site: 0.5.0 landing — prompt TUI, Claude, Muse 1.3, list/billed spend

site/docs/index.html

@@ -48,6 +48,8 @@ <h1>Agent Looper developer resources</h1>
         <li><a href="https://github.com/dancingteeth/agent-looper/blob/main/docs/dsh-plugin.md">DSH plugin</a> — DeepSeek Harness</li>
         <li><a href="https://github.com/dancingteeth/agent-looper/blob/main/docs/muse-runtime.md">Muse runtime</a> — Muse Code</li>
         <li><a href="https://github.com/dancingteeth/agent-looper/blob/main/docs/claude-runtime.md">Claude runtime</a> — Claude Code CLI</li>
+        <li><a href="https://github.com/dancingteeth/agent-looper/blob/main/docs/embed-api.md">Embed API</a> — contract for embedding Agent Looper</li>
+        <li><a href="https://github.com/dancingteeth/agent-looper/blob/main/SECURITY.md">Security</a> — reporting vulnerabilities</li>
         <li><a href="../llms.txt">Agent Looper llms.txt</a></li>
         <li><a href="../index.md">Agent Looper homepage (markdown)</a></li>
       </ul>

site/docs/index.md

@@ -12,6 +12,8 @@ Agent Looper is the npm package `@dancingteeth/agent-looper` and the CLI `agent-
 - DSH plugin (DeepSeek Harness): https://github.com/dancingteeth/agent-looper/blob/main/docs/dsh-plugin.md
 - Muse runtime (Muse Code): https://github.com/dancingteeth/agent-looper/blob/main/docs/muse-runtime.md
 - Claude runtime (Claude Code CLI): https://github.com/dancingteeth/agent-looper/blob/main/docs/claude-runtime.md
+- Embed API (contract for embedding Agent Looper): https://github.com/dancingteeth/agent-looper/blob/main/docs/embed-api.md
+- Security (reporting vulnerabilities): https://github.com/dancingteeth/agent-looper/blob/main/SECURITY.md
 - Agent Looper llms.txt: https://looper.dancingteeth.net/llms.txt
 - Agent Looper homepage (markdown): https://looper.dancingteeth.net/index.md
 

site/harnesses/index.html

@@ -230,7 +230,7 @@ <h3>Claude</h3>
                 <span class="harness-card__pill"><code>--runtime claude</code></span>
               </div>
             </header>
-            <p class="harness-card__job">PATH <code>claude</code> CLI — Claude Code subscription. Spawn uses <code>--safe-mode</code>. Not on <code>costPreset</code> minmax.</p>
+            <p class="harness-card__job">PATH <code>claude</code> CLI — Claude Code subscription. <code>--safe-mode</code> so the harness prompt is the only instruction source (strips project hooks and auto-memory). Not on <code>costPreset</code> minmax.</p>
             <p class="harness-card__chips">
               <span class="harness-chip">Worker <code>sonnet</code> → <code>opus</code></span>
               <span class="harness-chip">Judge any runtime, optional</span>

site/harnesses/index.md

@@ -86,7 +86,7 @@ DeepSeek Harness CLI — worker is `dsh --profile headless`; `dsh-agent-looper`
 
 ### Claude (`--runtime claude`)
 
-PATH `claude` CLI — Claude Code subscription. Spawn uses `--safe-mode`. Not on `costPreset` minmax. See [docs/claude-runtime.md](https://github.com/dancingteeth/agent-looper/blob/main/docs/claude-runtime.md).
+PATH `claude` CLI — Claude Code subscription. `--safe-mode` so the harness prompt is the only instruction source (strips project hooks and auto-memory). Not on `costPreset` minmax. See [docs/claude-runtime.md](https://github.com/dancingteeth/agent-looper/blob/main/docs/claude-runtime.md).
 
