Skip to content

fix: Fix stale context usage after compaction + usage calculation #1548

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
charlesvien wants to merge 2 commits into
base: 04-07-show_both_local_and_remote_repo_name_when_they_differ
Choose a base branch
from
+16 −0
Open

fix: Fix stale context usage after compaction + usage calculation #1548

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
20b1ebb
Include output_tokens in context usage calculation
charlesvien Apr 8, 2026
9e310fa
Reset context usage to zero on compaction
charlesvien Apr 8, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 16 additions & 0 deletions packages/agent/src/adapters/claude/claude-agent.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -371,8 +371,18 @@ export class ClaudeAcpAgent extends BaseAcpAgent {
switch (message.type) {
case "system":
if (message.subtype === "compact_boundary") {
// Send used:0 immediately so the client doesn't keep showing
// the stale pre-compaction context size until the next turn.
lastAssistantTotalUsage = 0;
promptReplayed = true;
await this.client.sessionUpdate({
sessionId: params.sessionId,
update: {
sessionUpdate: "usage_update",
used: 0,
size: lastContextWindowSize,
},
});
}
if (message.subtype === "local_command_output") {
promptReplayed = true;
Expand Down Expand Up @@ -530,6 +540,11 @@ export class ClaudeAcpAgent extends BaseAcpAgent {
}

// Store latest assistant usage (excluding subagents)
// Sum all token types as a proxy for post-turn context occupancy:
// current turn's output will become next turn's input.
// Note: per the Anthropic API, input_tokens excludes cache tokens —
// cache_read and cache_creation are reported separately, so summing
// all four fields is not double-counting.
if (
"usage" in message.message &&
message.parent_tool_use_id === null
Expand All @@ -544,6 +559,7 @@ export class ClaudeAcpAgent extends BaseAcpAgent {
};
lastAssistantTotalUsage =
usage.input_tokens +
usage.output_tokens +
usage.cache_read_input_tokens +
usage.cache_creation_input_tokens;

Expand Down
Loading