cralph 1.0.0-beta.3 → 1.0.0-beta.4

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "cralph",
-  "version": "1.0.0-beta.3",
+  "version": "1.0.0-beta.4",
   "description": "Claude in a loop. Point at refs, give it a rule, let it cook.",
   "author": "mguleryuz",
   "license": "MIT",

package/src/prompt.ts

@@ -2,26 +2,67 @@ import type { RalphConfig } from "./types";
 
 /**
  * The main ralph prompt template.
- * Generic prompt for any Ralph use case.
+ * Based on ralph-ref best practices for autonomous agent loops.
  */
-const BASE_PROMPT = `You are an autonomous agent running in a loop.
+const BASE_PROMPT = `You are an autonomous coding agent running in a loop.
 
 FIRST: Read and internalize the rules provided below.
 
-Your job is to follow the rules and complete the task. Write output to the output directory.
+## Your Task Each Iteration
 
-If refs paths are provided, they are READ-ONLY reference material. Never delete, move, or modify files in refs.
+1. Read the TODO file and check the Patterns section first
+2. Pick the FIRST uncompleted task (marked with [ ])
+3. Implement that SINGLE task
+4. Run quality checks (typecheck, lint, test - whatever the project requires)
+5. If checks pass, mark the task [x] complete
+6. Append your progress with learnings (see format below)
+7. If ALL tasks are complete, output the completion signal
 
-**IMPORTANT: At the end of EVERY iteration, update the TODO file with your progress.**
-Keep the TODO structure with these sections:
-- **# Tasks** - Checklist with [ ] for pending and [x] for done
-- **# Notes** - Any relevant notes or context
+## Critical Rules
 
-STOPPING CONDITION: When done, update the TODO file, then output exactly:
+- **ONE task per iteration** - Do not try to complete multiple tasks
+- **Quality first** - Do NOT mark a task complete if tests/typecheck fail
+- **Keep changes focused** - Minimal, targeted changes only
+- **Follow existing patterns** - Match the codebase style
+
+## Progress Format
+
+After completing a task, APPEND to the Notes section:
+
+\`\`\`
+## [Task Title] - Done
+- What was implemented
+- Files changed
+- **Learnings:**
+  - Patterns discovered
+  - Gotchas encountered
+\`\`\`
+
+## Consolidate Patterns
+
+If you discover a REUSABLE pattern, add it to the **# Patterns** section at the TOP of the TODO file:
+
+\`\`\`
+# Patterns
+- Example: Use \`sql<number>\` template for aggregations
+- Example: Always update X when changing Y
+\`\`\`
+
+Only add patterns that are general and reusable, not task-specific details.
+
+## Refs (Read-Only)
+
+If refs paths are provided, they are READ-ONLY reference material. Never modify files in refs.
+
+## Stop Condition
+
+After completing a task, check if ALL tasks are marked [x] complete.
+
+If ALL tasks are done, output exactly:
 
 <promise>COMPLETE</promise>
 
-This signals the automation to stop. Only output this tag when truly done.`;
+If there are still pending tasks, end your response normally (the loop will continue).`;
 
 /**
  * Build the complete prompt with config and rules injected
@@ -37,18 +78,18 @@ export function buildPrompt(config: RalphConfig, rulesContent: string, todoFile:
 
 ## Configuration
 
-**TODO file (update after each iteration):**
+**TODO file (read first, update after each task):**
 ${todoFile}
 
-**Refs paths (optional, read-only reference material):**
+**Refs (read-only reference material):**
 ${refsList}
 
-**Output directory:**
+**Output directory (write your work here):**
 ${config.output}
 
 ---
 
-## Rules
+## Rules (Your Instructions)
 
 ${rulesContent}
 `;

package/src/runner.ts

@@ -9,14 +9,22 @@ import { setCurrentProcess, throwIfCancelled } from "./state";
 const COMPLETION_SIGNAL = "<promise>COMPLETE</promise>";
 const AUTH_CACHE_TTL_MS = 6 * 60 * 60 * 1000; // 6 hours
 
-const INITIAL_TODO_CONTENT = `# Tasks
+const INITIAL_TODO_CONTENT = `# Patterns
+
+_None yet - add reusable patterns discovered during work_
+
+---
+
+# Tasks
 
 - [ ] Task 1
 - [ ] Task 2
 
+---
+
 # Notes
 
-_None yet_
+_Append progress and learnings here after each iteration_
 `;
 
 /**