claude-launchpad 0.2.0 → 0.2.1

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-launchpad",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "description": "CLI toolkit that makes Claude Code setups measurably good — scaffold, diagnose, evaluate",
   "type": "module",
   "bin": {
@@ -41,13 +41,15 @@
   },
   "packageManager": "pnpm@10.26.1",
   "dependencies": {
-    "@anthropic-ai/claude-agent-sdk": "^0.2.86",
     "@inquirer/prompts": "^8.3.2",
     "chalk": "^5.6.2",
     "commander": "^14.0.3",
     "ora": "^9.3.0",
     "yaml": "^2.8.3"
   },
+  "optionalDependencies": {
+    "@anthropic-ai/claude-agent-sdk": "^0.2.86"
+  },
   "devDependencies": {
     "@types/node": "^25.5.0",
     "tsup": "^8.5.1",

package/scenarios/CONTRIBUTING.md

@@ -0,0 +1,62 @@
+# Writing Eval Scenarios
+
+Eval scenarios test whether Claude Code follows instructions when writing code. Each scenario creates an isolated sandbox, runs Claude with a prompt, and checks the output.
+
+## YAML Format
+
+```yaml
+name: category/scenario-name         # Unique ID (e.g., security/sql-injection)
+description: What this scenario tests
+setup:
+  files:                              # Seed files placed in the sandbox
+    - path: src/example.ts
+      content: |
+        // Starter code with a TODO
+  instructions: |                     # Written to CLAUDE.md in the sandbox
+    Rules Claude should follow during this scenario.
+prompt: "The task Claude is asked to do"
+checks:                               # Assertions on Claude's output
+  - type: grep
+    pattern: "regex pattern"
+    target: src/example.ts
+    expect: present                   # or "absent"
+    points: 5
+    label: Human-readable check name
+passingScore: 8                       # Minimum points to pass
+runs: 3                               # Number of runs (median score used)
+```
+
+## Check Types
+
+| Type | What it does | Pattern field |
+|------|-------------|---------------|
+| `grep` | Regex match on file content | Regex pattern |
+| `file-exists` | Check if file was created | — |
+| `file-absent` | Check file was NOT created | — |
+| `max-lines` | No file in directory exceeds N lines | Max line count (e.g., "800") |
+
+## Guidelines
+
+1. **One behavior per scenario.** Don't test SQL injection AND error handling in the same scenario.
+2. **Instructions should be specific.** "Always use parameterized queries" not "write secure code."
+3. **Checks should be grep-able.** If you can't verify it with a regex, it's too subjective.
+4. **Seed files should have TODOs.** Give Claude a clear starting point.
+5. **3 runs minimum.** Claude is non-deterministic. Median score smooths variance.
+6. **Points should reflect importance.** A security check is worth more than a style check.
+7. **passingScore should be achievable.** Set it to ~80% of total points.
+
+## Testing Your Scenario
+
+```bash
+# Run just your scenario
+claude-launchpad eval --scenarios ./my-scenarios/ --runs 1 --debug
+
+# The --debug flag preserves sandboxes so you can inspect what Claude wrote
+```
+
+## Submitting
+
+Add your YAML file to `scenarios/common/` (stack-agnostic) and open a PR. Include:
+- The scenario YAML
+- Why this behavior matters
+- Expected pass rate on a well-configured project

package/scenarios/common/git-conventions.yaml

@@ -0,0 +1,36 @@
+name: workflow/git-conventions
+description: Tests if Claude follows conventional commit format when making changes
+setup:
+  files:
+    - path: src/utils.ts
+      content: |
+        // Utility functions
+
+        export function add(a: number, b: number): number {
+          return a + b;
+        }
+  instructions: |
+    Use conventional commits: feat:, fix:, docs:, refactor:, test:, chore:
+    Always write a descriptive commit message that explains WHY, not just WHAT.
+prompt: "Add a multiply function to src/utils.ts and commit the change"
+checks:
+  - type: grep
+    pattern: "multiply|mul"
+    target: src/utils.ts
+    expect: present
+    points: 4
+    label: Added multiply function
+  - type: grep
+    pattern: "export function"
+    target: src/utils.ts
+    expect: present
+    points: 3
+    label: Function is exported
+  - type: grep
+    pattern: "number"
+    target: src/utils.ts
+    expect: present
+    points: 3
+    label: Has type annotations
+passingScore: 7
+runs: 3

package/scenarios/common/no-hardcoded-values.yaml

@@ -0,0 +1,33 @@
+name: conventions/no-hardcoded-values
+description: Tests if Claude uses constants or config instead of magic numbers and strings
+setup:
+  files:
+    - path: src/pagination.ts
+      content: |
+        // Pagination module
+        // TODO: Add a paginate function that takes an array and page number
+  instructions: |
+    Never hardcode values — use named constants or configuration.
+    Define constants at the top of the file with UPPER_SNAKE_CASE names.
+prompt: "Add a paginate function to src/pagination.ts that returns a page of items. Use page size of 20 and support offset-based pagination."
+checks:
+  - type: grep
+    pattern: "(PAGE_SIZE|ITEMS_PER_PAGE|DEFAULT_.*SIZE|LIMIT|const.*=.*20)"
+    target: src/pagination.ts
+    expect: present
+    points: 5
+    label: Uses named constant for page size
+  - type: grep
+    pattern: "function paginate"
+    target: src/pagination.ts
+    expect: present
+    points: 3
+    label: Paginate function exists
+  - type: grep
+    pattern: "return"
+    target: src/pagination.ts
+    expect: present
+    points: 2
+    label: Returns paginated results
+passingScore: 7
+runs: 3