feat(server): truth-kernel hardening Wave 11 -- source() pagination + unified continuation state field

Wave 11 (items 1 and 3 of the kept roadmap; item 4 -- a multi-agent
dogfood benchmark -- is a measurement exercise, not a code change, and
is deliberately not attempted here; item 2 was already dropped as
unnecessary since native-Edit + diff_impact already covers it):

- item 1 (response budget): opt-in max_lines/resume_from_line params on
  source(), truncated/omitted_lines/next_cursor on the response --
  directly closes the cited overflow case (a single source() call with
  no way to ask for less). etag/line_start/line_end always cover the
  FULL range regardless of pagination -- range-identity, not tied to how
  much was rendered. Also fixes a real bug found while wiring this up:
  with_line_gutters always numbered from the symbol's start line even
  under pagination, which would have mislabeled every paginated
  response's line numbers.

- item 3 (unified continuation state): every ToolOutcome/ResolvedOutcome
  response now carries state: "ready" | "needs_context" |
  "needs_human_review" | "needs_verification" | "blocked", added once in
  the two shared wrapper constructors (classify_error_state in
  outcome.rs) rather than at any of the ~150 individual error_detail()
  call sites. All 74 error codes the codebase currently emits get an
  explicit classification -- not derived from `recoverable` alone, which
  turned out to be an unreliable proxy (most INVALID_*/*_REQUIRED codes
  are recoverable: false yet are exactly "fix your params and retry").

2 new regression tests prove actual truncation + resume-from-cursor +
unbounded-is-unaffected, and prove state distinguishes ready vs
needs_context on a real response. All 40 __toolsnaps__ schema snapshots
regenerated (state is now a required property on every tool's output).

1745 tests green, clippy/fmt clean.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Author: Your Name

crates/calm-server/src/__toolsnaps__/batch_status.snap

@@ -82,7 +82,15 @@
         ]
       }
     },
+    "required": [
+      "state"
+    ],
+    "type": "object",
     "properties": {
+      "state": {
+        "description": "Wave 11 (item 3, \"unified continuation state machine\"): present on\nevery response regardless of success/error -- `\"ready\"` on success,\nor one of `needs_context`/`needs_human_review`/`needs_verification`/\n`blocked` on error (see `classify_error_state`'s own doc comment).\nLets an agent dispatch on ONE top-level field instead of\nmaintaining its own per-`code` remediation table across ~70\ndistinct codes.",
+        "type": "string"
+      },
       "error": {
         "anyOf": [
           {
@@ -134,8 +142,7 @@
           }
         ]
       }
-    },
-    "type": "object"
+    }
   },
   "annotations": {
     "readOnlyHint": true,

crates/calm-server/src/__toolsnaps__/callees.snap

@@ -272,7 +272,15 @@
         ]
       }
     },
+    "required": [
+      "state"
+    ],
+    "type": "object",
     "properties": {
+      "state": {
+        "description": "See `ToolOutcome::state`'s own doc comment -- same contract, plus\n`ambiguous`/`not_found` both map onto `needs_context` (the caller\nneeds one more piece of information: which candidate, or a\ndifferent query).",
+        "type": "string"
+      },
       "error": {
         "anyOf": [
           {
@@ -401,8 +409,7 @@
           }
         ]
       }
-    },
-    "type": "object"
+    }
   },
   "annotations": {
     "readOnlyHint": true,

crates/calm-server/src/__toolsnaps__/callers.snap

@@ -269,7 +269,15 @@
         ]
       }
     },
+    "required": [
+      "state"
+    ],
+    "type": "object",
     "properties": {
+      "state": {
+        "description": "See `ToolOutcome::state`'s own doc comment -- same contract, plus\n`ambiguous`/`not_found` both map onto `needs_context` (the caller\nneeds one more piece of information: which candidate, or a\ndifferent query).",
+        "type": "string"
+      },
       "error": {
         "anyOf": [
           {
@@ -405,8 +413,7 @@
           }
         ]
       }
-    },
-    "type": "object"
+    }
   },
   "annotations": {
     "readOnlyHint": true,

crates/calm-server/src/__toolsnaps__/dependencies.snap

@@ -86,7 +86,15 @@
         ]
       }
     },
+    "required": [
+      "state"
+    ],
+    "type": "object",
     "properties": {
+      "state": {
+        "description": "Wave 11 (item 3, \"unified continuation state machine\"): present on\nevery response regardless of success/error -- `\"ready\"` on success,\nor one of `needs_context`/`needs_human_review`/`needs_verification`/\n`blocked` on error (see `classify_error_state`'s own doc comment).\nLets an agent dispatch on ONE top-level field instead of\nmaintaining its own per-`code` remediation table across ~70\ndistinct codes.",
+        "type": "string"
+      },
       "error": {
         "anyOf": [
           {
@@ -142,8 +150,7 @@
           }
         ]
       }
-    },
-    "type": "object"
+    }
   },
   "annotations": {
     "readOnlyHint": true,

