@trac3er/oh-my-god 1.0.2

package/hooks/_agent_registry.py

@@ -0,0 +1,421 @@
+#!/usr/bin/env python3
+"""Agent Registry — Central dispatch table for OMG domain agents.
+
+Maps domain keywords to agents with model preferences, skills, and MCP tools.
+"""
+import json
+import os
+import shutil
+
+# Agent registry: domain → agent config
+AGENT_REGISTRY = {
+    'frontend-designer': {
+        'preferred_model': 'gemini-cli',
+        'task_category': 'visual-engineering',
+        'skills': ['frontend-design', 'frontend-ui-ux'],
+        'trigger_keywords': {'ui', 'ux', 'css', 'layout', 'responsive', 'visual', 'frontend', 'component', 'style', 'design', 'animation', 'color', 'theme'},
+        'mcp_tools': ['puppeteer_screenshot', 'puppeteer_navigate'],
+        'description': 'Frontend/UI specialist. Uses Gemini for visual tasks.',
+        'agent_file': 'agents/omg-frontend-designer.md',
+        'model_version': 'gemini-3.1-pro-preview',
+    },
+    'backend-engineer': {
+        'preferred_model': 'codex-cli',
+        'task_category': 'deep',
+        'skills': ['backend-patterns', 'api-design'],
+        'trigger_keywords': {'api', 'server', 'database', 'logic', 'algorithm', 'backend', 'endpoint', 'route', 'middleware', 'service'},
+        'mcp_tools': [],
+        'description': 'Backend/logic specialist. Uses Codex for deep reasoning.',
+        'agent_file': 'agents/omg-backend-engineer.md',
+        'model_version': 'gpt-5.3',
+    },
+    'api-builder': {
+        'preferred_model': 'codex-cli',
+        'task_category': 'deep',
+        'skills': ['api-design', 'backend-patterns'],
+        'trigger_keywords': {'openapi', 'swagger', 'rest', 'graphql', 'api-spec', 'schema', 'contract', 'endpoint-design'},
+        'mcp_tools': ['context7_query-docs'],
+        'description': 'API design/build specialist. Contracts, endpoint shape, and versioning.',
+        'agent_file': 'agents/omg-api-builder.md',
+        'model_version': 'gpt-5.3',
+    },
+    'security-auditor': {
+        'preferred_model': 'codex-cli',
+        'task_category': 'deep',
+        'skills': ['security-review'],
+        'trigger_keywords': {'auth', 'encrypt', 'cors', 'jwt', 'vulnerability', 'security', 'xss', 'csrf', 'injection', 'secret', 'password', 'token'},
+        'mcp_tools': ['sentry_search_issues', 'sentry_get_issue_details'],
+        'description': 'Security specialist. Uses Codex for deep security analysis.',
+        'agent_file': 'agents/omg-security-auditor.md',
+        'model_version': 'gpt-5.3',
+    },
+    'database-engineer': {
+        'preferred_model': 'codex-cli',
+        'task_category': 'unspecified-high',
+        'skills': [],
+        'trigger_keywords': {'sql', 'migration', 'schema', 'query', 'index', 'database', 'postgres', 'mongo', 'redis', 'orm'},
+        'mcp_tools': [],
+        'description': 'Database specialist. Schema design, query optimization, migrations.',
+        'agent_file': 'agents/omg-database-engineer.md',
+        'model_version': 'gpt-5.3',
+    },
+    'testing-engineer': {
+        'preferred_model': 'claude',
+        'task_category': 'unspecified-high',
+        'skills': ['python-testing', 'e2e-testing'],
+        'trigger_keywords': {'test', 'spec', 'coverage', 'fixture', 'mock', 'playwright', 'e2e', 'unit', 'integration', 'pytest', 'jest'},
+        'mcp_tools': ['puppeteer_navigate', 'puppeteer_screenshot'],
+        'description': 'Testing specialist. Unit tests, integration tests, E2E with Playwright.',
+        'agent_file': 'agents/omg-testing-engineer.md',
