| 1 | //! Deterministic integrity repair for persisted model-visible tool history. |
| 2 | //! |
| 3 | //! Session JSON predates a durable per-call journal, so process exit can leave |
| 4 | //! a `tool_use` without its terminal `tool_result`. Provider APIs reject that |
| 5 | //! shape. This module repairs the existing message format without changing its |
| 6 | //! schema and returns a bounded diagnostic receipt for every mutation. |
| 7 | |
| 8 | use std::collections::{HashMap, HashSet}; |
| 9 | |
| 10 | use crate::models::{ContentBlock, Message}; |
| 11 | |
| 12 | const CRASH_REPAIR_CONTENT: &str = |
| 13 | "Tool call interrupted by process exit; terminal status: crashed_and_repaired."; |
| 14 | |
| 15 | #[derive(Debug, Clone, Default, PartialEq, Eq)] |
| 16 | pub(crate) struct ToolRepairReceipt { |
| 17 | pub(crate) repaired_call_ids: Vec<String>, |
| 18 | pub(crate) duplicate_result_ids: Vec<String>, |
| 19 | pub(crate) orphan_result_ids: Vec<String>, |
| 20 | } |
| 21 | |
| 22 | impl ToolRepairReceipt { |
| 23 | #[must_use] |
| 24 | pub(crate) fn is_empty(&self) -> bool { |
| 25 | self.repaired_call_ids.is_empty() |
| 26 | && self.duplicate_result_ids.is_empty() |
| 27 | && self.orphan_result_ids.is_empty() |
| 28 | } |
| 29 | |
| 30 | fn visible_message(&self) -> String { |
| 31 | format!( |
| 32 | "[tool_history_repair] Repaired {} crashed tool call(s); quarantined {} duplicate and {} orphan terminal result(s).", |
| 33 | self.repaired_call_ids.len(), |
| 34 | self.duplicate_result_ids.len(), |
| 35 | self.orphan_result_ids.len(), |
| 36 | ) |
| 37 | } |
| 38 | } |
| 39 | |
| 40 | /// Repair tool-use/result integrity in place. |
| 41 | /// |
| 42 | /// The first terminal result after a known call and before the next assistant |
| 43 | /// turn is retained. Results that precede their call, arrive after a later |
| 44 | /// assistant turn, reference no call, or repeat a retained result are |
| 45 | /// quarantined by removing them from model-visible history. Every dangling |
| 46 | /// call receives a synthetic error result directly after its assistant call |
| 47 | /// message. A visible system receipt makes the repair apparent after resume. |
| 48 | pub(crate) fn repair_tool_call_pairs(messages: &mut Vec<Message>) -> ToolRepairReceipt { |
| 49 | repair_tool_call_pairs_inner(messages, true) |
| 50 | } |
| 51 | |
| 52 | /// Repair an ephemeral provider request without appending a trailing receipt. |
| 53 | /// |
| 54 | /// Anthropic-style APIs interpret a final assistant message as a completion |
| 55 | /// prefill. Pair repair must therefore leave the synthetic user tool result as |
| 56 | /// the request tail; the durable session-facing path owns the visible receipt. |
| 57 | pub(crate) fn repair_tool_call_pairs_for_provider( |
| 58 | messages: &mut Vec<Message>, |
| 59 | ) -> ToolRepairReceipt { |
| 60 | repair_tool_call_pairs_inner(messages, false) |
| 61 | } |
| 62 | |
| 63 | fn repair_tool_call_pairs_inner( |
| 64 | messages: &mut Vec<Message>, |
| 65 | append_visible_receipt: bool, |
| 66 | ) -> ToolRepairReceipt { |
| 67 | let mut call_message_by_id = HashMap::new(); |
| 68 | let mut call_ids_in_order = Vec::new(); |
| 69 | for (message_index, message) in messages.iter().enumerate() { |
| 70 | if message.role != "assistant" && message.role != crate::models::INTERRUPTED_ASSISTANT_ROLE |
| 71 | { |
| 72 | continue; |
| 73 | } |
| 74 | for block in &message.content { |
