@trac3er/oh-my-god 2.1.8 → 2.2.2

package/runtime/tool_plan_gate.py

@@ -9,10 +9,14 @@ from uuid import uuid4
 import json
 
 from runtime.context_engine import ContextEngine
-from runtime.compliance_governor import evaluate_tool_compliance
+from runtime.compliance_governor import classify_bash_command_mode, evaluate_tool_compliance
 from runtime.complexity_scorer import score_complexity
-from runtime.release_run_coordinator import resolve_current_run_id as resolve_coordinator_run_id
+from runtime.release_run_coordinator import (
+    get_active_coordinator_run_id,
+    resolve_current_run_id as resolve_coordinator_run_id,
+)
 from runtime.runtime_contracts import read_run_state
+from runtime.test_intent_lock import verify_done_when, verify_lock
 
 
 _TOOL_KEYWORDS: dict[str, tuple[str, ...]] = {
@@ -21,6 +25,7 @@ _TOOL_KEYWORDS: dict[str, tuple[str, ...]] = {
     "omg_security_check": ("security", "scan", "vulnerability", "audit", "auth", "token", "secret"),
     "omg-control": ("policy", "governance", "release", "proof", "control plane"),
 }
+_MUTATION_TOOLS = frozenset({"write", "edit", "multiedit", "bash"})
 
 def build_tool_plan(
     goal: str,
@@ -102,33 +107,85 @@ def has_tool_plan_for_run(project_dir: str, run_id: str | None) -> bool:
     return any(plans_dir.glob(f"{run_prefix}-plan-*.json"))
 
 
-def tool_plan_gate_check(project_dir: str, run_id: str | None, tool: str) -> dict[str, object]:
-    if not run_id:
+def tool_plan_gate_check(
+    project_dir: str,
+    run_id: str | None,
+    tool: str,
+    tool_input: dict[str, object] | None = None,
+) -> dict[str, object]:
+    effective_run_id = get_active_coordinator_run_id(project_dir=project_dir) or run_id
+    if not effective_run_id:
         return {"status": "allowed", "reason": "run_id unavailable; skip tool plan gate"}
 
+    tool_input_obj = tool_input if isinstance(tool_input, dict) else {}
+    has_plan = has_tool_plan_for_run(project_dir, effective_run_id)
+
     context_packet = _resolve_bounded_context_packet(
         project_dir=project_dir,
-        run_id=run_id,
+        run_id=effective_run_id,
         fallback_context=None,
     )
     clarification = _clarification_status(context_packet)
     decision = evaluate_tool_compliance(
         project_dir=project_dir,
-        run_id=run_id,
+        run_id=effective_run_id,
         tool=tool,
-        has_tool_plan=has_tool_plan_for_run(project_dir, run_id),
+        has_tool_plan=has_plan,
         clarification_status=clarification,
+        tool_input=tool_input,
     )
     reason = str(decision.get("reason", ""))
     if decision.get("status") == "blocked":
-        _persist_gate_error(project_dir, run_id, tool, reason)
+        _persist_gate_error(project_dir, effective_run_id, tool, reason)
+        return {
+            **decision,
+            "run_id": effective_run_id,
+            "tool": tool,
+        }
+
+    if _is_mutation_capable_tool(tool, tool_input_obj) and not _is_docs_exemption(tool_input_obj):
+        lock_id = _extract_lock_id(tool_input_obj)
+        lock_verdict = verify_lock(project_dir, run_id=effective_run_id, lock_id=lock_id)
+        if lock_verdict.get("status") != "ok":
+            return {
+                "status": "blocked",
+                "authority": "test_intent_lock",
+                "reason": "test_intent_lock_required_before_mutation",
+                "run_id": effective_run_id,
+                "tool": tool,
+            }
+
+        metadata = _extract_metadata(tool_input_obj)
+        done_when_verdict = verify_done_when(metadata, run_id=effective_run_id)
+        if done_when_verdict.get("status") != "ok":
+            return {
+                "status": "blocked",
+                "authority": "done_when",
+                "reason": "done_when_required_before_mutation",
+                "run_id": effective_run_id,
+                "tool": tool,
+            }
+
     return {
         **decision,
-        "run_id": run_id,
+        "run_id": effective_run_id,
         "tool": tool,
     }
 
