edsger 0.28.2 → 0.28.4

package/dist/phases/chat-processor/index.js

@@ -112,7 +112,8 @@ export async function processHumanMessages(messages, featureId, config, verbose)
             channelSessions.set(channelId, result.sessionId);
         }
         // If the agent didn't send a message via tools, send the response text
-        if (result.text && !result.text.includes('[tool_used:send_chat_message]')) {
+        const sentViaTool = result.toolsUsed.has('mcp__edsger-chat__send_chat_message');
+        if (result.text && !sentViaTool) {
             await sendAiMessage(channelId, result.text, {
                 in_response_to: messages[messages.length - 1].id,
             });
@@ -218,7 +219,9 @@ export async function processPhaseCompletion(featureId, phase, summary, phaseOut
             channelSessions.set(channelId, result.sessionId);
         }
         // If the agent didn't send a message via tools, send the raw response
-        if (result.text && !result.text.includes('[tool_used:send_chat_message]') && !result.text.includes('[tool_used:provide_options]')) {
+        const sentMessage = result.toolsUsed.has('mcp__edsger-chat__send_chat_message');
+        const sentOptions = result.toolsUsed.has('mcp__edsger-chat__provide_options');
+        if (result.text && !sentMessage && !sentOptions) {
             await sendAiMessage(channelId, result.text, {
                 phase_completion_suggestion: true,
                 phase,
@@ -297,13 +300,6 @@ async function runChatAgent(systemPrompt, userPrompt, config, resumeSessionId, v
             }
         }
     }
-    // Annotate response with tool usage for the caller
-    if (toolUsed.size > 0) {
-        const annotations = Array.from(toolUsed)
-            .map((t) => `[tool_used:${t}]`)
-            .join('');
-        lastAssistantResponse += annotations;
-    }
     return {
         text: lastAssistantResponse.trim(),
         sessionId,

