@trac3er/oh-my-god 2.2.3 → 2.3.0

package/build/lib/runtime/untrusted_content.py

@@ -1,307 +0,0 @@
-"""State and provenance tracking for untrusted external content."""
-from __future__ import annotations
-
-from dataclasses import dataclass
-from datetime import datetime, timezone
-from hashlib import sha256
-import json
-from pathlib import Path
-import re
-from enum import Enum
-from typing import Any
-
-
-STATE_REL_PATH = Path(".omg") / "state" / "untrusted-content.json"
-EVIDENCE_REL_DIR = Path(".omg") / "evidence"
-_INSTRUCTION_PATTERNS: tuple[re.Pattern[str], ...] = (
-    re.compile(r"ignore\s+previous\s+instructions", re.IGNORECASE),
-    re.compile(r"\b(system|assistant|developer)\s*:", re.IGNORECASE),
-    re.compile(r"\b(run|execute|commit|push|apply_patch|edit)\b", re.IGNORECASE),
-    re.compile(r"<\s*/?(?:system|assistant|user)\s*>", re.IGNORECASE),
-    re.compile(r"\[INST\]|\[/INST\]", re.IGNORECASE),
-    re.compile(r"###\s*(?:System|Human|Assistant)\s*:", re.IGNORECASE),
-    re.compile(r"OVERRIDE\s*(?:INSTRUCTIONS|SYSTEM)", re.IGNORECASE),
-    re.compile(r"disregard\s+(?:your|previous|all)\s+(?:instructions|rules)", re.IGNORECASE),
-    re.compile(r"(?:you are|you're)\s+(?:now|actually|really)\s+(?:a|an)\s+", re.IGNORECASE),
-    re.compile(r"(?:jailbreak|DAN|AIM|UCAR)\s*(?:mode|prompt|persona)?", re.IGNORECASE),
-    re.compile(r"(?:bypass|disable)\s+(?:safety|guardrails|filters?)", re.IGNORECASE),
-)
-
-
-class TrustTier(str, Enum):
-    LOCAL = "local"
-    BALANCED = "balanced"
-    RESEARCH = "research"
-    BROWSER = "browser"
-
-
-@dataclass(frozen=True)
-class TrustTierConfig:
-    tier: TrustTier
-    label: str
-    score: float
-    trusted: bool
-
-
-TRUST_TIER_CONFIG: dict[TrustTier, TrustTierConfig] = {
-    TrustTier.LOCAL: TrustTierConfig(
-        tier=TrustTier.LOCAL,
-        label="TRUSTED_LOCAL_CONTENT",
-        score=1.0,
-        trusted=True,
-    ),
-    TrustTier.BALANCED: TrustTierConfig(
-        tier=TrustTier.BALANCED,
-        label="TRUSTED_LOCAL_CONTENT",
-        score=0.7,
-        trusted=True,
-    ),
-    TrustTier.RESEARCH: TrustTierConfig(
-        tier=TrustTier.RESEARCH,
-        label="UNTRUSTED_EXTERNAL_CONTENT",
-        score=0.0,
-        trusted=False,
-    ),
-    TrustTier.BROWSER: TrustTierConfig(
-        tier=TrustTier.BROWSER,
-        label="UNTRUSTED_EXTERNAL_CONTENT",
-        score=0.0,
-        trusted=False,
-    ),
-}
-
-
-_SOURCE_TO_TIER: dict[str, TrustTier] = {
-    "local": TrustTier.LOCAL,
-    "workspace": TrustTier.LOCAL,
-    "balanced": TrustTier.BALANCED,
-    "tool": TrustTier.BALANCED,
-    "web": TrustTier.RESEARCH,
-    "research": TrustTier.RESEARCH,
-    "mcp": TrustTier.RESEARCH,
-    "remote_document": TrustTier.RESEARCH,
-    "browser": TrustTier.BROWSER,
-    "dom": TrustTier.BROWSER,
-    "screenshot": TrustTier.BROWSER,
-}
-
-
-def normalize_trust_tier(tier: TrustTier | str) -> TrustTier:
-    if isinstance(tier, TrustTier):
-        return tier
-    normalized = str(tier).strip().lower()
-    if normalized in TrustTier._value2member_map_:
-        return TrustTier(normalized)
-    raise ValueError(f"Unsupported trust tier: {tier}")
-
-
-def trust_tier_for_source(source_type: str) -> TrustTier:
-    source_key = str(source_type).strip().lower()
-    return _SOURCE_TO_TIER.get(source_key, TrustTier.RESEARCH)
-
-
-def trust_tier_for_preset(preset: str) -> TrustTier:
-    mapping = {
-        "safe": TrustTier.LOCAL,
-        "balanced": TrustTier.BALANCED,
-        "interop": TrustTier.RESEARCH,
-        "labs": TrustTier.BROWSER,
-    }
-    return mapping.get(str(preset).strip().lower(), TrustTier.BALANCED)
-
-
-def tag_content(payload: Any, tier: TrustTier | str) -> dict[str, Any]:
-    normalized_tier = normalize_trust_tier(tier)
-    config = TRUST_TIER_CONFIG[normalized_tier]
-    tagged: dict[str, Any]
