lt;skill-name&gt;/
        SKILL.md
</code></pre>
<p dir="auto">User-level configuration:</p>
<pre class="notranslate"><code class="notranslate">~/.codex/config.toml
~/.codex/AGENTS.override.md
</code></pre>
<hr>
<h2 dir="auto">10. Compliance</h2>
<p dir="auto">A repository is FCIS-compliant if:</p>
<ul dir="auto">
<li>all applicable layers are present</li>
<li>responsibilities are orthogonal</li>
<li>plans exist for complex work</li>
<li>skills are modular and declarative</li>
</ul>
<hr>
<h2 dir="auto">11. Rationale</h2>
<p dir="auto">FCIS treats Codex as a <strong>compiler-like system</strong>:</p>
<ul dir="auto">
<li>law → intent → execution → capability<br>
This mirrors successful separation patterns from compilers, operating systems, and safety-critical automation.</li>
</ul>
<hr>
<h2 dir="auto">12. Security Considerations</h2>
<p dir="auto">Non-orthogonal instruction systems increase:</p>
<ul dir="auto">
<li>unintended behavior</li>
<li>silent policy drift</li>
<li>non-reproducible execution</li>
</ul>
<p dir="auto">FCIS reduces these risks through explicit separation and review.</p>
<hr>
<h2 dir="auto">13. Versioning</h2>
<ul dir="auto">
<li>RFC 0001 defines FCIS v1.0.0</li>
<li>Backward-incompatible changes require a new RFC</li>
</ul>
<hr>
<h2 dir="auto">14. Summary</h2>
<p dir="auto">FCIS establishes a <strong>contractual architecture</strong> for Codex:</p>
<ul dir="auto">
<li><code class="notranslate">AGENTS.md</code> defines law</li>
<li><code class="notranslate">PLANS.md</code> defines intent</li>
<li>Skills define execution</li>
<li>MCP defines capability</li>
</ul>
<p dir="auto">Adherence is mandatory for all fountain.coach Codex repositories.</p>

<!-- END MIRROR: ORG_PROFILE_README -->
          </div>
          <footer class="paper__footer">
            <p>© 2025 Fountain Coach</p>
          </footer>
        </div>
      </div>
    </main>
    <script defer src="/assets/js/mobile-tables.js"></script>
  </body>
</html>