package/dist/phases/chat-processor/prompts.d.ts

@@ -5,8 +5,13 @@
  * 1. CHAT_RESPONSE_PROMPT — responding to human messages
  * 2. NEXT_STEP_ADVISOR_PROMPT — proactive suggestions after phase completion
  */
-export declare const CHAT_RESPONSE_PROMPT = "You are an AI assistant embedded in Edsger, a software development platform. You are helping a team develop a feature.\n\n## Your Capabilities\nYou can see the feature's current state, user stories, test cases, workflow phases, and code. You have tools to:\n- Modify feature status, execution mode, and workflow phases\n- Create/update user stories and test cases\n- Read and search source code files\n- Send follow-up messages and present options to the user\n- Trigger phase reruns\n\n## How to Respond\n1. **Understand the intent** \u2014 Is this feedback, a question, a request to change something, or just a comment?\n2. **Take action if needed** \u2014 Use the appropriate tools to make changes\n3. **Respond concisely** \u2014 Summarize what you understood and what you did (or why you didn't do anything)\n4. **Ask for clarification** \u2014 If the message is ambiguous, use provide_options to present choices\n\n## Communication Style\n- Respond in the same language the user writes in\n- Be concise but thorough \u2014 no filler text\n- Reference specific items by name (e.g., \"User Story #3: Login flow\")\n- When making changes, always explain what you changed and why\n\n## Important Rules\n- Never make destructive changes without confirmation (deleting stories, resetting phases)\n- For ambiguous feedback, present options rather than guessing\n- If you can't do something, explain why clearly\n";
-export declare const NEXT_STEP_ADVISOR_PROMPT = "You are an AI advisor in Edsger, a software development platform. A workflow phase just completed for a feature you're helping develop.\n\n## Your Job\nAnalyze the completed phase output and the feature's current state, then give a concrete, data-backed suggestion for what to do next.\n\n## Critical Rules\n1. **Reference specific data** \u2014 \"8 user stories generated, 3 involve complex auth logic\" not \"several stories were created\"\n2. **Explain your reasoning** \u2014 \"Because Story #3 involves concurrent editing, I suggest writing test cases first to define edge cases before implementation\"\n3. **Present actionable options** \u2014 Always use the provide_options tool to give 2-4 choices the user can click\n4. **Do NOT follow a fixed phase order** \u2014 Adapt based on:\n   - Feature size and complexity\n   - What was just produced (quality, coverage, gaps)\n   - Previous human feedback in the chat\n   - Whether certain phases can be skipped for simple features\n5. **Flag issues proactively** \u2014 If the phase output has gaps, incomplete coverage, or potential problems, call them out\n\n## Context You Receive\n- Feature description, size, and current state\n- The completed phase name and its full output\n- Remaining workflow phases\n- User story and test case counts\n- Code change scope (if applicable)\n- Recent chat history (human feedback)\n\n## Communication Style\n- Respond in the same language as recent chat messages (default to the feature's language context)\n- Be specific and data-driven\n- Structure: brief summary \u2192 reasoning \u2192 options\n";
+export declare const CHAT_RESPONSE_PROMPT = "You are an AI assistant embedded in Edsger, a software development platform. You are helping a team develop a feature.\n\n## Your Capabilities\nYou can see the feature's current state, user stories, test cases, workflow phases, and code. You have tools to:\n- Modify feature status, execution mode, and workflow phases\n- Create/update user stories and test cases\n- Read and search source code files\n- Send follow-up messages and present options to the user\n- Trigger phase reruns\n\n## How to Respond\n1. **Understand the intent** \u2014 Is this feedback, a question, a request to change something, or just a comment?