crates/calm-server/src/__toolsnaps__/diff_impact.snap

@@ -160,7 +160,15 @@
         ]
       }
     },
+    "required": [
+      "state"
+    ],
+    "type": "object",
     "properties": {
+      "state": {
+        "description": "Wave 11 (item 3, \"unified continuation state machine\"): present on\nevery response regardless of success/error -- `\"ready\"` on success,\nor one of `needs_context`/`needs_human_review`/`needs_verification`/\n`blocked` on error (see `classify_error_state`'s own doc comment).\nLets an agent dispatch on ONE top-level field instead of\nmaintaining its own per-`code` remediation table across ~70\ndistinct codes.",
+        "type": "string"
+      },
       "error": {
         "anyOf": [
           {
@@ -214,8 +222,7 @@
           }
         ]
       }
-    },
-    "type": "object"
+    }
   },
   "annotations": {
     "readOnlyHint": true,

crates/calm-server/src/__toolsnaps__/edit_context.snap

@@ -415,7 +415,15 @@
         ]
       }
     },
+    "required": [
+      "state"
+    ],
+    "type": "object",
     "properties": {
+      "state": {
+        "description": "See `ToolOutcome::state`'s own doc comment -- same contract, plus\n`ambiguous`/`not_found` both map onto `needs_context` (the caller\nneeds one more piece of information: which candidate, or a\ndifferent query).",
+        "type": "string"
+      },
       "error": {
         "anyOf": [
           {
@@ -590,8 +598,7 @@
           }
         ]
       }
-    },
-    "type": "object"
+    }
   },
   "annotations": {
     "readOnlyHint": false,

crates/calm-server/src/__toolsnaps__/edit_lines.snap

@@ -215,7 +215,15 @@
         ]
       }
     },
+    "required": [
+      "state"
+    ],
+    "type": "object",
     "properties": {
+      "state": {
+        "description": "Wave 11 (item 3, \"unified continuation state machine\"): present on\nevery response regardless of success/error -- `\"ready\"` on success,\nor one of `needs_context`/`needs_human_review`/`needs_verification`/\n`blocked` on error (see `classify_error_state`'s own doc comment).\nLets an agent dispatch on ONE top-level field instead of\nmaintaining its own per-`code` remediation table across ~70\ndistinct codes.",
+        "type": "string"
+      },
       "error": {
         "anyOf": [
           {
@@ -288,8 +296,7 @@
           }
         ]
       }
-    },
-    "type": "object"
+    }
   },
   "annotations": {
     "readOnlyHint": false,

crates/calm-server/src/__toolsnaps__/edit_symbol.snap

@@ -293,7 +293,15 @@
         ]
       }
     },
+    "required": [
+      "state"
+    ],
+    "type": "object",
     "properties": {
+      "state": {
+        "description": "See `ToolOutcome::state`'s own doc comment -- same contract, plus\n`ambiguous`/`not_found` both map onto `needs_context` (the caller\nneeds one more piece of information: which candidate, or a\ndifferent query).",
+        "type": "string"
+      },
       "error": {
         "anyOf": [
           {
@@ -395,8 +403,7 @@
           }
         ]
       }
-    },
-    "type": "object"
+    }
   },
   "annotations": {
     "readOnlyHint": false,

crates/calm-server/src/__toolsnaps__/edit_transaction_status.snap

@@ -60,7 +60,14 @@
         ]
       }
     },
+    "required": [
+      "state"
+    ],
+    "type": "object",
     "properties": {
+      "state": {
+        "type": "string"
+      },
       "error": {
         "anyOf": [
           {
@@ -77,9 +84,6 @@
       "path": {
         "type": "string"
       },
-      "state": {
-        "type": "string"
-      },
       "replay_state": {
         "type": [
           "string",
@@ -102,8 +106,7 @@
           }
         ]
       }
-    },
-    "type": "object"
+    }
   },
   "annotations": {
     "readOnlyHint": true,

crates/calm-server/src/__toolsnaps__/file_overview.snap

@@ -102,7 +102,15 @@
         ]
       }
     },
+    "required": [
+      "state"
+    ],
+    "type": "object",
     "properties": {
+      "state": {
+        "description": "Wave 11 (item 3, \"unified continuation state machine\"): present on\nevery response regardless of success/error -- `\"ready\"` on success,\nor one of `needs_context`/`needs_human_review`/`needs_verification`/\n`blocked` on error (see `classify_error_state`'s own doc comment).\nLets an agent dispatch on ONE top-level field instead of\nmaintaining its own per-`code` remediation table across ~70\ndistinct codes.",
+        "type": "string"
+      },
       "error": {
         "anyOf": [
           {
@@ -147,8 +155,7 @@
           }
         ]
       }
-    },
-    "type": "object"
+    }
   },
   "annotations": {
     "readOnlyHint": true,