+        'model_version': 'claude-sonnet-4-5',
+    },
+    'infra-engineer': {
+        'preferred_model': 'codex-cli',
+        'task_category': 'unspecified-high',
+        'skills': ['docker-patterns'],
+        'trigger_keywords': {'docker', 'ci', 'cd', 'deploy', 'terraform', 'k8s', 'kubernetes', 'nginx', 'pipeline', 'container', 'cloud'},
+        'mcp_tools': [],
+        'description': 'Infrastructure specialist. Docker, CI/CD, deployment, cloud.',
+        'agent_file': 'agents/omg-infra-engineer.md',
+        'model_version': 'gpt-5.3',
+    },
+    # Cognitive modes
+    'research-mode': {
+        'preferred_model': 'claude',
+        'task_category': None,
+        'subagent_type': 'librarian',
+        'skills': [],
+        'trigger_keywords': {'research', 'find', 'how to', 'explain', 'documentation', 'docs', 'lookup'},
+        'mcp_tools': ['web_search_exa', 'google_search', 'context7_query-docs'],
+        'description': 'Research mode. Web search, docs lookup, library exploration.',
+        'agent_file': 'agents/omg-research-mode.md',
+        'model_version': 'claude-haiku-3-5',
+    },
+    'architect-mode': {
+        'preferred_model': 'claude',
+        'task_category': None,
+        'subagent_type': 'oracle',
+        'skills': [],
+        'trigger_keywords': {'architect', 'design', 'plan', 'structure', 'system', 'architecture', 'tradeoff'},
+        'mcp_tools': [],
+        'description': 'Architecture mode. System design, trade-off analysis.',
+        'agent_file': 'agents/omg-architect-mode.md',
+        'model_version': 'claude-sonnet-4-5',
+    },
+    'implement-mode': {
+        'preferred_model': 'domain-dependent',
+        'task_category': 'deep',
+        'skills': [],
+        'trigger_keywords': {'implement', 'build', 'create', 'add', 'develop', 'write', 'code'},
+        'mcp_tools': [],
+        'description': 'Implementation mode. Model chosen based on domain of task.',
+        'agent_file': 'agents/omg-implement-mode.md',
+        'model_version': 'claude-sonnet-4-5',
+    },
+    'implement-mode': {
+        'preferred_model': 'domain-dependent',
+        'task_category': 'deep',
+        'skills': [],
+        'trigger_keywords': {'implement', 'build', 'create', 'add', 'develop', 'write', 'code'},
+        'mcp_tools': [],
+        'description': 'Implementation mode. Model chosen based on domain of task.',
+        'agent_file': 'agents/omg-implement-mode.md',
+        'model_version': 'claude-sonnet-4-5',
+    },
+    # Bundled agents (Task 2.3)
+    'explore': {
+        'preferred_model': 'claude',
+        'task_category': 'quick',
+        'skills': [],
+        'trigger_keywords': {'find', 'search', 'grep', 'locate', 'where', 'which', 'lookup', 'explore', 'discover'},
+        'mcp_tools': [],
+        'description': 'Fast codebase search agent. Read-only: grep, glob, file reading, pattern matching.',
+        'agent_file': 'agents/explore.md',
+        'model_version': 'claude-haiku-4-5',
+        'model_role': 'smol',
+        'bundled': True,
+    },
+    'plan': {
+        'preferred_model': 'claude',
+        'task_category': 'unspecified-high',
+        'skills': [],
+        'trigger_keywords': {'plan', 'architect', 'design', 'decompose', 'strategy', 'roadmap', 'breakdown', 'structure'},
+        'mcp_tools': [],
+        'description': 'Strategic planning agent. Architecture design, task decomposition, risk analysis.',
+        'agent_file': 'agents/plan.md',
+        'model_version': 'claude-opus-4-5',
+        'model_role': 'slow',
+        'bundled': True,
+    },
