fix: power-steering incorrectly blocks Q&A/PM sessions with dev checks (#2914)

.claude/tools/amplihack/considerations.yaml

@@ -115,7 +115,7 @@
   severity: warning
   checker: _check_documentation_updates
   enabled: true
-  applicable_session_types: ["DEVELOPMENT", "MAINTENANCE", "INVESTIGATION"]
+  applicable_session_types: ["DEVELOPMENT", "MAINTENANCE"]
   guidance: |
     This check should only fire when PUBLIC-FACING code was changed.
 
@@ -170,7 +170,7 @@
   severity: blocker
   checker: _check_next_steps
   enabled: true
-  applicable_session_types: ["DEVELOPMENT", "INVESTIGATION"]
+  applicable_session_types: ["DEVELOPMENT"]
   guidance: |
     This check is about distinguishing COMPLETION SUMMARIES from REMAINING WORK.
 
@@ -222,7 +222,30 @@
   severity: blocker
   checker: _check_workflow_invocation
   enabled: true
-  applicable_session_types: ["DEVELOPMENT", "INVESTIGATION"]
+  applicable_session_types: ["DEVELOPMENT"]
+
+- id: skill_invocation
+  category: Workflow Process Adherence
+  question: If the session was started with a skill command (e.g. /pm-architect, /dev), did the agent actually invoke that skill using the Skill tool?
+  description: Detects when a user requests a skill via slash command but the agent bypasses it and responds directly without invoking the skill (Issue #2914)
+  severity: blocker
+  checker: _check_skill_invocation
+  enabled: true
+  applicable_session_types: ["DEVELOPMENT", "INVESTIGATION", "MAINTENANCE"]
+  guidance: |
+    Check if a <command-name> tag appears in the transcript (indicating the user
+    ran a slash command that maps to a skill). If so, check if the Skill tool
+    was called for that skill name.
+
+    SATISFIED when:
+    - No <command-name> tag in transcript (no skill was requested)
+    - A <command-name> tag exists AND the Skill tool was called for that skill
+    - The skill was loaded directly into the conversation (some skills inject
+      their content via the command system without needing a Skill tool call)
+
+    NOT SATISFIED when:
+    - A <command-name> tag exists but the agent never called the Skill tool
+      and instead responded directly, bypassing the skill entirely
 
 - id: no_direct_main_commit
   category: Workflow Process Adherence

.claude/tools/amplihack/hooks/power_steering_checker/checks_workflow.py

@@ -120,6 +120,54 @@ def _check_workflow_invocation(self, transcript: list[dict], session_id: str) ->
             self._log(f"Error in _check_workflow_invocation: {e}", "WARNING", exc_info=True)
             return True
 
+    def _check_skill_invocation(self, transcript: list[dict], session_id: str) -> bool:
+        """Check if a requested skill was actually invoked.
+
+        If the session was started with a slash command (indicated by a
+        <command-name> tag in the transcript), verify the Skill tool was
+        called for that skill. If no command-name tag exists, the check
+        is automatically satisfied (no skill was requested). (Issue #2914)
+
+        Args:
+            transcript: List of message dictionaries
+            session_id: Session identifier
+
+        Returns:
+            True if no skill was requested or if the requested skill was invoked
+        """
+        # Find <command-name> tag in user messages
+        requested_skill = None
+        for msg in transcript:
+            if msg.get("type") != "user":
+                continue
+            content_str = str(msg.get("message", {}).get("content", ""))
+            match = re.search(r"<command-name>/?([\w:.-]+)</command-name>", content_str)
+            if match:
+                requested_skill = match.group(1)
+                break
+
+        if not requested_skill:
+            return True  # No skill requested — check not applicable
+
+        # Check if the Skill tool was called for this skill
+        for msg in transcript:
+            if msg.get("type") != "assistant":
+                continue
+            content = msg.get("message", {}).get("content", [])
+            if not isinstance(content, list):
+                continue
+            for block in content:
+                if not isinstance(block, dict) or block.get("type") != "tool_use":
+                    continue
+                if block.get("name") == "Skill":
+                    invoked = block.get("input", {}).get("skill", "")
+                    if invoked == requested_skill:
+                        self._log(f"Skill '{requested_skill}' was invoked", "DEBUG")
+                        return True
+
+        self._log(f"Skill '{requested_skill}' was requested but not invoked", "WARNING")
+        return False
+
     def _check_no_direct_main_commit(self, transcript: list[dict], session_id: str) -> bool:
         """Check that the agent did not commit directly to main.
 

amplifier-bundle/tools/amplihack/considerations.yaml

@@ -115,7 +115,7 @@
   severity: warning
   checker: _check_documentation_updates
   enabled: true
-  applicable_session_types: ["DEVELOPMENT", "MAINTENANCE", "INVESTIGATION"]
+  applicable_session_types: ["DEVELOPMENT", "MAINTENANCE"]
   guidance: |
     This check should only fire when PUBLIC-FACING code was changed.
 