-    if isinstance(payload, dict):
-        tagged = dict(payload)
-    else:
-        tagged = {"content": payload}
-    tagged["_trust_tier"] = normalized_tier.value
-    tagged["_trust_label"] = config.label
-    tagged["_trust_score"] = config.score
-    return tagged
-
-
-def write_trust_evidence(inputs: list[dict[str, Any]], output_dir: str | Path) -> str:
-    output_path = Path(output_dir)
-    output_path.mkdir(parents=True, exist_ok=True)
-    timestamp = datetime.now(timezone.utc).strftime("%Y%m%dT%H%M%S%fZ")
-    evidence_path = output_path / f"trust-{timestamp}.json"
-
-    normalized_inputs: list[dict[str, Any]] = []
-    for item in inputs:
-        data = dict(item)
-        trust_tier = str(data.get("trust_tier", data.get("_trust_tier", "research")))
-        trust_label = str(data.get("_trust_label", "UNTRUSTED_EXTERNAL_CONTENT"))
-        try:
-            trust_score = float(data.get("_trust_score", 0.0))
-        except (TypeError, ValueError):
-            trust_score = 0.0
-        content_repr = data.get("sanitized_content", data.get("content", ""))
-        if not isinstance(content_repr, str):
-            content_repr = json.dumps(content_repr, sort_keys=True, ensure_ascii=True)
-        content_hash = sha256(content_repr.encode("utf-8")).hexdigest()
-        normalized_inputs.append(
-            {
-                "source_type": str(data.get("source_type", "unknown")),
-                "source_ref": str(data.get("source_ref", "")),
-                "trust_tier": trust_tier,
-                "trust_label": trust_label,
-                "trust_score": trust_score,
-                "content_hash": content_hash,
-            }
-        )
-
-    payload = {
-        "schema": "TrustEvidence",
-        "generated_at": datetime.now(timezone.utc).isoformat(),
-        "input_count": len(normalized_inputs),
-        "inputs": normalized_inputs,
-    }
-    evidence_path.write_text(json.dumps(payload, indent=2, ensure_ascii=True) + "\n", encoding="utf-8")
-    return evidence_path.as_posix()
-
-
-def write_sandbox_budget_evidence(
-    *,
-    project_dir: str | Path,
-    run_id: str,
-    budget: dict[str, Any],
-    isolation: dict[str, Any],
-) -> str:
-    tagged = tag_content(
-        {
-            "source_type": "workspace",
-            "source_ref": f"forge:{run_id}",
-            "content": {
-                "run_id": run_id,
-                "budget": budget,
-                "isolation": isolation,
-            },
-        },
-        TrustTier.BALANCED,
-    )
-    return write_trust_evidence(
-        [tagged],
-        output_dir=Path(project_dir) / EVIDENCE_REL_DIR,
-    )
-
-
-def mark_untrusted_content(
-    project_dir: str,
-    *,
-    source_type: str,
-    content: str,
-    source_ref: str = "",
-    tier: TrustTier | str | None = None,
-) -> dict[str, Any]:
-    state_path = _state_path(project_dir)
-    state = get_untrusted_content_state(project_dir)
-    sanitized, quarantined = quarantine_instruction_like_text(content)
-    resolved_tier = normalize_trust_tier(tier) if tier is not None else trust_tier_for_source(source_type)
-    entry = tag_content(
-        {
-            "source_type": source_type,
-            "source_ref": source_ref,
-            "quarantined_fragments": quarantined,
-            "sanitized_content": sanitized,
-        },
-        resolved_tier,
-    )
-    entry["trust_tier"] = resolved_tier.value
-    entry["trust_score"] = entry.get("_trust_score", 0.0)
-
-    evidence_path = write_trust_evidence(
-        [entry],
-        output_dir=Path(project_dir) / EVIDENCE_REL_DIR,
-    )
-    entry["evidence_artifact"] = evidence_path
-
-    provenance = list(state.get("provenance", []))
-    provenance.append(entry)
-
-    prior_scores = state.get("trust_scores", {})
-    next_external_score = min(
-        [
-            float(prior_scores.get("external_content", 1.0)),