+    'designer': {
+        'preferred_model': 'gemini-cli',
+        'task_category': 'visual-engineering',
+        'skills': ['frontend-design', 'frontend-ui-ux'],
+        'trigger_keywords': {'component', 'layout', 'accessibility', 'responsive', 'tailwind', 'css', 'aria', 'wcag', 'breakpoint'},
+        'mcp_tools': ['puppeteer_screenshot', 'puppeteer_navigate'],
+        'description': 'UI/UX design agent. Component design, layout, accessibility, responsive design.',
+        'agent_file': 'agents/designer.md',
+        'model_version': 'claude-opus-4-5',
+        'model_role': 'default',
+        'bundled': True,
+    },
+    'reviewer': {
+        'preferred_model': 'codex-cli',
+        'task_category': 'deep',
+        'skills': ['security-review'],
+        'trigger_keywords': {'review', 'audit', 'check', 'inspect', 'critique', 'feedback', 'pr', 'pull-request', 'quality'},
+        'mcp_tools': ['sentry_search_issues', 'sentry_get_issue_details'],
+        'description': 'Code review agent. Security, performance, quality, best practices, test coverage.',
+        'agent_file': 'agents/reviewer.md',
+        'model_version': 'claude-opus-4-5',
+        'model_role': 'slow',
+        'bundled': True,
+    },
+    'task': {
+        'preferred_model': 'claude',
+        'task_category': 'unspecified-high',
+        'skills': [],
+        'trigger_keywords': {'fix', 'implement', 'feature', 'bug', 'patch', 'update', 'change', 'modify', 'refactor'},
+        'mcp_tools': [],
+        'description': 'General task execution agent. Implement features, fix bugs, write tests.',
+        'agent_file': 'agents/task.md',
+        'model_version': 'claude-opus-4-5',
+        'model_role': 'default',
+        'bundled': True,
+    },
+    'quick_task': {
+        'preferred_model': 'claude',
+        'task_category': 'quick',
+        'skills': [],
+        'trigger_keywords': {'typo', 'rename', 'label', 'caption', 'spelling', 'minor', 'small', 'quick', 'simple'},
+        'mcp_tools': [],
+        'description': 'Fast task execution agent. Simple fixes, typo corrections, single-file changes.',
+        'agent_file': 'agents/quick_task.md',
+        'model_version': 'claude-haiku-4-5',
+        'model_role': 'smol',
+        'bundled': True,
+    },
+}
+
+# ═══════════════════════════════════════════════════════════
+# Intent-to-Agent Routing Table (Magic Keyword Router)
+# Maps LEADER_HINT intents from intentgate-keyword-detector
+# to target agent names. None = halt (no agent dispatch).
+# ═══════════════════════════════════════════════════════════
+INTENT_ROUTING = {
+    "INTENT_MAX_EFFORT":  "sisyphus",     # ultrawork → full-effort agent
+    "INTENT_AUTONOMOUS":  "sisyphus",     # autopilot → autonomous agent
+    "INTENT_LOOP":        "sisyphus",     # ralph → loop agent
+    "INTENT_PLAN":        "prometheus",   # plan this → planning agent
+    "INTENT_TEST_DRIVEN": "sisyphus",     # tdd → TDD agent
+    "INTENT_SEARCH":      "librarian",    # search → search agent
+    "INTENT_STOP":        None,           # stop → halt (no agent)
+    "INTENT_CRAZY":       "sisyphus",     # crazy → aggressive agent
+    # Bundled agent intents (Task 2.3)
+    "INTENT_EXPLORE":     "explore",      # explore/find → fast search agent
+    "INTENT_REVIEW":      "reviewer",     # review/audit → code review agent
+    "INTENT_QUICK":       "quick_task",   # quick/simple → fast task agent
+}
+
+# Core agent model preferences. NOT keyword-matched — used by orchestration pipeline only. model_version is informational (not passed to CLI).