 - Worker: `sonnet` → `opus`
 - Judge: any runtime, optional

site/index.html

@@ -147,7 +147,23 @@
             "name": "What do I get when a loop finishes?",
             "acceptedAnswer": {
               "@type": "Answer",
-              "text": "A report card, not a chat dump. How often the check sent work back, whether the worker hung, whether the judge bounced it, what it cost, and whether it needed you."
+              "text": "A report card, not a chat dump. How often the check sent work back, whether the worker hung, whether the judge bounced it, list and billed spend when they differ, and whether it needed you."
+            }
+          },
+          {
+            "@type": "Question",
+            "name": "How do I start a loop from an idea?",
+            "acceptedAnswer": {
+              "@type": "Answer",
+              "text": "Type an idea, freeze a loop. agent-loop-prompt is an Ink TUI: you describe the goal; the judge (not the worker) drafts GOAL.md and verify.sh; freeze lint rejects gameable greps; you confirm; then the watch TUI grinds. Resume lines keep Doppler when the shell was started that way. Optional preview after green is trust-gated and is not executed by agent-loop run."
+            }
+          },
+          {
+            "@type": "Question",
+            "name": "What do the spend numbers mean?",
+            "acceptedAnswer": {
+              "@type": "Answer",
+              "text": "Watch and the report card show two numbers when they differ: list (public API rates, including prompt-cache) and billed (what the runtime invoice says). $0 on a subscription quota is billed $0, not free. Budget caps use billed when you are on PAYG and list when the invoice is $0."
             }
           }
         ]
@@ -179,18 +195,12 @@
           {
             "@type": "HowToStep",
             "position": 3,
-            "name": "Initialize a loop",
-            "text": "Run pnpm exec agent-loop-init to scaffold the loop folder."
+            "name": "Freeze a goal and check",
+            "text": "Prefer pnpm exec agent-loop-prompt --out .cursor/loops/my-task — describe your idea; the judge drafts GOAL.md and verify.sh; confirm freeze. Or run pnpm exec agent-loop-init and edit GOAL.md and verify.sh until bash verify.sh is an honest pass/fail."
           },
           {
             "@type": "HowToStep",
             "position": 4,
-            "name": "Write the finish line",
-            "text": "Edit GOAL.md and verify.sh until bash verify.sh is an honest determined check."
-          },
-          {
-            "@type": "HowToStep",
-            "position": 5,
             "name": "Run until green",
             "text": "Run pnpm exec agent-loop run .cursor/loops/my-task --runtime cursor --review-gate."
           }
@@ -355,7 +365,19 @@ <h2>How do I know if a loop is still alive?</h2>
           <article class="tui-answer tui-answer--report">
             <h2>What do I get when a loop finishes?</h2>
             <p>
-              A report card, not a chat dump. How often the check sent work back, whether the worker hung, whether the judge bounced it, what it cost, and whether it needed you.
+              A report card, not a chat dump. How often the check sent work back, whether the worker hung, whether the judge bounced it, <strong>list</strong> and <strong>billed</strong> spend when they differ, and whether it needed you.
+            </p>
+          </article>
+          <article class="tui-answer tui-answer--prompt">
+            <h2>How do I start a loop from an idea?</h2>
+            <p>
+              Type an idea, freeze a loop. <code>agent-loop-prompt</code> is an Ink TUI: you describe the goal; the <strong>judge</strong> (not the worker) drafts <code>GOAL.md</code> and <code>verify.sh</code>; freeze lint rejects gameable greps; you confirm; then the watch TUI grinds. Resume lines keep Doppler when the shell was started that way. Optional <code>preview</code> after green is trust-gated and is <strong>not</strong> executed by <code>agent-loop run</code>.
+            </p>
+          </article>
+          <article class="tui-answer tui-answer--spend">
+            <h2>What do the spend numbers mean?</h2>
+            <p>
+              Watch and the report card show two numbers when they differ: <strong>list</strong> (public API rates, including prompt-cache) and <strong>billed</strong> (what the runtime invoice says). <code>$0</code> on a subscription quota is billed <code>$0</code>, not “free.” Budget caps use billed when you are on PAYG and list when the invoice is <code>$0</code>.
             </p>
           </article>
         </div>
@@ -432,7 +454,7 @@ <h2>How is Agent Looper different from looping in chat?</h2>
       <div class="wrap">
         <h2>Open spine, not rented brain</h2>
         <p class="section-intro">
-          MIT. Inspectable. Models stay whoever you already pay — Cursor, OpenCode, OpenRouter, Codex, DSH, Muse.
+          MIT. Inspectable. Models stay whoever you already pay — Cursor, OpenCode, OpenRouter, Codex, DSH, Muse, Claude.
           What you own is the loop: frozen <code>GOAL.md</code>, measurable <code>verify.sh</code>,
           sparse <code>REVIEWS.md</code> / <code>AGENTS.md</code>, and orchestration that keeps workers fresh and spend bounded.
         </p>
@@ -448,8 +470,7 @@ <h2>Install Agent Looper</h2>
         <ol class="howto-steps">
           <li>Add <code>@dancingteeth/agent-looper</code> and the SDK for your coding agent (Cursor: <code>@cursor/sdk</code>).</li>
           <li>Export an API key, or run under Doppler.</li>
-          <li>Run <code>pnpm exec agent-loop-init</code>.</li>
-          <li>Edit <code>GOAL.md</code> and <code>verify.sh</code> until <code>bash verify.sh</code> is an honest determined check.</li>
+          <li>Prefer <code>pnpm exec agent-loop-prompt --out .cursor/loops/my-task</code> — describe your idea; the judge drafts <code>GOAL.md</code> and <code>verify.sh</code>; confirm freeze. Or <code>pnpm exec agent-loop-init</code> and edit the goal and check yourself.</li>
           <li>Run <code>pnpm exec agent-loop run</code> until that check is green.</li>
         </ol>
         <div class="install-switcher">
@@ -469,25 +490,24 @@ <h2>Install Agent Looper</h2>
 