-            float(entry.get("_trust_score", 0.0)),
-        ]
-    )
-    next_state = {
-        "active": resolved_tier in {TrustTier.RESEARCH, TrustTier.BROWSER},
-        "last_source_type": source_type,
-        "last_source_ref": source_ref,
-        "last_trust_tier": resolved_tier.value,
-        "last_trust_label": entry.get("_trust_label"),
-        "last_trust_score": entry.get("_trust_score"),
-        "last_evidence_artifact": evidence_path,
-        "sanitized_content": sanitized,
-        "quarantined_fragments": quarantined,
-        "provenance": provenance[-20:],
-        "trust_scores": {
-            **(prior_scores if isinstance(prior_scores, dict) else {}),
-            "external_content": next_external_score,
-            resolved_tier.value: float(entry.get("_trust_score", 0.0)),
-        },
-        "evidence_artifacts": (list(state.get("evidence_artifacts", [])) + [evidence_path])[-20:],
-    }
-    _write_state(state_path, next_state)
-    return next_state
-
-
-def clear_untrusted_content(project_dir: str, *, reason: str) -> dict[str, Any]:
-    state_path = _state_path(project_dir)
-    existing = get_untrusted_content_state(project_dir)
-    next_state = {
-        **existing,
-        "active": False,
-        "cleared_reason": reason,
-    }
-    _write_state(state_path, next_state)
-    return next_state
-
-
-def get_untrusted_content_state(project_dir: str) -> dict[str, Any]:
-    state_path = _state_path(project_dir)
-    if not state_path.exists():
-        return {
-            "active": False,
-            "provenance": [],
-            "trust_scores": {},
-        }
-    try:
-        payload = json.loads(state_path.read_text(encoding="utf-8"))
-    except (OSError, json.JSONDecodeError):
-        return {
-            "active": False,
-            "provenance": [],
-            "trust_scores": {},
-        }
-    return payload if isinstance(payload, dict) else {"active": False, "provenance": [], "trust_scores": {}}
-
-
-def is_untrusted_content_mode_active(project_dir: str) -> bool:
-    return bool(get_untrusted_content_state(project_dir).get("active", False))
-
-
-def quarantine_instruction_like_text(content: str) -> tuple[str, list[str]]:
-    sanitized_lines: list[str] = []
-    quarantined: list[str] = []
-    for raw_line in content.splitlines():
-        line = raw_line.strip()
-        if any(pattern.search(line) for pattern in _INSTRUCTION_PATTERNS):
-            quarantined.append(raw_line)
-            continue
-        sanitized_lines.append(raw_line)
-    return "\n".join(sanitized_lines).strip(), quarantined
-
-
-def _state_path(project_dir: str) -> Path:
-    return Path(project_dir) / STATE_REL_PATH
-
-
-def _write_state(path: Path, payload: dict[str, Any]) -> None:
-    path.parent.mkdir(parents=True, exist_ok=True)
-    path.write_text(json.dumps(payload, indent=2, ensure_ascii=True) + "\n", encoding="utf-8")

package/build/lib/runtime/validate.py

@@ -1,293 +0,0 @@
-"""Canonical validation engine — composes doctor, contract, profile, and install checks.
-
-This module is the single entrypoint for ``omg validate``.  It delegates to
-existing checkers (``run_doctor``, ``validate_contract_registry``, profile
-governor, install integrity) rather than duplicating their logic.
-"""
-from __future__ import annotations
-
-import os
-import shutil
-from pathlib import Path
-from typing import Any
-
-from runtime.adoption import CANONICAL_VERSION
-from runtime.compat import run_doctor, _doctor_check
-from runtime.contract_compiler import validate_contract_registry
-from runtime.plugin_diagnostics import run_plugin_diagnostics
-from runtime.profile_io import load_profile, ensure_governed_preferences, assess_profile_risk
-
-
-def _check_contract_registry(root_dir: Path) -> dict[str, Any]:
-    """Compose contract registry validation into a single doctor-style check."""
