@cephalization/math 0.3.0 → 0.3.2

package/README.md

@@ -24,9 +24,22 @@ curl -fsSL https://bun.sh/install | bash
 ```
 
 Why Bun?
+
 - This tool is written in TypeScript and uses Bun's native TypeScript execution (no compilation step)
 - The CLI uses a `#!/usr/bin/env bun` shebang for direct execution
-- Bun's speed makes the agent loop faster and more responsive
+
+
+**[OpenCode](https://opencode.ai) is required** to run this tool.
+
+```bash
+# Install OpenCode
+curl -fsSL https://opencode.ai/install | bash
+```
+
+Why OpenCode?
+
+- OpenCode provides a consistent and reliable interface for running the agent loop
+- It supports many models, is easy to use, and is free to use
 
 ## Installation
 
@@ -94,6 +107,15 @@ Options:
 - `--max-iterations <n>` - Safety limit (default: 100)
 - `--pause <seconds>` - Pause between iterations (default: 3)
 
+Iteratively run the agent loop until all tasks are complete. Each iteration will:
+
+- Read the `TASKS.md` file to find the next task to complete
+- Invoke the agent with the `PROMPT.md` file and the `TASKS.md` file
+- The agent will complete the task and update the `TASKS.md` file
+- The agent will log its learnings to the `LEARNINGS.md` file
+- The agent will commit the changes to the repository
+- The agent will exit
+
 ### Check status
 
 ```bash
@@ -178,7 +200,7 @@ Signs accumulate over time, making the agent increasingly reliable.
 
 | Variable | Default | Description |
 |----------|---------|-------------|
-| `MODEL`  | `anthropic/claude-opus-4-20250514` | Model to use |
+| `MODEL`  | `anthropic/claude-opus-4-5` | Model to use |
 
 ## Credits
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@cephalization/math",
-  "version": "0.3.0",
+  "version": "0.3.2",
   "author": "Tony Powell <powell.anthonyd@proton.me>",
   "access": "public",
   "repository": {

package/src/plan.ts

@@ -46,6 +46,13 @@ FIRST, examine the project to understand its technology stack:
 - Identify the linter/formatter (e.g., eslint, prettier, rustfmt, black)
 - Note any existing scripts or commands defined in the project
 
+NOTE:
+- Distinguish between frontend and backend tooling. 
+- For frontend tooling, check if 'agent-browser' cli is installed. If so, it should be used for validating visual changes.
+- Distinguish between unit tests and integration tests.
+- Prefer code generation tools over manual coding when there are scripts to generate code (e.g. relay, openapi-codegen, etc).
+- If typed languages are used, prefer concrete, safe types (e.g. unknown instead of any in TypeScript)
+
 ## Step 2: Plan the Tasks
 
 Break the user's goal into discrete, implementable tasks using this format:

package/src/templates.ts

@@ -10,7 +10,7 @@ Implement ONE task from TASKS.md, test it, commit it, log your learnings, then E
 
 1. **Read TASKS.md** - Find the first task with \`status: pending\` where ALL dependencies have \`status: complete\`
 2. **Mark in_progress** - Update the task's status to \`in_progress\` in TASKS.md
-3. **Implement** - Write the code following the project's patterns
+3. **Implement** - Write the code following the project's patterns. Use prior learnings to your advantage.
 4. **Write tests** - For behavioral code changes, create unit tests in the appropriate directory. Skip for documentation-only tasks.
 5. **Run tests** - Execute tests from the package directory (ensures existing tests still pass)
 6. **Fix failures** - If tests fail, debug and fix. DO NOT PROCEED WITH FAILING TESTS.