@@ -170,7 +170,7 @@
   severity: blocker
   checker: _check_next_steps
   enabled: true
-  applicable_session_types: ["DEVELOPMENT", "INVESTIGATION"]
+  applicable_session_types: ["DEVELOPMENT"]
   guidance: |
     This check is about distinguishing COMPLETION SUMMARIES from REMAINING WORK.
 
@@ -222,7 +222,30 @@
   severity: blocker
   checker: _check_workflow_invocation
   enabled: true
-  applicable_session_types: ["DEVELOPMENT", "INVESTIGATION"]
+  applicable_session_types: ["DEVELOPMENT"]
+
+- id: skill_invocation
+  category: Workflow Process Adherence
+  question: If the session was started with a skill command (e.g. /pm-architect, /dev), did the agent actually invoke that skill using the Skill tool?
+  description: Detects when a user requests a skill via slash command but the agent bypasses it and responds directly without invoking the skill (Issue #2914)
+  severity: blocker
+  checker: _check_skill_invocation
+  enabled: true
+  applicable_session_types: ["DEVELOPMENT", "INVESTIGATION", "MAINTENANCE"]
+  guidance: |
+    Check if a <command-name> tag appears in the transcript (indicating the user
+    ran a slash command that maps to a skill). If so, check if the Skill tool
+    was called for that skill name.
+
+    SATISFIED when:
+    - No <command-name> tag in transcript (no skill was requested)
+    - A <command-name> tag exists AND the Skill tool was called for that skill
+    - The skill was loaded directly into the conversation (some skills inject
+      their content via the command system without needing a Skill tool call)
+
+    NOT SATISFIED when:
+    - A <command-name> tag exists but the agent never called the Skill tool
+      and instead responded directly, bypassing the skill entirely
 
 - id: no_direct_main_commit
   category: Workflow Process Adherence

amplifier-bundle/tools/amplihack/hooks/power_steering_checker/checks_workflow.py

@@ -120,6 +120,54 @@ def _check_workflow_invocation(self, transcript: list[dict], session_id: str) ->
             self._log(f"Error in _check_workflow_invocation: {e}", "WARNING", exc_info=True)
             return True
 
+    def _check_skill_invocation(self, transcript: list[dict], session_id: str) -> bool:
+        """Check if a requested skill was actually invoked.
+
+        If the session was started with a slash command (indicated by a
+        <command-name> tag in the transcript), verify the Skill tool was
+        called for that skill. If no command-name tag exists, the check
+        is automatically satisfied (no skill was requested). (Issue #2914)
+
+        Args:
+            transcript: List of message dictionaries
+            session_id: Session identifier
+
+        Returns:
+            True if no skill was requested or if the requested skill was invoked
+        """
+        # Find <command-name> tag in user messages
+        requested_skill = None
+        for msg in transcript:
+            if msg.get("type") != "user":
+                continue
+            content_str = str(msg.get("message", {}).get("content", ""))
+            match = re.search(r"<command-name>/?([\w:.-]+)</command-name>", content_str)
+            if match:
+                requested_skill = match.group(1)
+                break
+
+        if not requested_skill:
+            return True  # No skill requested — check not applicable
+
+        # Check if the Skill tool was called for this skill
+        for msg in transcript:
+            if msg.get("type") != "assistant":
+                continue
+            content = msg.get("message", {}).get("content", [])
+            if not isinstance(content, list):
+                continue
+            for block in content:
+                if not isinstance(block, dict) or block.get("type") != "tool_use":
+                    continue
+                if block.get("name") == "Skill":
+                    invoked = block.get("input", {}).get("skill", "")
+                    if invoked == requested_skill:
+                        self._log(f"Skill '{requested_skill}' was invoked", "DEBUG")
+                        return True
+
+        self._log(f"Skill '{requested_skill}' was requested but not invoked", "WARNING")
+        return False
+
     def _check_no_direct_main_commit(self, transcript: list[dict], session_id: str) -> bool:
         """Check that the agent did not commit directly to main.
 

pyproject.toml

@@ -5,7 +5,7 @@ backend-path = ["."]
 
 [project]
 name = "amplihack"
-version = "0.5.113"
+version = "0.5.114"
 description = "Amplifier bundle for agentic coding with comprehensive skills, recipes, and workflows"
 requires-python = ">=3.11"
 dependencies = [