bms-speckit-plugin 2.0.0 → 2.1.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bms-speckit-plugin",
-  "version": "2.0.0",
+  "version": "2.1.0",
   "description": "Single-command automated development pipeline: /bms-speckit takes requirements and runs constitution → specify → plan → tasks → analyze → implement → verify",
   "files": [
     ".claude-plugin/",

package/skills/speckit/SKILL.md

@@ -11,10 +11,6 @@ allowed-tools: Read, Write, Edit, Bash, Glob, Grep, Skill, Agent, WebSearch, Web
 Takes a single requirement and runs the complete engineering workflow automatically:
 **constitution → specify → plan → tasks → analyze → implement → verify**
 
-## Input Requirement
-
-$ARGUMENTS
-
 ## Engineering Constitution
 
 All work follows these principles:
@@ -38,7 +34,8 @@ All work follows these principles:
 - Commit: `feat: establish engineering constitution`
 
 ### Step 2: Create Specification
-- Analyze the requirement: $ARGUMENTS
+- The user's requirement is provided as the command argument: $ARGUMENTS
+- Analyze this requirement thoroughly
 - If database access is needed and MCP tools are available (list_tables, describe_table, query), explore the actual database schema
 - Create `specs/{feature-name}/specification.md` containing:
   - Overview and user requirements
@@ -91,30 +88,24 @@ All work follows these principles:
 - Fix any issues found in tasks.md
 - Commit if changes made
 
-### Step 6: Implement ALL Tasks
-For each pending task, execute the TDD cycle:
-
-1. Update task status to `in_progress` in tasks.md
-2. **Write the test first** — meaningful tests that verify behavior
-3. **Run tests — verify they FAIL** (if they pass, tests are wrong)
-4. **Write minimal implementation** to make tests pass
-5. **Run ALL tests — verify they PASS** (fix regressions)
-6. **Commit** with the task's commit message
-7. Update task status to `completed` and increment completed count
-8. Move to next pending task
-
-If a task is blocked, mark as `blocked` with reason and continue.
-
-### Step 7: Final Verification
-- Run the full test suite
-- Review all specification requirements are met
-- Check for security issues
-- Run the analyze step again on completed code
-- If issues found, create new tasks and implement them
-- Commit any final fixes
-
-### Step 8: Summary Report
-Output a final report:
+### Step 6: Commit and push
+- Commit all spec files created so far
+- Push to remote if configured
+- This preserves work before the long implementation phase
+
+### Step 7: Compact context
+- Run `/compact` to free up context window for implementation
+
+### Step 8: Implement with ralph-loop
+Invoke ralph-loop to implement all tasks automatically:
+
+```
+/ralph-loop:ralph-loop "Read specs/*/tasks.md and implement the next pending task using TDD (write test first, verify fail, implement, verify pass, commit). Update task status to completed after each task. Do not ask for any confirmation. Once all tasks are done, run analysis: read specs/*/tasks.md and verify all tasks are completed, run the full test suite, check all specification requirements are met, fix any issues found, and output <promise>FINISHED</promise> only after all tasks are completed and revalidated" --completion-promise "FINISHED" --max-iterations 10
+```
+
+### After ralph-loop completes
+
+Output a final summary:
 - What was built
 - Total tests passing
 - Files created/modified