feat(cortex): revamp cognitive phases to v2 architecture

Rearchitect cortex agent phases to version 2 with unified cognitive cycle and improved modularity. Removed old phases (categorize, gather, analyze, synthesize, understand, formulate, act) and replaced them with new v2 phases (1.categorize, 2.gather, 3.reason, 5.act) that always run all phases in sequence. Merged analyze and synthesize into a single reason phase.

Added new atomic capabilities for action deduplication, evidence and fact management, and memory with composite keys for namespace isolation. Updated cortex main.yaml to reflect new cognitive cycle with depth and iteration tracking, pipeline config overrides, and enhanced permissions and capabilities registry.

Added state-management modules for actions, evidence, facts, and memory with SQLite backend supporting deduplication, FTS search, conflict detection, and composite key memory.

Author: ibacalu

src/workflows_mcp/templates/agents/cortex/capabilities/atomic/action-complete.yaml

@@ -0,0 +1,52 @@
+name: cortex-action-complete
+description: |
+  Atomic capability: Mark action as completed with result.
+  Used by ACT phase after successful execution.
+
+tags: [cortex, capability, atomic, actions, deduplication]
+
+inputs:
+  state:
+    type: str
+    description: Path to SQLite state database.
+    required: true
+
+  action_id:
+    type: str
+    description: ID of the action to complete.
+    required: true
+
+  result_json:
+    type: str
+    description: Action result as JSON string.
+    required: false
+    default: "{}"
+
+blocks:
+  - id: complete
+    type: Workflow
+    inputs:
+      workflow: agent-state-management
+      inputs:
+        state: "{{ inputs.state }}"
+        op: actions
+        actions_op: complete
+        actions_action_id: "{{ inputs.action_id }}"
+        actions_result_json: "{{ inputs.result_json }}"
+        caller: "cortex-action-complete"
+
+outputs:
+  success:
+    value: "{{ blocks.complete.outputs.actions.success | default(false) }}"
+    type: bool
+    description: Whether operation succeeded.
+
+  action_id:
+    value: "{{ inputs.action_id }}"
+    type: str
+    description: ID of the completed action.
+
+  status:
+    value: "{{ blocks.complete.outputs.actions.status | default('') }}"
+    type: str
+    description: New status of the action (should be "completed").

src/workflows_mcp/templates/agents/cortex/capabilities/atomic/action-dispatch.yaml

