mirror of
https://github.com/instructkr/claw-code.git
synced 2026-11-18 00:00:00 +08:00
fix(runtime): bound tool-result payloads to the actual context window
Two related gaps let a single agentic turn exceed a small context window, with no recovery path once it happened. 1. read_file's character ceiling was a hardcoded 64 KB (~16k tokens), sized for a 200k window. Two reads in one turn reach ~48k tokens, which a 32k backend rejects outright. Make it configurable via CLAW_MAX_READ_CHARS, defaulting to the existing 64 KB so behavior is unchanged unless set. Reads remain paged, not refused: callers still get truncated/nextOffset. 2. Compaction could not help, because it works at message granularity and an agentic turn is five messages (user / assistant+tool_use / tool_result / assistant+tool_use / tool_result). With preserve_recent_messages: 4 the only droppable message is the short user prompt, so the two oversized tool results were preserved verbatim by design. Trim preserved tool-result payloads over 16 KB to a head/tail excerpt with an explicit "[... N characters elided ...]" marker, so the model knows its view is partial and can re-read. Small payloads pass through untouched. Also apply trimming on the early-return path: a session can exceed the window while holding fewer messages than preserve_recent_messages, where should_compact is false and message-granular compaction has no move to make. compact_now no longer gates on removed_message_count alone, which would have computed the smaller session and then discarded it. Also fixes a pre-existing flaky test: repl_executes_python_code resolves the python runtime through PATH, which the PowerShell tests blank while they run. It now takes the same env_lock, so it no longer intermittently fails with "python runtime not found" on a machine that has python. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013bVvRYn7CQKR6hT2j1L6Cs (cherry picked from commit b0d4d41e8132f8dc75717b1f74554d649639de10)
This commit is contained in:
1 parent
910ac4e326
commit
fa0d53a969
4 files changed
+378
-11
No files matched your search
@@ -625,7 +625,7 @@ where
|
||||
self.compact_now()
|
||||
}
|
||||
|
||||
/// Compacts the session unconditionally, returning `None` when there was nothing to remove.
|
||||
/// Compacts the session unconditionally, returning `None` when nothing actually changed.
|
||||
fn compact_now(&mut self) -> Option<AutoCompactionEvent> {
|
||||
let result = compact_session(
|
||||
&self.session,
|
||||
@@ -635,7 +635,12 @@ where
|
||||
},
|
||||
);
|
||||
|
||||
if result.removed_message_count == 0 {
|
||||
// `removed_message_count` alone is the wrong test for "did anything happen": compaction
|
||||
// also trims oversized tool-result payloads in messages it must PRESERVE, and on a
|
||||
// session too short to have anything droppable that trimming is the entire result.
|
||||
// Gating on the count would compute the smaller session and then throw it away.
|
||||
let shrank = result.compacted_session.messages != self.session.messages;
|
||||
if result.removed_message_count == 0 && !shrank {
|
||||
return None;
|
||||
}
|
||||
|
||||
|
||||
Reference in new issue
Block a user