| 75 | if let ContentBlock::ToolUse { id, .. } = block |
| 76 | && !call_message_by_id.contains_key(id) |
| 77 | { |
| 78 | call_message_by_id.insert(id.clone(), message_index); |
| 79 | call_ids_in_order.push(id.clone()); |
| 80 | } |
| 81 | } |
| 82 | } |
| 83 | |
| 84 | let mut retained_results = HashSet::new(); |
| 85 | let mut duplicate_result_ids = Vec::new(); |
| 86 | let mut orphan_result_ids = Vec::new(); |
| 87 | let mut keep_results = HashSet::new(); |
| 88 | let mut result_ordinal = 0usize; |
| 89 | let mut latest_assistant_message = None; |
| 90 | |
| 91 | for (message_index, message) in messages.iter().enumerate() { |
| 92 | if message.role == "assistant" || message.role == crate::models::INTERRUPTED_ASSISTANT_ROLE |
| 93 | { |
| 94 | latest_assistant_message = Some(message_index); |
| 95 | } |
| 96 | for block in &message.content { |
| 97 | let ContentBlock::ToolResult { tool_use_id, .. } = block else { |
| 98 | continue; |
| 99 | }; |
| 100 | let ordinal = result_ordinal; |
| 101 | result_ordinal = result_ordinal.saturating_add(1); |
| 102 | |
| 103 | let follows_known_call = |
| 104 | call_message_by_id |
| 105 | .get(tool_use_id) |
| 106 | .is_some_and(|call_index| { |
| 107 | *call_index < message_index && latest_assistant_message == Some(*call_index) |
| 108 | }); |
| 109 | if !follows_known_call { |
| 110 | orphan_result_ids.push(tool_use_id.clone()); |
| 111 | } else if !retained_results.insert(tool_use_id.clone()) { |
| 112 | duplicate_result_ids.push(tool_use_id.clone()); |
| 113 | } else { |
| 114 | keep_results.insert(ordinal); |
| 115 | } |
| 116 | } |
| 117 | } |
| 118 | |
| 119 | let repaired_call_ids: Vec<_> = call_ids_in_order |
| 120 | .into_iter() |
| 121 | .filter(|id| !retained_results.contains(id)) |
| 122 | .collect(); |
| 123 | let repaired_set: HashSet<_> = repaired_call_ids.iter().cloned().collect(); |
| 124 | |
| 125 | let receipt = ToolRepairReceipt { |
| 126 | repaired_call_ids, |
| 127 | duplicate_result_ids, |
| 128 | orphan_result_ids, |
| 129 | }; |
| 130 | if receipt.is_empty() { |
| 131 | return receipt; |
| 132 | } |
| 133 | |
| 134 | let original = std::mem::take(messages); |
| 135 | let mut rebuilt = Vec::with_capacity( |
| 136 | original |
| 137 | .len() |
| 138 | .saturating_add(receipt.repaired_call_ids.len()), |
| 139 | ); |
| 140 | let mut seen_result_ordinal = 0usize; |
| 141 | |
| 142 | for message in original { |
| 143 | let missing_after_message: Vec<_> = if message.role == "assistant" |
| 144 | || message.role == crate::models::INTERRUPTED_ASSISTANT_ROLE |
| 145 | { |
| 146 | message |
| 147 | .content |
| 148 | .iter() |
| 149 | .filter_map(|block| match block { |
| 150 | ContentBlock::ToolUse { id, .. } if repaired_set.contains(id) => { |
| 151 | Some(id.clone()) |
| 152 | } |
| 153 | _ => None, |
| 154 | }) |
| 155 | .collect() |
| 156 | } else { |
| 157 | Vec::new() |
| 158 | }; |
| 159 | |
| 160 | let mut filtered = message; |
| 161 | filtered.content.retain(|block| { |
| 162 | if matches!(block, ContentBlock::ToolResult { .. }) { |
| 163 | let keep = keep_results.contains(&seen_result_ordinal); |
| 164 | seen_result_ordinal = seen_result_ordinal.saturating_add(1); |
| 165 | keep |
| 166 | } else { |
| 167 | true |
| 168 | } |
| 169 | }); |
| 170 | if !filtered.content.is_empty() { |
| 171 | rebuilt.push(filtered); |
| 172 | } |
| 173 | |
| 174 | if !missing_after_message.is_empty() { |
| 175 | rebuilt.push(Message { |