+CORE_AGENT_MODELS = {
+    'architect': {
+        'preferred_model': 'codex-cli',
+        'model_version': 'gpt-5.2',
+        'task_category': None,
+        'description': 'System design + planning + delegation routing.',
+        'agent_file': 'agents/omg-architect.md',
+    },
+    'critic': {
+        'preferred_model': 'codex-cli',
+        'model_version': 'gpt-5.3',
+        'task_category': None,
+        'description': 'Code review — 3 perspectives, no LGTM allowed.',
+        'agent_file': 'agents/omg-critic.md',
+    },
+    'executor': {
+        'preferred_model': 'claude',
+        'model_version': 'claude-sonnet-4-5',
+        'task_category': 'deep',
+        'description': 'Implements code with evidence, auto-escalates when stuck.',
+        'agent_file': 'agents/omg-executor.md',
+    },
+    'qa-tester': {
+        'preferred_model': 'claude',
+        'model_version': 'claude-sonnet-4-5',
+        'task_category': 'unspecified-high',
+        'description': 'User-journey test writer — no boilerplate.',
+        'agent_file': 'agents/omg-qa-tester.md',
+    },
+    'escalation-router': {
+        'preferred_model': 'claude',
+        'model_version': 'claude-haiku-3-5',
+        'task_category': None,
+        'description': 'Routes problems to Codex/Gemini/CCG based on domain.',
+        'agent_file': 'agents/omg-escalation-router.md',
+    },
+}
+
+# Cache for model availability (per process)
+_model_cache: dict[str, bool] | None = None
+
+def resolve_agent(prompt_keywords: set[str]):
+    """Match prompt keywords to best agent. Returns registry entry or None.
+
+    Scoring: count of matching trigger_keywords. Returns highest-scoring agent.
+    Ties broken by order in registry (first wins).
+    """
+    best_agent = None
+    best_score = 0
+    for name, config in AGENT_REGISTRY.items():
+        triggers_raw = config.get('trigger_keywords', set())
+        triggers = triggers_raw if isinstance(triggers_raw, set) else set()
+        score = len(prompt_keywords & triggers)
+        if score > best_score:
+            best_score = score
+            best_agent = dict(config)
+            best_agent['name'] = name
+    return best_agent if best_score > 0 else None
+
+
+def get_dispatch_params(agent_name: str):
+    """Get task() parameters for dispatching this agent.
+
+    Returns dict with 'category', 'skills', and optionally 'subagent_type'.
+    Falls back to claude if preferred model not available.
+    """
+    config = AGENT_REGISTRY.get(agent_name, {})
+    available = detect_available_models()
+    preferred = config.get('preferred_model', 'claude')
+
+    # Resolve model availability
+    if preferred == 'gemini-cli' and not available.get('gemini-cli'):
+        preferred = 'claude'
+    elif preferred == 'codex-cli' and not available.get('codex-cli'):
+        preferred = 'claude'
+    elif preferred == 'domain-dependent':
+        preferred = 'claude'
+
+    params = {
+        'category': config.get('task_category', 'unspecified-high'),
+        'skills': config.get('skills', []),
+        'preferred_model': preferred,
+        'available_models': available,
+        'model_version': config.get('model_version', 'unknown'),
+    }
+    if 'subagent_type' in config:
+        params['subagent_type'] = config['subagent_type']
+    return params
+
+
+def detect_available_models() -> dict[str, bool]:
+    """Check which CLIs are available: codex-cli, gemini-cli.
+
+    Returns dict: {'claude': True, 'codex-cli': bool, 'gemini-cli': bool}
+    Caches result per process.
+    """
+    global _model_cache
+    if _model_cache is not None:
+        return _model_cache
+
+    result = {'claude': True}  # Claude is always available
+    result['codex-cli'] = shutil.which('codex') is not None
+    result['gemini-cli'] = shutil.which('gemini') is not None
+    _model_cache = result
+    return result
+
+
+def discover_mcp_tools() -> list[str]:
+    """Read MCP config to find available tool names.
+
+    Checks both project-level .mcp.json and ~/.claude/settings.json for mcpServers keys.
+    Returns list of server names (not individual tool names).
+    """
+    mcp_servers = {}
+    project_dir = os.getcwd()
+    
+    # Read from both sources: project .mcp.json and user settings
+    for mcp_loc in [
+        os.path.join(project_dir, '.mcp.json'),
+        os.path.expanduser('~/.claude/settings.json'),
+    ]:
+        if not os.path.exists(mcp_loc):
+            continue
+        try:
+            with open(mcp_loc) as f:
+                config = json.load(f)
+            servers = config.get('mcpServers', {})
+            mcp_servers.update(servers)
+        except (json.JSONDecodeError, OSError, KeyError):
+            continue
+    
+    return list(mcp_servers.keys())
+
+
+# --- Custom Agent Loading (Task 2.4) ---
+
+
+def load_custom_agents_into_registry(project_dir: str = ".") -> int:
+    """Load custom agents from user/project dirs into AGENT_REGISTRY.