 Requirements: Node 22+ and pnpm.
 
-1. Add @dancingteeth/agent-looper and the SDK for the coding agent on this machine. Cursor uses @cursor/sdk. If you're not sure, detect what's installed (Cursor, Cline, OpenCode, Pi, Codex, DSH, Muse).
+1. Add @dancingteeth/agent-looper and the SDK for the coding agent on this machine. Cursor uses @cursor/sdk. If you're not sure, detect what's installed (Cursor, Cline, OpenCode, Pi, Codex, DSH, Muse, Claude).
 2. Keep @dancingteeth/agent-looper in package.json even if you gitignore `.cursor/loops`. Don't only npx it.
 3. Use an API key already on this machine, or run under Doppler. Don't invent keys.
-4. Run: pnpm exec agent-loop-init
-5. Write GOAL.md for the feature and freeze it. Don't rewrite the goal mid-loop.
-6. Write a determined check until `bash verify.sh` is an honest pass/fail that the feature actually works. The model does not get to say done.
-7. Start the loop:
-   pnpm exec agent-loop run .cursor/loops/&lt;task&gt; --runtime &lt;cursor|cline|opencode|pi|codex|dsh|muse&gt; --review-gate
-8. Review skill: blockers come back for the next iteration. Advisory is my call. Don't loop on subjective taste.
+4. Prefer: pnpm exec agent-loop-prompt --out .cursor/loops/&lt;task&gt;
+   — describe the idea; the judge (not the worker) drafts GOAL.md and verify.sh; freeze lint rejects gameable greps; confirm; then the watch TUI grinds.
+   Or manually: pnpm exec agent-loop-init, write GOAL.md, and a determined check until `bash verify.sh` is an honest pass/fail. Don't rewrite the goal mid-loop.
+5. Start the loop:
+   pnpm exec agent-loop run .cursor/loops/&lt;task&gt; --runtime &lt;cursor|cline|opencode|pi|codex|dsh|muse|claude&gt; --review-gate
+6. Review skill: blockers come back for the next iteration. Advisory is my call. Don't loop on subjective taste.
 
 Cursor example (swap the SDK, key, and --runtime if this machine isn't Cursor):
 
 pnpm add -D @dancingteeth/agent-looper @cursor/sdk
 
 export CURSOR_API_KEY=…   # or: doppler run -- …
 
-pnpm exec agent-loop-init
-# edit .cursor/loops/my-task/GOAL.md
-# edit verify.sh until `bash .cursor/loops/my-task/verify.sh` is honest
+pnpm exec agent-loop-prompt --out .cursor/loops/my-task
+# or: pnpm exec agent-loop-init and edit GOAL.md + verify.sh yourself
 
 pnpm exec agent-loop run .cursor/loops/my-task --runtime cursor --review-gate</code></pre>
             </div>
@@ -503,9 +523,9 @@ <h2>Install Agent Looper</h2>
 