| 176 | role: "user".to_string(), |
| 177 | content: missing_after_message |
| 178 | .into_iter() |
| 179 | .map(|tool_use_id| ContentBlock::ToolResult { |
| 180 | tool_use_id, |
| 181 | content: CRASH_REPAIR_CONTENT.to_string(), |
| 182 | is_error: Some(true), |
| 183 | content_blocks: None, |
| 184 | }) |
| 185 | .collect(), |
| 186 | }); |
| 187 | } |
| 188 | } |
| 189 | |
| 190 | if append_visible_receipt { |
| 191 | rebuilt.push(Message { |
| 192 | role: "assistant".to_string(), |
| 193 | content: vec![ContentBlock::Text { |
| 194 | text: receipt.visible_message(), |
| 195 | cache_control: None, |
| 196 | }], |
| 197 | }); |
| 198 | } |
| 199 | *messages = rebuilt; |
| 200 | receipt |
| 201 | } |
| 202 | |
| 203 | #[cfg(test)] |
| 204 | mod tests { |
| 205 | use serde_json::json; |
| 206 | |
| 207 | use super::*; |
| 208 | |
| 209 | fn tool_call(id: &str) -> Message { |
| 210 | Message { |
| 211 | role: "assistant".to_string(), |
| 212 | content: vec![ContentBlock::ToolUse { |
| 213 | id: id.to_string(), |
| 214 | name: "read_file".to_string(), |
| 215 | input: json!({"path": "README.md"}), |
| 216 | caller: None, |
| 217 | }], |
| 218 | } |
| 219 | } |
| 220 | |
| 221 | fn tool_result(id: &str, content: &str) -> Message { |
| 222 | Message { |
| 223 | role: "user".to_string(), |
| 224 | content: vec![ContentBlock::ToolResult { |
| 225 | tool_use_id: id.to_string(), |
| 226 | content: content.to_string(), |
| 227 | is_error: None, |
| 228 | content_blocks: None, |
| 229 | }], |
| 230 | } |
| 231 | } |
| 232 | |
| 233 | fn text(role: &str, content: &str) -> Message { |
| 234 | Message { |
| 235 | role: role.to_string(), |
| 236 | content: vec![ContentBlock::Text { |
| 237 | text: content.to_string(), |
| 238 | cache_control: None, |
| 239 | }], |
| 240 | } |
| 241 | } |
| 242 | |
| 243 | #[test] |
| 244 | fn well_formed_history_is_unchanged() { |
| 245 | let mut messages = vec![tool_call("call-1"), tool_result("call-1", "ok")]; |
| 246 | let before = messages.clone(); |
| 247 | |
| 248 | let receipt = repair_tool_call_pairs(&mut messages); |
| 249 | |
| 250 | assert!(receipt.is_empty()); |
| 251 | assert_eq!(messages, before); |
| 252 | } |
| 253 | |
| 254 | #[test] |
| 255 | fn repairs_dangling_calls_beside_their_assistant_message() { |
| 256 | let mut messages = vec![ |
| 257 | tool_call("call-1"), |
| 258 | text("assistant", "later assistant text"), |
| 259 | ]; |
| 260 | |
| 261 | let receipt = repair_tool_call_pairs(&mut messages); |
| 262 | |
| 263 | assert_eq!(receipt.repaired_call_ids, vec!["call-1"]); |
| 264 | assert_eq!(messages[1].role, "user"); |
| 265 | assert!(matches!( |
| 266 | &messages[1].content[0], |
| 267 | ContentBlock::ToolResult { |
| 268 | tool_use_id, |
| 269 | content, |
| 270 | is_error: Some(true), |
| 271 | .. |
| 272 | } if tool_use_id == "call-1" && content.contains("crashed_and_repaired") |
| 273 | )); |
| 274 | assert_eq!(messages.last().expect("receipt").role, "assistant"); |
| 275 | } |
| 276 | |
| 277 | #[test] |
| 278 | fn provider_repair_never_appends_an_assistant_prefill_receipt() { |
| 279 | let mut messages = vec![tool_call("call-1")]; |
| 280 | |
| 281 | let receipt = repair_tool_call_pairs_for_provider(&mut messages); |
| 282 | |
| 283 | assert_eq!(receipt.repaired_call_ids, vec!["call-1"]); |
| 284 | assert_eq!(messages.last().expect("synthetic result").role, "user"); |
| 285 | assert!(!messages.iter().any(|message| { |