-    try:
-        result = validate_contract_registry(root_dir)
-    except Exception as exc:
-        return _doctor_check(
-            "contract_registry",
-            ok=False,
-            message=f"contract validation error: {exc}",
-        )
-    if result.get("status") == "ok":
-        error_count = len(result.get("errors", []))
-        return _doctor_check(
-            "contract_registry",
-            ok=True,
-            message=f"contract registry valid ({error_count} errors)",
-        )
-    errors = result.get("errors", [])
-    summary = errors[0] if errors else "unknown error"
-    if len(errors) > 1:
-        summary += f" (+{len(errors) - 1} more)"
-    return _doctor_check(
-        "contract_registry",
-        ok=False,
-        message=f"contract registry invalid: {summary}",
-    )
-
-
-def _check_profile_governor(root_dir: Path) -> dict[str, Any]:
-    """Validate that governed profile state is consistent."""
-    profile_path = os.path.join(root_dir, ".omg", "state", "profile.yaml")
-    profile = load_profile(profile_path)
-
-    if not profile:
-        return _doctor_check(
-            "profile_governor",
-            ok=True,
-            message="no profile found (optional — run /OMG:init to create)",
-            required=False,
-        )
-
-    try:
-        ensure_governed_preferences(profile)
-    except Exception as exc:
-        return _doctor_check(
-            "profile_governor",
-            ok=False,
-            message=f"governed preferences malformed: {exc}",
-            required=False,
-        )
-
-    governed = profile.get("governed_preferences", {})
-    governed = governed if isinstance(governed, dict) else {}
-    style_entries = governed.get("style", [])
-    style_entries = style_entries if isinstance(style_entries, list) else []
-    safety_entries = governed.get("safety", [])
-    safety_entries = safety_entries if isinstance(safety_entries, list) else []
-
-    pending = sum(
-        1 for e in style_entries + safety_entries
-        if isinstance(e, dict) and e.get("confirmation_state") == "pending_confirmation"
-    )
-
-    risk = assess_profile_risk(profile)
-    risk_level = str(risk.get("risk_level", "low"))
-
-    parts = [f"{len(style_entries)} style, {len(safety_entries)} safety"]
-    if pending:
-        parts.append(f"{pending} pending confirmation")
-    parts.append(f"risk={risk_level}")
-
-    return _doctor_check(
-        "profile_governor",
-        ok=True,
-        message=f"profile governor ok ({', '.join(parts)})",
-        required=False,
-    )
-
-
-def _check_install_integrity(root_dir: Path) -> dict[str, Any]:
-    """Verify critical install paths exist and are coherent."""
-    issues: list[str] = []
-
-    # Check scripts/omg.py exists
-    omg_script = root_dir / "scripts" / "omg.py"
-    if not omg_script.exists():
-        issues.append("scripts/omg.py missing")
-
-    # Check runtime/ package exists
-    runtime_init = root_dir / "runtime" / "__init__.py"
-    runtime_dir = root_dir / "runtime"
-    if not runtime_dir.exists():
-        issues.append("runtime/ directory missing")
-
-    # Check commands/ directory exists
-    commands_dir = root_dir / "commands"
-    if not commands_dir.exists():
-        issues.append("commands/ directory missing")
-
-    # Check plugins/core/plugin.json exists
-    plugin_json = root_dir / "plugins" / "core" / "plugin.json"
-    if not plugin_json.exists():
-        issues.append("plugins/core/plugin.json missing")
-
-    # Check pyproject.toml exists
-    pyproject = root_dir / "pyproject.toml"
-    if not pyproject.exists():
-        issues.append("pyproject.toml missing")
-
-    if issues:
-        return _doctor_check(
-            "install_integrity",
-            ok=False,
-            message=f"install issues: {'; '.join(issues)}",
-        )
-    return _doctor_check(
-        "install_integrity",
-        ok=True,
-        message="install integrity ok (scripts, runtime, commands, plugins present)",
-    )
-
-
-def _check_plugin_compatibility(root_dir: Path) -> dict[str, Any]:
-    try:
-        result = run_plugin_diagnostics(str(root_dir))
-    except Exception as exc:
-        return _doctor_check(
-            "plugin_compatibility",
-            ok=False,
-            message=f"plugin diagnostics error: {exc}",
-        )
-
-    status = str(result.get("status", "error"))
-    summary = result.get("summary", {})
-    summary = summary if isinstance(summary, dict) else {}
-    total_records = int(summary.get("total_records", 0))
-    total_conflicts = int(summary.get("total_conflicts", 0))
-    blockers = int(summary.get("blockers", 0))
-
-    return _doctor_check(
-        "plugin_compatibility",
-        ok=status in {"ok", "warn"},
-        message=(
-            f"plugin compatibility: {total_records} records, "
-            f"{total_conflicts} conflicts, {blockers} blockers"
-        ),
-    )
-
-
-def _load_selected_mcps(root_dir: Path) -> list[str]:
-    """Load selected MCP IDs from cli-config.yaml.
