Merge pull request #52285 from NousResearch/bb/verify-ledger

feat(agent): record coding verification evidence

tools/file_tools.py

@@ -1266,8 +1266,43 @@ def _check_file_staleness(filepath: str, task_id: str) -> str | None:
     return None
 
 
+def _mark_verification_stale(
+    task_id: str,
+    resolved_paths: list[str],
+    session_id: str | None = None,
+) -> None:
+    """Best-effort note that successful edits made prior verification stale."""
+    paths = [p for p in resolved_paths if p]
+    if not paths:
+        return
+    try:
+        from agent.coding_context import project_facts_for
+        from agent.verification_evidence import mark_workspace_edited
+
+        cwd = None
+        for path in paths:
+            try:
+                candidate = str(Path(path).parent)
+            except Exception:
+                continue
+            if project_facts_for(candidate):
+                cwd = candidate
+                break
+        if cwd is None:
+            cwd = _authoritative_workspace_root(task_id)
+        if cwd is None:
+            try:
+                cwd = str(Path(paths[0]).parent)
+            except Exception:
+                cwd = None
+        mark_workspace_edited(session_id=session_id or task_id, cwd=cwd, paths=paths)
+    except Exception:
+        logger.debug("verification stale marker failed", exc_info=True)
+
+
 def write_file_tool(path: str, content: str, task_id: str = "default",
-                    cross_profile: bool = False) -> str:
+                    cross_profile: bool = False,
+                    session_id: str | None = None) -> str:
     """Write content to a file.
 
     ``cross_profile`` opts out of the soft cross-Hermes-profile guard. The
@@ -1305,6 +1340,8 @@ def write_file_tool(path: str, content: str, task_id: str = "default",
             result_dict = result.to_dict()
             if stale_warning:
                 result_dict["_warning"] = stale_warning
+            if not result_dict.get("error"):
+                _mark_verification_stale(task_id, [path], session_id=session_id)
             _update_read_timestamp(path, task_id)
             return json.dumps(result_dict, ensure_ascii=False)
 
@@ -1331,6 +1368,7 @@ def write_file_tool(path: str, content: str, task_id: str = "default",
             result_dict["resolved_path"] = _resolved
             if not result_dict.get("error"):
                 result_dict["files_modified"] = [_resolved]
+                _mark_verification_stale(task_id, [_resolved], session_id=session_id)
             # Refresh stamps after the successful write so consecutive
             # writes by this task don't trigger false staleness warnings.
             _update_read_timestamp(path, task_id)
@@ -1347,7 +1385,8 @@ def write_file_tool(path: str, content: str, task_id: str = "default",
 
 def patch_tool(mode: str = "replace", path: str = None, old_string: str = None,
                new_string: str = None, replace_all: bool = False, patch: str = None,
-               task_id: str = "default", cross_profile: bool = False) -> str:
+               task_id: str = "default", cross_profile: bool = False,
+               session_id: str | None = None) -> str:
     """Patch a file using replace mode or V4A patch format.
 
     ``cross_profile`` opts out of the soft cross-Hermes-profile guard for
@@ -1465,6 +1504,7 @@ def patch_tool(mode: str = "replace", path: str = None, old_string: str = None,
                 result_dict["files_modified"] = _resolved_modified
                 if len(_resolved_modified) == 1:
                     result_dict["resolved_path"] = _resolved_modified[0]
+                _mark_verification_stale(task_id, _resolved_modified, session_id=session_id)
                 for _p in _paths_to_check:
                     _update_read_timestamp(_p, task_id)
                     _r = _path_to_resolved.get(_p)
@@ -1730,6 +1770,7 @@ def _handle_write_file(args, **kw):
     return write_file_tool(
         path=args["path"], content=args["content"], task_id=tid,
         cross_profile=bool(args.get("cross_profile", False)),
+        session_id=kw.get("session_id"),
     )
 
 
@@ -1740,6 +1781,7 @@ def _handle_patch(args, **kw):
         old_string=args.get("old_string"), new_string=args.get("new_string"),
         replace_all=args.get("replace_all", False), patch=args.get("patch"), task_id=tid,
         cross_profile=bool(args.get("cross_profile", False)),
+        session_id=kw.get("session_id"),
     )
 
 

tools/terminal_tool.py

@@ -1872,6 +1872,7 @@ def terminal_tool(
     background: bool = False,
     timeout: Optional[int] = None,
     task_id: Optional[str] = None,
+    session_id: Optional[str] = None,
     force: bool = False,
     workdir: Optional[str] = None,
     pty: bool = False,
@@ -1886,6 +1887,7 @@ def terminal_tool(
         background: Whether to run in background (default: False)
         timeout: Command timeout in seconds (default: from config)
         task_id: Unique identifier for environment isolation (optional)
+        session_id: Conversation/session identifier for durable observability
         force: If True, skip dangerous command check (use after user confirms)
         workdir: Working directory for this command (optional, uses session cwd if not set)
         pty: If True, use pseudo-terminal for interactive CLI tools (local backend only)
@@ -2441,16 +2443,18 @@ def terminal_tool(
             max_retries = 3
             retry_count = 0
             result = None
+            command_cwd = None
             
             while retry_count <= max_retries:
                 try:
+                    command_cwd = _resolve_command_cwd(
+                        workdir=workdir,
+                        env=env,
+                        default_cwd=cwd,
+                    )
                     execute_kwargs = {
                         "timeout": effective_timeout,
-                        "cwd": _resolve_command_cwd(
-                            workdir=workdir,
-                            env=env,
-                            default_cwd=cwd,
-                        ),
+                        "cwd": command_cwd,
                     }
                     result = env.execute(command, **execute_kwargs)
                 except Exception as e:
@@ -2541,6 +2545,25 @@ def terminal_tool(
                 "exit_code": returncode,
                 "error": None,
             }
+            try:
+                from agent.verification_evidence import record_terminal_result
+
+                evidence = record_terminal_result(
+                    command=command,
+                    cwd=command_cwd,
+                    session_id=session_id or task_id or effective_task_id or "default",
+                    exit_code=returncode,
+                    output=output,
+                )
+                if evidence:
+                    result_dict["verification_evidence"] = {
+                        "status": evidence.get("status"),
+                        "kind": evidence.get("kind"),
+                        "scope": evidence.get("scope"),
+                        "canonical_command": evidence.get("canonical_command"),
+                    }
+            except Exception:
+                logger.debug("verification evidence recording failed", exc_info=True)
             if approval_note:
                 result_dict["approval"] = approval_note
             if exit_note:
@@ -2774,6 +2797,7 @@ def _handle_terminal(args, **kw):
         background=args.get("background", False),
         timeout=args.get("timeout"),
         task_id=kw.get("task_id"),
+        session_id=kw.get("session_id"),
         workdir=args.get("workdir"),
         pty=args.get("pty", False),
         notify_on_complete=args.get("notify_on_complete", False),