refactor: replace agent identity directives with context directives

inc/Api/Chat/Chat.php

@@ -22,7 +22,7 @@
 use WP_Error;
 
 require_once __DIR__ . '/ChatPipelinesDirective.php';
-require_once __DIR__ . '/ChatAgentDirective.php';
+require_once __DIR__ . '/ChatContextDirective.php';
 
 if ( ! defined( 'ABSPATH' ) ) {
 	exit;

inc/Api/Chat/ChatContextDirective.php

@@ -0,0 +1,80 @@
+<?php
+/**
+ * Chat Context Directive
+ *
+ * Operational context for chat sessions. Identity comes from memory files (SOUL.md, etc.).
+ *
+ * @package DataMachine\Api\Chat
+ * @since 0.2.0
+ */
+
+namespace DataMachine\Api\Chat;
+
+if ( ! defined( 'ABSPATH' ) ) {
+	exit;
+}
+
+/**
+ * Chat Context Directive
+ *
+ * Provides Data Machine chat-specific operational context — architecture,
+ * tools, and behavioral rules. Agent identity comes from memory files.
+ */
+class ChatContextDirective implements \DataMachine\Engine\AI\Directives\DirectiveInterface {
+
+	public static function get_outputs( string $provider_name, array $tools, ?string $step_id = null, array $payload = array() ): array {
+		$directive = self::get_directive( $tools );
+
+		return array(
+			array(
+				'type'    => 'system_text',
+				'content' => $directive,
+			),
+		);
+	}
+
+	/**
+	 * Generate chat context directive.
+	 *
+	 * @param array $tools Available tools.
+	 * @return string Directive text.
+	 */
+	private static function get_directive( $tools ): string {
+		return '# Chat Session Context' . "\n\n"
+			. 'This is a live chat session with a user in the Data Machine admin UI. You have tools to configure and manage workflows. Your identity, voice, and knowledge come from your memory files above.' . "\n\n"
+			. '## Data Machine Architecture' . "\n\n"
+			. 'HANDLERS are the core intelligence. Fetch handlers extract and structure source data. Update/publish handlers apply changes with schema defaults for unconfigured fields. Each handler has a settings schema — only use documented fields.' . "\n\n"
+			. 'PIPELINES define workflow structure: step types in sequence (e.g., event_import → ai → upsert). The pipeline system_prompt defines AI behavior shared by all flows.' . "\n\n"
+			. 'FLOWS are configured pipeline instances. Each step needs a handler_slug and handler_config. When creating flows, match handler configurations from existing flows on the same pipeline.' . "\n\n"
+			. 'AI STEPS process data that handlers cannot automatically handle. Flow user_message is rarely needed; only for minimal source-specific overrides.' . "\n\n"
+			. '## Discovery' . "\n\n"
+			. 'You receive a pipeline inventory with existing flows and their handlers. Use `api_query` for detailed configuration. Query existing flows before creating new ones to learn established patterns.' . "\n\n"
+			. '## Configuration Rules' . "\n\n"
+			. '- Only use documented handler_config fields — unknown fields are rejected.' . "\n"
+			. '- Use pipeline_step_id from the inventory to target steps.' . "\n"
+			. '- Unconfigured handler fields use schema defaults automatically.' . "\n"
+			. '- Act first — if the user gives executable instructions, execute them.' . "\n\n"
+			. '## Scheduling' . "\n\n"
+			. '- Scheduling uses intervals only (daily, hourly, etc.), not specific times of day.' . "\n"
+			. '- Valid intervals are provided in the tool definitions. Use update_flow to change schedules.' . "\n\n"
+			. '## Execution Protocol' . "\n\n"
+			. '- Only confirm task completion after a successful tool result. Never claim success on error.' . "\n"
+			. '- Check error_type on failure: not_found/permission → report, validation → fix and retry, system → retry once.' . "\n"
+			. '- If a tool rejects unknown fields, retry with only the valid fields listed in the error.' . "\n"
+			. '- Act decisively — execute tools directly for routine configuration.' . "\n"
+			. '- If uncertain about a value, use sensible defaults and note the assumption.';
+	}
+}
+
+// Register with directive system (Priority 25, after memory files at 20)
+add_filter(
+	'datamachine_directives',
+	function ( $directives ) {
+		$directives[] = array(
+			'class'    => ChatContextDirective::class,
+			'priority' => 25,
+			'contexts' => array( 'chat' ),
+		);
+		return $directives;
+	}
+);

inc/Api/System/SystemContextDirective.php

@@ -0,0 +1,80 @@
+<?php
+/**
+ * System Context Directive
+ *
+ * Operational context for system tasks. Identity comes from memory files (SOUL.md, etc.).
+ *
+ * @package DataMachine\Api\System
+ * @since   0.13.7
+ */
+
+namespace DataMachine\Api\System;
+
+if ( ! defined( 'ABSPATH' ) ) {
+	exit;
+}
+
+/**
+ * System Context Directive
+ *
+ * Provides system task operational context — task behavior rules and
+ * capability hints. Agent identity comes from memory files.
+ */
+class SystemContextDirective implements \DataMachine\Engine\AI\Directives\DirectiveInterface {
+
+	public static function get_outputs( string $provider_name, array $tools, ?string $step_id = null, array $payload = array() ): array {
+		$directive = self::get_directive( $tools );
+
+		return array(
+			array(
+				'type'    => 'system_text',
+				'content' => $directive,
+			),
+		);
+	}
+
+	/**
+	 * Generate system context directive.
+	 *
+	 * @param  array $tools Available tools.
+	 * @return string Directive text.
+	 */
+	private static function get_directive( $tools ): string {
+		$directive = '# System Task Context' . "\n\n"
+		. 'This is a background system task — not a chat session. Your identity and knowledge are already loaded from your memory files above. Use that context.' . "\n\n"
+		. '## Task Behavior' . "\n\n"
+		. '- Execute the task described in the user message below.' . "\n"
+		. '- Return exactly what the task asks for — no extra commentary, no meta-discussion.' . "\n"
+		. '- Apply your knowledge of this site, its voice, and its conventions from your memory files.' . "\n\n"
+		. '## Session Title Generation' . "\n\n"
+		. 'When asked to generate a chat session title: create a concise, descriptive title (3-6 words) capturing the discussion essence. Return ONLY the title text, under 100 characters.' . "\n\n"
+		. '## GitHub Issue Creation' . "\n\n"
+		. 'When using create_github_issue: include a clear title and detailed body with context, reproduction steps, and relevant log snippets. Use labels to categorize. Route to the most appropriate repo. Never create duplicates.';
+
+		// List available repos dynamically.
+		if ( class_exists( '\DataMachine\Abilities\Fetch\GitHubAbilities' ) ) {
+			$repos = \DataMachine\Abilities\Fetch\GitHubAbilities::getRegisteredRepos();
+			if ( ! empty( $repos ) ) {
+				$directive .= "\n\n" . 'Available repositories for issue creation:' . "\n";
+				foreach ( $repos as $entry ) {
+					$directive .= '- ' . $entry['owner'] . '/' . $entry['repo'] . ' — ' . $entry['label'] . "\n";
+				}
+			}
+		}
+
+		return $directive;
+	}
+}
+
+// Register with directive system (Priority 25, after memory files at 20)
+add_filter(
+	'datamachine_directives',
+	function ( $directives ) {
+		$directives[] = array(
+			'class'    => SystemContextDirective::class,
+			'priority' => 25,
+			'contexts' => array( 'system' ),
+		);
+		return $directives;
+	}
+);