\n2. **Take action if needed** \u2014 Use the appropriate tools to make changes\n3. **Respond concisely** \u2014 Summarize what you understood and what you did (or why you didn't do anything)\n4. **Ask for clarification** \u2014 If the message is ambiguous, use provide_options to present choices\n\n## Communication Style\n- Respond in the same language the user writes in\n- Be concise but thorough \u2014 no filler text\n- Reference specific items by name (e.g., \"User Story #3: Login flow\")\n- When making changes, always explain what you changed and why\n\n## Phase Reference (know what each phase does before suggesting it)\n- **code-implementation**: Writes/updates production code (creates or modifies code)\n- **pr-execution**: Syncs already-written code from dev branch to split PR branches (does NOT write new code)\n- **bug-fixing**: Fixes code bugs and test failures\n- To fix bugs or update code \u2192 suggest **code-implementation** or **bug-fixing**, NOT pr-execution\n\n## Important Rules\n- Never make destructive changes without confirmation (deleting stories, resetting phases)\n- For ambiguous feedback, present options rather than guessing\n- If you can't do something, explain why clearly\n";
+export declare const NEXT_STEP_ADVISOR_PROMPT = "You are an AI advisor in Edsger, a software development platform. A workflow phase just completed for a feature you're helping develop.\n\n## Your Job\nAnalyze the completed phase output and the feature's current state, then give a concrete, data-backed suggestion for what to do next.\n\n## Critical Rules\n1. **Reference specific data** \u2014 \"8 user stories generated, 3 involve complex auth logic\" not \"several stories were created\"\n2. **Explain your reasoning** \u2014 \"Because Story #3 involves concurrent editing, I suggest writing test cases first to define edge cases before implementation\"\n3. **Present actionable options** \u2014 Always use the provide_options tool to give 2-4 choices the user can click\n4. **Do NOT follow a fixed phase order** \u2014 Adapt based on:\n   - Feature size and complexity\n   - What was just produced (quality, coverage, gaps)\n   - Previous human feedback in the chat\n   - Whether certain phases can be skipped for simple features\n5. **Flag issues proactively** \u2014 If the phase output has gaps, incomplete coverage, or potential problems, call them out\n\n## Phase Reference (know what each phase does before suggesting it)\n- **code-implementation**: Writes/updates production code (the phase that creates or modifies code)\n- **pr-splitting**: Plans how to split code changes into reviewable PRs\n- **pr-execution**: Syncs already-written code from the dev branch to split PR branches (does NOT write new code)\n- **code-testing**: Writes automated tests for implemented code\n- **functional-testing**: Runs end-to-end tests with Playwright\n- **bug-fixing**: Fixes code bugs and test failures\n- **code-review**: Reviews PR code for issues\n- **code-refine**: Updates code based on PR review feedback\n\n**Important distinctions:**\n- To fix bugs or update code \u2192 use **code-implementation** or **bug-fixing**, NOT pr-execution\n- pr-execution only moves existing code to PR branches \u2014 it never creates or modifies implementation code\n\n## Context You Receive\n- Feature description, size, and current state\n- The completed phase name and its full output\n- Remaining workflow phases (with descriptions)\n- User story and test case counts\n- Code change scope (if applicable)\n- Recent chat history (human feedback)\n\n## Communication Style\n- Respond in the same language as recent chat messages (default to the feature's language context)\n- Be specific and data-driven\n- Structure: brief summary \u2192 reasoning \u2192 options\n";
+/**
+ * Phase descriptions so the AI advisor understands what each phase does.
+ * This prevents misinterpretation (e.g., thinking pr-execution writes code).
+ */
+export declare const PHASE_DESCRIPTIONS: Record<string, string>;
 /**
  * Build the user message for the next-step advisor based on phase completion context.
  */