@@ -0,0 +1,224 @@
+name: cortex-action-dispatch
+description: |
+  Atomic capability: Dispatch action with deduplication.
+
+  This is a wrapper that provides transparent action deduplication:
+  1. Logs the action (creates hash from capability + inputs)
+  2. If action already completed, returns cached result (skip execution)
+  3. If new or pending, executes the capability workflow
+  4. Marks action complete with result
+  5. Returns result (cached or fresh)
+
+  Use this wrapper instead of calling capability workflows directly
+  to prevent duplicate mutations (idempotency).
+
+  The wrapper handles these statuses:
+  - pending: Execute (new action logged)
+  - completed: Skip (return cached result)
+  - failed: Re-execute (retry failed action)
+  - rolled_back: Re-execute (action was undone)
+
+tags: [cortex, capability, atomic, actions, deduplication, wrapper]
+
+inputs:
+  state:
+    type: str
+    description: Path to SQLite state database.
+    required: true
+
+  cell_id:
+    type: str
+    description: Cell ID requesting the action.
+    required: true
+
+  capability:
+    type: str
+    description: Capability workflow name to execute.
+    required: true
+
+  inputs:
+    type: dict
+    description: Inputs to pass to the capability workflow.
+    required: true
+
+  reason:
+    type: str
+    description: Why this action is being executed (for audit).
+    required: false
+    default: ""
+
+  base_path:
+    type: str
+    description: Base path for file operations.
+    required: false
+    default: "."
+
+  permissions:
+    type: dict
+    description: Available permissions.
+    required: false
+    default: { "read": true, "write": false, "execute": false }
+
+blocks:
+  # ===========================================================================
+  # STEP 1: Log action for deduplication
+  # ===========================================================================
+  - id: log_action
+    type: Workflow
+    description: Log action and check for duplicates.
+    inputs:
+      workflow: cortex-action-log
+      inputs:
+        state: "{{ inputs.state }}"
+        cell_id: "{{ inputs.cell_id }}"
+        capability: "{{ inputs.capability }}"
+        inputs_json: "{{ inputs.inputs | tojson }}"
+
+  # ===========================================================================
+  # STEP 2: Check if should execute
+  # ===========================================================================
+  - id: check_execute
+    type: Shell
+    description: Determine if action should be executed or use cached result.
+    depends_on: [log_action]
+    inputs:
+      command: |
+        python3 << 'EOF'
+        import json
+        import os
+
+        already_exists = os.environ.get('ALREADY_EXISTS', 'false').lower() == 'true'
+        status = os.environ.get('STATUS', 'pending')
+        cached_result = os.environ.get('CACHED_RESULT', '')
+        action_id = os.environ.get('ACTION_ID', '')
+
+        should_execute = True
+        result = None
+        reason = 'new_action'
+
+        if already_exists:
+            if status == 'completed':
+                # Action already completed - use cached result
+                should_execute = False
+                reason = 'cached_completed'
+                try:
+                    result = json.loads(cached_result) if cached_result else {}
+                except json.JSONDecodeError:
+                    result = {'raw': cached_result}
+            elif status == 'failed':
+                # Previous attempt failed - retry
+                should_execute = True
+                reason = 'retry_failed'
+            elif status == 'pending':
+                # Action logged but not completed - execute
+                should_execute = True
+                reason = 'pending'
+            elif status == 'rolled_back':
+                # Action was rolled back - re-execute
+                should_execute = True
+                reason = 'retry_rolled_back'
+            else:
+                # Unknown status - execute to be safe
+                should_execute = True
+                reason = f'unknown_status:{status}'
+
+        print(json.dumps({
+            'should_execute': should_execute,
+            'reason': reason,
+            'action_id': action_id,
+            'cached_result': result,
+            'already_exists': already_exists,
+            'status': status
+        }))
+        EOF
+      env:
+        ALREADY_EXISTS: "{{ blocks.log_action.outputs.already_exists | default(false) }}"
+        STATUS: "{{ blocks.log_action.outputs.status | default('pending') }}"
+        CACHED_RESULT: "{{ blocks.log_action.outputs.cached_result | default({}) | tojson }}"
+        ACTION_ID: "{{ blocks.log_action.outputs.action_id | default('') }}"
+
+  # ===========================================================================
+  # STEP 3: Execute capability (if not cached)
+  # ===========================================================================
+  - id: execute_capability
+    type: Workflow
+    description: Execute the actual capability workflow.
+    depends_on: [check_execute]
+    condition: "{{ (blocks.check_execute.outputs.stdout | fromjson).should_execute }}"
+    continue_on_error: true
+    inputs:
+      workflow: "{{ inputs.capability }}"
+      inputs: "{{ {'state': inputs.state, 'base_path': inputs.base_path, 'permissions': inputs.permissions} | combine(inputs.inputs) }}"
+
+  # ===========================================================================
+  # STEP 4: Mark action complete (on success)
+  # ===========================================================================
+  - id: complete_action
+    type: Workflow
+    description: Mark action as completed with result.
+    depends_on: [execute_capability]
+    condition: "{{ blocks.execute_capability.succeeded }}"
+    inputs:
+      workflow: cortex-action-complete
+      inputs:
+        state: "{{ inputs.state }}"
+        action_id: "{{ (blocks.check_execute.outputs.stdout | fromjson).action_id }}"
+        result_json: "{{ blocks.execute_capability.outputs | tojson }}"
+
+  # ===========================================================================
+  # STEP 5: Mark action failed (on failure)
+  # ===========================================================================
+  - id: fail_action
+    type: Workflow
+    description: Mark action as failed.
+    depends_on: [execute_capability]
+    condition: "{{ blocks.execute_capability.failed and (blocks.check_execute.outputs.stdout | fromjson).should_execute }}"
+    inputs:
+      workflow: agent-state-management
+      inputs:
+        state: "{{ inputs.state }}"
+        op: actions
+        actions_op: fail
+        actions_action_id: "{{ (blocks.check_execute.outputs.stdout | fromjson).action_id }}"
+        actions_result_json: "{{ {'error': blocks.execute_capability.metadata.error_message | default('Unknown error')} | tojson }}"
+        caller: "cortex-action-dispatch"
+
+outputs:
+  success:
+    value: |
+      {{ blocks.execute_capability.succeeded
+         or (not (blocks.check_execute.outputs.stdout | fromjson).should_execute) }}
+    type: bool
+    description: Whether action succeeded (executed or cached).
+
+  executed:
+    value: "{{ (blocks.check_execute.outputs.stdout | fromjson).should_execute }}"
+    type: bool
+    description: Whether action was actually executed (vs cached).
+
+  cached:
+    value: "{{ not (blocks.check_execute.outputs.stdout | fromjson).should_execute }}"
+    type: bool
+    description: Whether cached result was used.
+
+  action_id:
+    value: "{{ (blocks.check_execute.outputs.stdout | fromjson).action_id }}"
+    type: str
+    description: ID of the logged action.
+
+  reason:
+    value: "{{ (blocks.check_execute.outputs.stdout | fromjson).reason }}"
+    type: str
+    description: Why execution decision was made.
+
+  result:
+    value: |
+      {{ blocks.execute_capability.outputs if blocks.execute_capability.succeeded
+         else ((blocks.check_execute.outputs.stdout | fromjson).cached_result | default({})) }}
+    type: dict
+    description: Action result (fresh or cached).
+
+  error:
+    value: "{{ blocks.execute_capability.metadata.error_message | default('') if blocks.execute_capability.failed else '' }}"
+    type: str
+    description: Error message if action failed.