| 286 | message.content.iter().any(|block| { |
| 287 | matches!( |
| 288 | block, |
| 289 | ContentBlock::Text { text, .. } |
| 290 | if text.contains("[tool_history_repair]") |
| 291 | ) |
| 292 | }) |
| 293 | })); |
| 294 | } |
| 295 | |
| 296 | #[test] |
| 297 | fn quarantines_orphan_and_duplicate_results_without_losing_other_blocks() { |
| 298 | let mut mixed_result = tool_result("call-1", "duplicate"); |
| 299 | mixed_result.content.push(ContentBlock::Text { |
| 300 | text: "keep me".to_string(), |
| 301 | cache_control: None, |
| 302 | }); |
| 303 | let mut messages = vec![ |
| 304 | tool_result("orphan", "bad"), |
| 305 | tool_call("call-1"), |
| 306 | tool_result("call-1", "first"), |
| 307 | mixed_result, |
| 308 | ]; |
| 309 | |
| 310 | let receipt = repair_tool_call_pairs(&mut messages); |
| 311 | |
| 312 | assert_eq!(receipt.orphan_result_ids, vec!["orphan"]); |
| 313 | assert_eq!(receipt.duplicate_result_ids, vec!["call-1"]); |
| 314 | let result_contents: Vec<_> = messages |
| 315 | .iter() |
| 316 | .flat_map(|message| &message.content) |
| 317 | .filter_map(|block| match block { |
| 318 | ContentBlock::ToolResult { content, .. } => Some(content.as_str()), |
| 319 | _ => None, |
| 320 | }) |
| 321 | .collect(); |
| 322 | assert_eq!(result_contents, vec!["first"]); |
| 323 | assert!(messages.iter().any(|message| { |
| 324 | message |
| 325 | .content |
| 326 | .iter() |
| 327 | .any(|block| matches!(block, ContentBlock::Text { text, .. } if text == "keep me")) |
| 328 | })); |
| 329 | } |
| 330 | |
| 331 | #[test] |
| 332 | fn repair_is_idempotent() { |
| 333 | let mut messages = vec![tool_call("call-1"), tool_result("orphan", "bad")]; |
| 334 | |
| 335 | let first = repair_tool_call_pairs(&mut messages); |
| 336 | let after_first = messages.clone(); |
| 337 | let second = repair_tool_call_pairs(&mut messages); |
| 338 | |
| 339 | assert!(!first.is_empty()); |
| 340 | assert!(second.is_empty()); |
| 341 | assert_eq!(messages, after_first); |
| 342 | } |
| 343 | |
| 344 | #[test] |
| 345 | fn result_preceding_its_call_is_orphaned_and_call_is_repaired() { |
| 346 | let mut messages = vec![tool_result("call-1", "too early"), tool_call("call-1")]; |
| 347 | |
| 348 | let receipt = repair_tool_call_pairs(&mut messages); |
| 349 | |
| 350 | assert_eq!(receipt.orphan_result_ids, vec!["call-1"]); |
| 351 | assert_eq!(receipt.repaired_call_ids, vec!["call-1"]); |
| 352 | assert!(messages.iter().any(|message| { |
| 353 | message.content.iter().any(|block| { |
| 354 | matches!( |
| 355 | block, |
| 356 | ContentBlock::ToolResult { content, .. } |
| 357 | if content.contains("crashed_and_repaired") |
| 358 | ) |
| 359 | }) |
| 360 | })); |
| 361 | } |
| 362 | |
| 363 | #[test] |
| 364 | fn result_after_a_later_assistant_turn_is_quarantined_as_too_late() { |
| 365 | let mut messages = vec![ |
| 366 | tool_call("call-1"), |
| 367 | text("assistant", "a later model turn"), |
| 368 | tool_result("call-1", "too late"), |
| 369 | ]; |
| 370 | |
| 371 | let receipt = repair_tool_call_pairs(&mut messages); |
| 372 | |
| 373 | assert_eq!(receipt.orphan_result_ids, vec!["call-1"]); |
| 374 | assert_eq!(receipt.repaired_call_ids, vec!["call-1"]); |
| 375 | assert!(!messages.iter().any(|message| { |
| 376 | message.content.iter().any(|block| { |
| 377 | matches!(block, ContentBlock::ToolResult { content, .. } if content == "too late") |
| 378 | }) |
| 379 | })); |
| 380 | } |
| 381 | } |
| 382 |