package/dist/phases/chat-processor/prompts.js

@@ -27,6 +27,12 @@ You can see the feature's current state, user stories, test cases, workflow phas
 - Reference specific items by name (e.g., "User Story #3: Login flow")
 - When making changes, always explain what you changed and why
 
+## Phase Reference (know what each phase does before suggesting it)
+- **code-implementation**: Writes/updates production code (creates or modifies code)
+- **pr-execution**: Syncs already-written code from dev branch to split PR branches (does NOT write new code)
+- **bug-fixing**: Fixes code bugs and test failures
+- To fix bugs or update code → suggest **code-implementation** or **bug-fixing**, NOT pr-execution
+
 ## Important Rules
 - Never make destructive changes without confirmation (deleting stories, resetting phases)
 - For ambiguous feedback, present options rather than guessing
@@ -48,10 +54,24 @@ Analyze the completed phase output and the feature's current state, then give a
    - Whether certain phases can be skipped for simple features
 5. **Flag issues proactively** — If the phase output has gaps, incomplete coverage, or potential problems, call them out
 
+## Phase Reference (know what each phase does before suggesting it)
+- **code-implementation**: Writes/updates production code (the phase that creates or modifies code)
+- **pr-splitting**: Plans how to split code changes into reviewable PRs
+- **pr-execution**: Syncs already-written code from the dev branch to split PR branches (does NOT write new code)
+- **code-testing**: Writes automated tests for implemented code
+- **functional-testing**: Runs end-to-end tests with Playwright
+- **bug-fixing**: Fixes code bugs and test failures
+- **code-review**: Reviews PR code for issues
+- **code-refine**: Updates code based on PR review feedback
+
+**Important distinctions:**
+- To fix bugs or update code → use **code-implementation** or **bug-fixing**, NOT pr-execution
+- pr-execution only moves existing code to PR branches — it never creates or modifies implementation code
+
 ## Context You Receive
 - Feature description, size, and current state
 - The completed phase name and its full output
-- Remaining workflow phases
+- Remaining workflow phases (with descriptions)
 - User story and test case counts
 - Code change scope (if applicable)
 - Recent chat history (human feedback)
@@ -61,6 +81,30 @@ Analyze the completed phase output and the feature's current state, then give a
 - Be specific and data-driven
 - Structure: brief summary → reasoning → options
 `;
+/**
+ * Phase descriptions so the AI advisor understands what each phase does.
+ * This prevents misinterpretation (e.g., thinking pr-execution writes code).
+ */
+export const PHASE_DESCRIPTIONS = {
+    'feature-analysis': 'Analyze a feature and generate user stories and test cases',
+    'user-stories-analysis': 'Write detailed, professional user stories for the feature',
+    'test-cases-analysis': 'Generate comprehensive test cases from user stories',
+    'feature-analysis-verification': 'Verify generated user stories and test cases against checklist requirements',
+    'technical-design': 'Create detailed technical design and architecture documentation',
+    'technical-design-verification': 'Verify technical design satisfies checklist requirements',
+    'branch-planning': 'Split implementation into multiple logical branches for incremental delivery',
+    'code-implementation': 'Write production-ready code to implement the feature (the phase that actually writes/updates code)',
+    'code-implementation-verification': 'Review code implementation against checklist requirements',
+    'pr-splitting': 'Analyze code changes and plan how to split them into reviewable PRs',
+    'pr-execution': 'Sync code from the dev branch to split PR branches (does NOT write new code — only moves existing changes to PR branches)',
+    'code-testing': 'Write automated tests (unit, integration) for the implemented code',
+    'functional-testing': 'Run end-to-end functional tests with headless Playwright',
+    'code-review': 'Review pull request code for issues, bugs, and quality concerns',
+    'code-refine': 'Refine code based on PR review feedback',
+    'code-refine-verification': 'Verify refined code addresses all review comments',
+    'bug-fixing': 'Analyze test failures and fix underlying code issues',
+    autonomous: 'Discover and implement improvements in a time-limited loop',
+};
 /**
  * Build the user message for the next-step advisor based on phase completion context.
  */
@@ -75,7 +119,10 @@ export function buildNextStepAdvisorMessage(context) {
         `## Current State`,
         `- User stories: ${context.userStoriesCount}`,
         `- Test cases: ${context.testCasesCount}`,
-        `- Remaining phases: ${context.remainingPhases.join(', ') || 'none'}`,
+        `- Remaining phases:`,
+        ...(context.remainingPhases.length > 0
+            ? context.remainingPhases.map((p) => `  - ${p}: ${PHASE_DESCRIPTIONS[p] || 'No description available'}`)
+            : ['  none']),
     ];
     if (context.codeChangesScope) {
         parts.push(`- Code changes: ${context.codeChangesScope.files} files, +${context.codeChangesScope.additions}/-${context.codeChangesScope.deletions} lines`);

package/dist/phases/chat-processor/tools.js

@@ -59,8 +59,11 @@ export function createChatMcpServer() {
                     content: [{ type: 'text', text: JSON.stringify(result, null, 2) }],
                 };
             }),
-            tool('update_user_stories', 'Create, update, or delete user stories based on feedback.', {
+            tool('update_user_stories', 'Create, update, or delete user stories based on feedback. Always provide a reason explaining why the change is being made.', {
                 feature_id: z.string().describe('Feature ID'),
+                reason: z
+                    .string()
+                    .describe('Why this change is being made — e.g. user feedback, bug found, requirement changed. Required for audit trail.'),
                 actions: z.array(z.object({
                     action: z.enum(['create', 'update', 'delete']),
                     user_story_id: z
@@ -98,6 +101,7 @@ export function createChatMcpServer() {
                             await callMcpEndpoint('user_stories/update_status', {
                                 user_story_id: action.user_story_id,
                                 status: action.status || 'draft',
+                                reason: args.reason,
                             });
                             results.push({ action: 'update', success: true });
                         }
@@ -117,8 +121,11 @@ export function createChatMcpServer() {
                     ],
                 };
             }),
-            tool('update_test_cases', 'Create, update, or delete test cases based on feedback.', {
+            tool('update_test_cases', 'Create, update, or delete test cases based on feedback. Always provide a reason explaining why the change is being made.', {
                 feature_id: z.string().describe('Feature ID'),
+                reason: z
+                    .string()
+                    .describe('Why this change is being made — e.g. user feedback, bug found, requirement changed. Required for audit trail.'),
                 actions: z.array(z.object({
                     action: z.enum(['create', 'update', 'delete']),
                     test_case_id: z
@@ -156,6 +163,7 @@ export function createChatMcpServer() {
                             await callMcpEndpoint('test_cases/update_status', {
                                 test_case_id: action.test_case_id,
                                 status: 'draft',
+                                reason: args.reason,
                             });
                             results.push({ action: 'update', success: true });
                         }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "edsger",
-  "version": "0.28.2",
+  "version": "0.28.4",
   "type": "module",
   "bin": {
     "edsger": "dist/index.js"