 
+def governed_tool_plan_gate_check(
+    project_dir: str,
+    run_id: str | None,
+    lane_name: str,
+    tool: str,
+) -> dict[str, object]:
+    decision = tool_plan_gate_check(project_dir=project_dir, run_id=run_id, tool=tool)
+    return {
+        **decision,
+        "lane_name": str(lane_name).strip().lower(),
+    }
+
+
 def _normalize_tools(available_tools: list[str]) -> list[str]:
     output: list[str] = []
     seen: set[str] = set()
@@ -334,3 +391,38 @@ def _clarification_status(context_packet: dict[str, object]) -> dict[str, object
         "clarification_prompt": prompt,
         "confidence": round(max(0.0, min(1.0, confidence)), 2),
     }
+
+
+def _is_mutation_capable_tool(tool: str, tool_input: dict[str, object]) -> bool:
+    token = str(tool or "").strip().lower()
+    if token == "bash":
+        return classify_bash_command_mode(str(tool_input.get("command", ""))) == "mutation"
+    return token in _MUTATION_TOOLS
+
+
+def _extract_metadata(tool_input: dict[str, object]) -> dict[str, object]:
+    metadata = tool_input.get("metadata")
+    if isinstance(metadata, dict):
+        return dict(metadata)
+    return {}
+
+
+def _extract_lock_id(tool_input: dict[str, object]) -> str | None:
+    lock_id = tool_input.get("lock_id")
+    if isinstance(lock_id, str) and lock_id.strip():
+        return lock_id.strip()
+    metadata = _extract_metadata(tool_input)
+    metadata_lock_id = metadata.get("lock_id")
+    if isinstance(metadata_lock_id, str) and metadata_lock_id.strip():
+        return metadata_lock_id.strip()
+    return None
+
+
+def _is_docs_exemption(tool_input: dict[str, object]) -> bool:
+    exemption = str(tool_input.get("exemption", "")).strip().lower()
+    if exemption == "docs":
+        return True
+    metadata = _extract_metadata(tool_input)
+    if metadata.get("exempt") is True and str(metadata.get("exemption", "")).strip().lower() == "docs":
+        return True
+    return False

package/runtime/untrusted_content.py

@@ -133,7 +133,7 @@ def write_trust_evidence(inputs: list[dict[str, Any]], output_dir: str | Path) -
     normalized_inputs: list[dict[str, Any]] = []
     for item in inputs:
         data = dict(item)
-        trust_tier = str(data.get("_trust_tier", "research"))
+        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))
@@ -203,20 +203,24 @@ def mark_untrusted_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,
+            "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)
-    provenance = list(state.get("provenance", []))
-    provenance.append(entry)
+
     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(
@@ -232,6 +236,7 @@ def mark_untrusted_content(
         "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:],

package/runtime/vision_artifacts.py

@@ -0,0 +1,31 @@
+"""Artifact writers for vision runtime jobs."""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+from typing import Any
+
+
+def write_vision_artifacts(
+    project_dir: str,
+    job: dict[str, Any],
+    deterministic_results: list[dict[str, Any]],
+) -> dict[str, Any]:
+    artifacts_dir = Path(project_dir) / ".omg" / "vision"
+    artifacts_dir.mkdir(parents=True, exist_ok=True)
+
+    artifact_path = artifacts_dir / f"{job['job_id']}.json"
+    payload = {
+        "schema": "VisionJobArtifact",
+        "job": job,
+        "results": deterministic_results,
+    }
+    artifact_path.write_text(json.dumps(payload, indent=2, sort_keys=True), encoding="utf-8")
+
+    return {
+        "artifact_path": str(artifact_path.relative_to(project_dir)),
+        "compare_count": sum(1 for item in deterministic_results if item.get("operation") == "compare"),
+        "ocr_count": sum(1 for item in deterministic_results if item.get("operation") == "ocr"),
+        "result_count": len(deterministic_results),
+    }