 export CURSOR_API_KEY=…   # or: doppler run -- …
 
-pnpm exec agent-loop-init
-# edit .cursor/loops/my-task/GOAL.md
-# edit verify.sh until `bash .cursor/loops/my-task/verify.sh` is honest
+pnpm exec agent-loop-prompt --out .cursor/loops/my-task
+# describe your idea; the judge drafts GOAL.md + verify.sh — confirm freeze, then watch
+# or: pnpm exec agent-loop-init and edit GOAL.md + verify.sh until `bash .cursor/loops/my-task/verify.sh` is honest
 
 pnpm exec agent-loop run .cursor/loops/my-task --runtime cursor --review-gate</code></pre>
             </div>

site/index.md

@@ -32,7 +32,15 @@ The IDE job list will lie. Ask your agent to use the check-running-loops skill.
 
 ## What do I get when a loop finishes?
 
-A report card, not a chat dump. How often the check sent work back, whether the worker hung, whether the judge bounced it, what it cost, and whether it needed you.
+A report card, not a chat dump. How often the check sent work back, whether the worker hung, whether the judge bounced it, **list** and **billed** spend when they differ, and whether it needed you.
+
+## How do I start a loop from an idea?
+
+Type an idea, freeze a loop. `agent-loop-prompt` is an Ink TUI: you describe the goal; the **judge** (not the worker) drafts `GOAL.md` and `verify.sh`; freeze lint rejects gameable greps; you confirm; then the watch TUI grinds. Resume lines keep Doppler when the shell was started that way. Optional `preview` after green is trust-gated and is **not** executed by `agent-loop run`.
+
+## What do the spend numbers mean?
+
+Watch and the report card show two numbers when they differ: **list** (public API rates, including prompt-cache) and **billed** (what the runtime invoice says). `$0` on a subscription quota is billed `$0`, not “free.” Budget caps use billed when you are on PAYG and list when the invoice is `$0`.
 
 ## How is Agent Looper different from looping in chat?
 
@@ -46,9 +54,8 @@ Node 22+, pnpm.
 
 1. Add `@dancingteeth/agent-looper` and the SDK for your coding agent (Cursor: `@cursor/sdk`).
 2. Export an API key, or run under Doppler.
-3. Run `pnpm exec agent-loop-init`.
-4. Edit `GOAL.md` and `verify.sh` until `bash verify.sh` is an honest determined check.
-5. Run `pnpm exec agent-loop run` until that check is green.
+3. Prefer `pnpm exec agent-loop-prompt --out .cursor/loops/my-task` — describe your idea; the judge drafts `GOAL.md` and `verify.sh`; confirm freeze. Or `pnpm exec agent-loop-init` and edit the goal and check yourself.
+4. Run `pnpm exec agent-loop run` until that check is green.
 
 ### For agent
 
@@ -60,22 +67,21 @@ Requirements: Node 22+ and pnpm.
 1. Add @dancingteeth/agent-looper and the SDK for the coding agent on this machine. Cursor uses @cursor/sdk. If you're not sure, detect what's installed (Cursor, Cline, OpenCode, Pi, Codex, DSH, Muse, Claude).
 2. Keep @dancingteeth/agent-looper in package.json even if you gitignore `.cursor/loops`. Don't only npx it.
 3. Use an API key already on this machine, or run under Doppler. Don't invent keys.
-4. Run: pnpm exec agent-loop-init
-5. Write GOAL.md for the feature and freeze it. Don't rewrite the goal mid-loop.
-6. Write a determined check until `bash verify.sh` is an honest pass/fail that the feature actually works. The model does not get to say done.
-7. Start the loop:
-   pnpm exec agent-loop run .cursor/loops/<task> --runtime <cursor|cline|opencode|pi|codex|dsh|muse> --review-gate
-8. Review skill: blockers come back for the next iteration. Advisory is my call. Don't loop on subjective taste.
+4. Prefer: pnpm exec agent-loop-prompt --out .cursor/loops/<task>
+   — describe the idea; the judge (not the worker) drafts GOAL.md and verify.sh; freeze lint rejects gameable greps; confirm; then the watch TUI grinds.
+   Or manually: pnpm exec agent-loop-init, write GOAL.md, and a determined check until `bash verify.sh` is an honest pass/fail. Don't rewrite the goal mid-loop.
+5. Start the loop:
+   pnpm exec agent-loop run .cursor/loops/<task> --runtime <cursor|cline|opencode|pi|codex|dsh|muse|claude> --review-gate
+6. Review skill: blockers come back for the next iteration. Advisory is my call. Don't loop on subjective taste.
 