+
+    If OMG_CUSTOM_AGENTS_ENABLED is disabled, does nothing.
+    Uses lazy import of runtime.custom_agent_loader to avoid circular deps.
+
+    Args:
+        project_dir: Project directory path.
+
+    Returns:
+        Number of custom agents loaded.
+    """
+    import sys as _sys
+
+    # Check feature flag via env var first (fast path)
+    env_val = os.environ.get("OMG_CUSTOM_AGENTS_ENABLED", "").lower()
+    if env_val in ("0", "false", "no"):
+        return 0
+
+    # If not explicitly enabled via env, check via _common
+    if env_val not in ("1", "true", "yes"):
+        try:
+            from _common import get_feature_flag  # pyright: ignore[reportMissingImports]
+            if not get_feature_flag("CUSTOM_AGENTS", default=False):
+                return 0
+        except ImportError:
+            return 0  # Can't check flag → disabled
+
+    # Lazy import custom_agent_loader from runtime/
+    _runtime_dir = os.path.join(os.path.dirname(os.path.abspath(__file__)), '..', 'runtime')
+    _runtime_dir = os.path.normpath(_runtime_dir)
+    if _runtime_dir not in _sys.path:
+        _sys.path.insert(0, _runtime_dir)
+
+    try:
+        from custom_agent_loader import load_custom_agents  # pyright: ignore[reportMissingImports]
+    except ImportError:
+        return 0
+
+    custom_agents = load_custom_agents(project_dir)
+    count = 0
+
+    for agent in custom_agents:
+        if not agent.get("validated", False):
+            continue  # Skip invalid agents
+
+        name = agent["name"]
+        AGENT_REGISTRY[name] = {
+            'preferred_model': 'claude',
+            'task_category': 'unspecified-high',
+            'skills': [],
+            'trigger_keywords': set(),
+            'mcp_tools': [],
+            'description': agent.get('description', ''),
+            'agent_file': agent.get('file', ''),
+            'model_version': 'claude-sonnet-4-5',
+            'model_role': agent.get('model_role'),
+            'source': 'custom',
+            'level': agent.get('level', 'unknown'),
+            'validated': True,
+        }
+        count += 1
+
+    return count

package/hooks/_budget.py

@@ -0,0 +1,31 @@
+#!/usr/bin/env python3
+"""
+Budget Constants — OMG Context & Prompt Budgets
+
+Named constants for token/character budgets across OMG hooks.
+Replaces magic numbers with semantic names for maintainability.
+"""
+
+# ═══════════════════════════════════════════════════════════
+# Session-start budgets (chars)
+# ═══════════════════════════════════════════════════════════
+BUDGET_SESSION_TOTAL = 2000          # Total context injection budget
+BUDGET_SESSION_IDLE = 200            # When no features active
+BUDGET_PROFILE = 200                 # Project profile section
+BUDGET_WORKING_MEMORY = 400          # Working memory section
+BUDGET_HANDOFF = 300                 # Handoff section
+BUDGET_MEMORY = 300                  # Memory/state section
+BUDGET_FAILURES = 200                # Active failures section
+BUDGET_TOOLS = 100                   # Tools inventory section
+BUDGET_PLANNING = 100                # Planning/checklist section
+BUDGET_RALPH = 100                   # Ralph/persistent mode section
+
+# ═══════════════════════════════════════════════════════════
+# Prompt-enhancer budgets (chars)
+# ═══════════════════════════════════════════════════════════
+BUDGET_PROMPT_TOTAL = 1000           # Total prompt enhancement budget
+BUDGET_INTENT_DISCIPLINE = 200       # Intent classification + discipline
+BUDGET_KNOWLEDGE = 300               # Knowledge retrieval section
+BUDGET_LEARNINGS = 200               # Learnings/patterns section
+BUDGET_AGENT_ROUTING = 200           # Agent routing directives
+BUDGET_MODE = 100                    # Mode detection (ulw/crazy/etc)