package/runtime/vision_cache.py

@@ -0,0 +1,38 @@
+"""Content-addressed cache helpers for vision jobs."""
+
+from __future__ import annotations
+
+import hashlib
+import json
+from pathlib import Path
+from typing import Any
+
+
+def _cache_dir(project_dir: str) -> Path:
+    path = Path(project_dir) / ".omg" / "cache" / "vision"
+    path.mkdir(parents=True, exist_ok=True)
+    return path
+
+
+def build_cache_key(mode: str, inputs: list[str]) -> str:
+    digest = hashlib.sha256()
+    digest.update(mode.encode("utf-8"))
+    for input_path in sorted(inputs):
+        digest.update(input_path.encode("utf-8"))
+        with open(input_path, "rb") as handle:
+            for chunk in iter(lambda: handle.read(65536), b""):
+                digest.update(chunk)
+    return digest.hexdigest()
+
+
+def load_cached_result(project_dir: str, cache_key: str) -> dict[str, Any] | None:
+    cache_path = _cache_dir(project_dir) / f"{cache_key}.json"
+    if not cache_path.exists():
+        return None
+    return json.loads(cache_path.read_text(encoding="utf-8"))
+
+
+def store_cached_result(project_dir: str, cache_key: str, payload: dict[str, Any]) -> str:
+    cache_path = _cache_dir(project_dir) / f"{cache_key}.json"
+    cache_path.write_text(json.dumps(payload, indent=2, sort_keys=True), encoding="utf-8")
+    return str(cache_path)

package/runtime/vision_jobs.py

@@ -0,0 +1,89 @@
+"""Runtime orchestration for vision jobs."""
+
+from __future__ import annotations
+
+import hashlib
+import itertools
+from pathlib import Path
+from typing import Any
+
+from omg_natives.image import image
+
+from runtime.vision_artifacts import write_vision_artifacts
+from runtime.vision_cache import build_cache_key, load_cached_result, store_cached_result
+
+
+VALID_VISION_MODES = frozenset({"ocr", "compare", "analyze", "batch", "eval"})
+IMPLEMENTED_VISION_MODES = frozenset({"ocr", "compare"})
+
+
+def _resolve_inputs(project_dir: str, inputs: list[str]) -> list[str]:
+    return [str((Path(project_dir) / item).resolve()) for item in inputs]
+
+
+def _job_id(mode: str, inputs: list[str]) -> str:
+    digest = hashlib.sha256()
+    digest.update(mode.encode("utf-8"))
+    for item in inputs:
+        digest.update(item.encode("utf-8"))
+    return f"vision-{digest.hexdigest()[:12]}"
+
+
+def normalize_vision_payload(project_dir: str, payload: dict[str, Any]) -> dict[str, Any]:
+    mode = str(payload.get("mode", "")).strip()
+    raw_inputs = payload.get("inputs")
+    if mode not in VALID_VISION_MODES:
+        raise ValueError(f"unsupported vision mode: {mode}")
+    if mode not in IMPLEMENTED_VISION_MODES:
+        raise ValueError(f"vision mode is not implemented yet: {mode}")
+    if not isinstance(raw_inputs, list) or not raw_inputs or not all(isinstance(item, str) and item.strip() for item in raw_inputs):
+        raise ValueError("inputs must be a non-empty list of file paths")
+    if mode == "compare" and len(raw_inputs) < 2:
+        raise ValueError("compare mode requires at least two input files")
+
+    resolved_inputs = _resolve_inputs(project_dir, raw_inputs)
+    for input_path in resolved_inputs:
+        if not Path(input_path).exists():
+            raise ValueError(f"missing vision input: {input_path}")
+
+    return {
+        "job_id": _job_id(mode, resolved_inputs),
+        "mode": mode,
+        "inputs": resolved_inputs,
+    }
+
+
+def expand_pairs(job: dict[str, Any]) -> list[tuple[str, str]]:
+    if job["mode"] != "compare":
+        return []
+    return list(itertools.combinations(job["inputs"], 2))
+
+
+def run_vision_job(project_dir: str, payload: dict[str, Any]) -> dict[str, Any]:
+    job = normalize_vision_payload(project_dir, payload)
+    cache_key = build_cache_key(job["mode"], job["inputs"])
+    cached = load_cached_result(project_dir, cache_key)
+    if cached is not None:
+        cached["cached"] = True
+        return cached
+
+    if job["mode"] == "compare":
+        deterministic_results = [
+            image(left_path, "compare", other_path=right_path)
+            for left_path, right_path in expand_pairs(job)
+        ]
+    elif job["mode"] == "ocr":
+        deterministic_results = [image(input_path, "ocr") for input_path in job["inputs"]]
+    else:
+        deterministic_results = []
+
+    artifacts = write_vision_artifacts(project_dir, job, deterministic_results)
+    result = {
+        "status": "ok",
+        "job": job,
+        "results": deterministic_results,
+        "artifacts": artifacts,
+        "cached": False,
+    }
+    store_cached_result(project_dir, cache_key, result)
+    return result