 Cursor example (swap the SDK, key, and --runtime if this machine isn't Cursor):
 
 pnpm add -D @dancingteeth/agent-looper @cursor/sdk
 
 export CURSOR_API_KEY=…   # or: doppler run -- …
 
-pnpm exec agent-loop-init
-# edit .cursor/loops/my-task/GOAL.md
-# edit verify.sh until `bash .cursor/loops/my-task/verify.sh` is honest
+pnpm exec agent-loop-prompt --out .cursor/loops/my-task
+# or: pnpm exec agent-loop-init and edit GOAL.md + verify.sh yourself
 
 pnpm exec agent-loop run .cursor/loops/my-task --runtime cursor --review-gate
 ```
@@ -88,9 +94,9 @@ pnpm add -D @dancingteeth/agent-looper @cursor/sdk
 
 export CURSOR_API_KEY=…   # or: doppler run -- …
 
-pnpm exec agent-loop-init
-# edit .cursor/loops/my-task/GOAL.md
-# edit verify.sh until `bash .cursor/loops/my-task/verify.sh` is honest
+pnpm exec agent-loop-prompt --out .cursor/loops/my-task
+# describe your idea; the judge drafts GOAL.md + verify.sh — confirm freeze, then watch
+# or: pnpm exec agent-loop-init and edit GOAL.md + verify.sh until `bash .cursor/loops/my-task/verify.sh` is honest
 
 pnpm exec agent-loop run .cursor/loops/my-task --runtime cursor --review-gate
 ```

site/llms.txt

@@ -6,7 +6,7 @@ When to use Agent Looper: a human wants a feature finished without babysitting C
 
 When not to use Agent Looper: there is no determined check (pure taste with no script or vision gate); you need a team-ops factory, shared org cloud, or multi-repo orchestration; you want the model to grade itself as “done.”
 
-How an agent should call it: install `@dancingteeth/agent-looper`, run `pnpm exec agent-loop-init`, freeze `.cursor/loops/<task>/GOAL.md` and `verify.sh`, then `pnpm exec agent-loop run .cursor/loops/<task> --runtime <cursor|cline|opencode|pi|codex|dsh|muse|claude>`. Optional `--review-gate` re-opens the loop only on blocking review findings. CLI binary: `agent-loop`.
+How an agent should call it: install `@dancingteeth/agent-looper`, prefer `pnpm exec agent-loop-prompt --out .cursor/loops/<task>` (describe the idea; judge drafts GOAL/verify; confirm freeze) or run `pnpm exec agent-loop-init` and edit `GOAL.md` and `verify.sh`, then `pnpm exec agent-loop run .cursor/loops/<task> --runtime <cursor|cline|opencode|pi|codex|dsh|muse|claude>`. Optional `--review-gate` re-opens the loop only on blocking review findings. CLI binary: `agent-loop`.
 
 ## Developer resources
 
@@ -21,6 +21,8 @@ How an agent should call it: install `@dancingteeth/agent-looper`, run `pnpm exe
 - [DSH plugin](https://github.com/dancingteeth/agent-looper/blob/main/docs/dsh-plugin.md): DeepSeek Harness companion
 - [Muse runtime](https://github.com/dancingteeth/agent-looper/blob/main/docs/muse-runtime.md): Muse Code CLI
 - [Claude runtime](https://github.com/dancingteeth/agent-looper/blob/main/docs/claude-runtime.md): Claude Code CLI (`--safe-mode`)
+- [Embed API](https://github.com/dancingteeth/agent-looper/blob/main/docs/embed-api.md): contract for embedding Agent Looper
+- [Security](https://github.com/dancingteeth/agent-looper/blob/main/SECURITY.md): reporting vulnerabilities
 - [Harnesses — Agent Looper worker runtimes](https://looper.dancingteeth.net/harnesses/index.md)
 
 ## Optional