-
-    Checks CLAUDE_PROJECT_DIR env first, then falls back to root_dir.
-    Returns an empty list when the config file is absent or unparseable.
-    """
-    import yaml as _yaml
-
-    project_dir = os.environ.get("CLAUDE_PROJECT_DIR", "")
-    if project_dir:
-        config_path = os.path.join(project_dir, ".omg", "state", "cli-config.yaml")
-    else:
-        config_path = os.path.join(root_dir, ".omg", "state", "cli-config.yaml")
-
-    if not os.path.isfile(config_path):
-        return []
-
-    try:
-        with open(config_path, "r", encoding="utf-8") as f:
-            data = _yaml.safe_load(f)
-    except Exception:
-        return []
-
-    if not isinstance(data, dict):
-        return []
-
-    mcps = data.get("selected_mcps", [])
-    return [str(m) for m in mcps] if isinstance(mcps, list) else []
-
-
-def _check_notebooklm(root_dir: Path) -> dict[str, Any] | None:
-    """Optional NotebookLM health check — only runs when notebooklm is selected.
-
-    Returns None when NotebookLM is not in the selected MCP set (meaning:
-    skip entirely, no check emitted).  When selected, checks that ``npx``
-    is reachable on PATH.  Missing npx is a *warning*, never a blocker.
-    """
-    selected = _load_selected_mcps(root_dir)
-    if "notebooklm" not in selected:
-        return None
-
-    npx_path = shutil.which("npx")
-    if npx_path is None:
-        return _doctor_check(
-            "notebooklm",
-            ok=False,
-            message="npx not found — install Node.js to use NotebookLM",
-            required=False,
-        )
-
-    return _doctor_check(
-        "notebooklm",
-        ok=True,
-        message="npx available, notebooklm-mcp callable",
-        required=False,
-    )
-
-
-def run_validate(*, root_dir: Path | None = None) -> dict[str, Any]:
-    """Run all validation checks and return a structured result.
-
-    Composes:
-    1. ``run_doctor()`` — install and runtime checks
-    2. Contract registry validation
-    3. Profile governor validation
-    4. Install integrity check
-    5. Optional NotebookLM check (only when selected in MCP config)
-
-    Returns a dict matching the ``ValidateResult`` schema:
-    ``{"schema": "ValidateResult", "status": "pass"|"fail", "checks": [...], "version": "..."}``
-    """
-    repo_root = root_dir or Path(__file__).resolve().parent.parent
-
-    checks: list[dict[str, Any]] = []
-
-    # 1. Compose doctor checks (never duplicate — delegate directly)
-    doctor_result = run_doctor(root_dir=repo_root)
-    checks.extend(doctor_result.get("checks", []))
-
-    # 2. Contract registry
-    checks.append(_check_contract_registry(repo_root))
-
-    # 3. Profile governor
-    checks.append(_check_profile_governor(repo_root))
-
-    # 4. Install integrity
-    checks.append(_check_install_integrity(repo_root))
-
-    if not any(c.get("name") == "plugin_compatibility" for c in checks):
-        checks.append(_check_plugin_compatibility(repo_root))
-
-    nb_check = _check_notebooklm(repo_root)
-    if nb_check is not None:
-        checks.append(nb_check)
-
-    has_blocker = any(c["status"] == "blocker" for c in checks)
-
-    return {
-        "schema": "ValidateResult",
-        "status": "fail" if has_blocker else "pass",
-        "checks": checks,
-        "version": CANONICAL_VERSION,
-    }
-
-
-def format_text(result: dict[str, Any]) -> str:
-    """Format a ValidateResult as human-readable text."""
-    lines: list[str] = []
-    for check in result.get("checks", []):
-        if check["status"] == "ok":
-            marker = "PASS"
-        elif check["status"] == "blocker":
-            marker = "BLOCKER"
-        else:
-            marker = "WARN"
-        req_tag = "" if check.get("required", True) else " (optional)"
-        lines.append(f"  {marker:>7} {check['name']}: {check['message']}{req_tag}")
-
-    blockers = sum(1 for c in result.get("checks", []) if c["status"] == "blocker")
-    warnings = sum(1 for c in result.get("checks", []) if c["status"] == "warning")
-    passed = sum(1 for c in result.get("checks", []) if c["status"] == "ok")
-    lines.append(f"\nPASS [{passed}] | WARN [{warnings}] | BLOCKER [{blockers}]")
-
-    return "\n".join(lines)