Hmbown/CodeWhale · error

Compaction summary response incomplete: provider stop reason

Error message

Compaction summary response incomplete: provider stop reason `{}`; the partial summary was not accepted.

What it means

The summary request finished with a provider stop reason classified as incomplete (e.g. length/max_tokens or content filter). Usage was already billed and recorded, but the fragment is not accepted as the session checkpoint (crates/tui/src/compaction.rs:1273); the message includes stop_reason_detail to distinguish causes.

Source

Thrown at crates/tui/src/compaction.rs:1273

            cost_scope,
            config.runtime_cost_owner.as_deref(),
            &format!(
                "compaction:dispatch:{}:response:{}",
                cost_route
                    .dispatched_at
                    .timestamp_nanos_opt()
                    .unwrap_or_default(),
                response.id
            ),
            &cost_route,
            &response.usage,
        );

        // Usage above is already billed; a provider-declared incomplete
        // summary must still fail rather than replace the session history
        // with a fragment.
        if crate::models::is_incomplete_stop_reason(response.stop_reason.as_deref()) {
            anyhow::bail!(
                "Compaction summary response incomplete: provider stop reason `{}`; the partial summary was not accepted.",
                crate::models::stop_reason_detail(response.stop_reason.as_deref())
            );
        }

        let summary = response
            .content
            .iter()
            .filter_map(|block| match block {
                ContentBlock::Text { text, .. } => Some(text.clone()),
                _ => None,
            })
            .collect::<Vec<_>>()
            .join("\n");

        if let Err(error) = validate_compaction_summary(&summary) {
            if quality_retry_used {
                return Err(error.context(

View on GitHub (pinned to 8880682c63)

Solutions

1. Raise the compaction output-token budget in config
2. If the stop reason is content_filter, compact earlier so less sensitive material accumulates, or change models
3. Retry once - truncation can be borderline and stochastic on similar lengths
4. Compact more frequently so each individual summary is smaller

// Before compacting, budget output tokens against session size
let out_tokens = cfg.compaction_max_output_tokens();
ensure!(out_tokens as u64 >= estimated_summary_tokens(&messages), "output budget too small; will truncate");

fn is_incomplete_summary(msg: &str) -> bool {
    msg.contains("Compaction summary response incomplete")
}

// Read stop_reason_detail from the message: raise budget for `length`, change model for `content_filter`
if is_incomplete_summary(&e.to_string()) {
    if e.to_string().contains("length") || e.to_string().contains("max") {
        raise_output_budget_and_retry();
    } else {
        switch_compaction_model();
    }
}

When it happens

Trigger: Compaction output-token budget smaller than the summary a long session needs; provider-side length caps; content-filter truncation of the summary itself.

Common situations: Very long sessions compacted in one shot; default max output tokens too low for the summary model; strict providers that truncate aggressively.

Related errors

• Compaction summary response was unusable: no text was return
• Compaction summary response was unusable: only whitespace or
• Model response incomplete: provider stop reason `{}`; the pa
• Model response incomplete: provider stop reason `{}`; the JS
• Codewhale terminal receipt did not use provider openai

AI-assisted analysis of Hmbown/CodeWhale@8880682c63 (2026-08-16). Data as JSON: /api/errors/54d79b1236b231f7. Report an issue: GitHub.