src/workflows_mcp/templates/agents/cortex/capabilities/atomic/action-log.yaml

@@ -0,0 +1,76 @@
+name: cortex-action-log
+description: |
+  Atomic capability: Log action for deduplication before execution.
+  Used by ACT phase to prevent duplicate mutations.
+
+  Deduplication uses SHA256 hash of (capability, inputs) to identify
+  duplicate requests. If an identical action exists, returns cached result.
+
+  Action statuses:
+  - pending: Logged but not yet executed
+  - completed: Successfully executed
+  - failed: Execution failed
+  - rolled_back: Action was reversed
+
+tags: [cortex, capability, atomic, actions, deduplication]
+
+inputs:
+  state:
+    type: str
+    description: Path to SQLite state database.
+    required: true
+
+  cell_id:
+    type: str
+    description: Cell ID requesting the action.
+    required: true
+
+  capability:
+    type: str
+    description: Capability workflow name (e.g., "action-write", "action-edit").
+    required: true
+
+  inputs_json:
+    type: str
+    description: Action inputs as JSON string (used for deduplication hash).
+    required: true
+
+blocks:
+  - id: log
+    type: Workflow
+    inputs:
+      workflow: agent-state-management
+      inputs:
+        state: "{{ inputs.state }}"
+        op: actions
+        actions_op: log
+        actions_cell_id: "{{ inputs.cell_id }}"
+        actions_capability: "{{ inputs.capability }}"
+        actions_inputs_json: "{{ inputs.inputs_json }}"
+        caller: "cortex-action-log"
+
+outputs:
+  success:
+    value: "{{ blocks.log.outputs.actions.success | default(false) }}"
+    type: bool
+    description: Whether operation succeeded.
+
+  action_id:
+    value: "{{ blocks.log.outputs.actions.action_id | default('') }}"
+    type: str
+    description: ID of the logged action.
+
+  already_exists:
+    value: "{{ blocks.log.outputs.actions.already_exists | default(false) }}"
+    type: bool
+    description: Whether action already existed (duplicate detected).
+
+  status:
+    value: "{{ blocks.log.outputs.actions.status | default('') }}"
+    type: str
+    description: Current status of the action.
+
+  cached_result:
+    value: "{{ blocks.log.outputs.actions.result.result | default(none) }}"
+    type: dict
+    description: Cached result if action was already completed.