feat(engine): context-limit handoff replaces routine compaction (closes #664) by merchloubna70-dot · Pull Request #667 · Hmbown/CodeWhale

merchloubna70-dot · 2026-05-05T01:05:38Z

Three-stage awareness 70/80/90%. Closes #664

All system prompts were English-only, causing DeepSeek V4 to reason and respond in English even when users wrote in Chinese or other languages. Add a Language Mirror section to base.md and base.txt that instructs the model to detect the user's primary language and use it for both reasoning (thinking tokens) and the final reply.

base.txt is not referenced via include_str! in prompts.rs. Only base.md is loaded (BASE_PROMPT). Remove the redundant change to base.txt as noted by Gemini Code Assist review.

gemini-code-assist · 2026-05-05T01:05:41Z

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

Hmbown · 2026-05-05T05:27:06Z

Review notes from v0.8.12 integration

Thanks for this PR. I've reviewed the code as part of merging it into the v0.8.12 branch. One finding:

`prefer_handoff` is dead code

The prefer_handoff: Option<bool> field was added to CompactionConfig, but zero code paths read it. There's no handoff-writing function, no trigger in should_compact or the engine turn loop, nothing that consults this field. Setting prefer_handoff = true in config would have no effect.

The idea itself is solid — when context pressure builds, write a .deepseek/handoff.md instead of paying for an LLM compaction call. That preserves the V4 prefix cache and avoids the compaction cost. But the plumbing behind the config knob was never built.

Recommendation: Either implement the handoff path (file writer + trigger in the compaction flow + context-clear mechanism) in a follow-up PR, or remove the field until that work is done. Shipping a config key that does nothing is misleading to users.

The rest of the PR (the auto_floor_tokens → prefer_handoff coexistence in the struct) merged cleanly.

Hmbown · 2026-05-05T05:55:29Z

v0.8.12: removed dead `prefer_handoff` field

Per the earlier review finding, the prefer_handoff field has been removed from CompactionConfig. The field was defined but zero code paths consulted it — it was a config knob wired to nothing.

The handoff idea itself (write .deepseek/handoff.md on context pressure instead of paying for LLM compaction) is a good one. When the implementation exists, the field can be re-added with the plumbing behind it.

Commit: c0e27485a

…follow-up) `prefer_handoff: Option<bool>` was added to CompactionConfig but never read — setting it had no effect. When `Some(true)`, skip automatic LLM compaction so the handoff threshold system in handoff.rs can instruct the model to write `.deepseek/handoff.md` instead. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

merchloubna70-dot · 2026-05-05T06:26:52Z

Fixed — should_compact now returns false when prefer_handoff == Some(true), directing context pressure to the handoff threshold system (handoff.rs) instead of triggering LLM compaction.

Updated per Hunter's v0.8.12 review: `prefer_handoff: Option<bool>` was removed from CompactionConfig rather than being partially wired. A half-wired field that prevents LLM compaction without writing a handoff file is worse than no compaction — context would accumulate unchecked. The field can be re-introduced once the full handoff implementation (file writer + context-clear) is ready. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

merchloubna70-dot · 2026-05-05T06:39:41Z

Updated to align with your v0.8.12 approach — removed prefer_handoff from CompactionConfig entirely rather than wiring it into should_compact. Agreed that a half-wired knob (skip LLM compaction but write no handoff file) is worse than no knob. The field can come back once the full implementation (file writer + context-clear) is ready.

The match guard at tui/ui.rs:1603 used `&& let Some(...) = ...` inside an `if` guard, which requires the `if_let_guard` nightly feature on Rust < 1.94. Reported by an external user attempting `cargo install deepseek-tui` on stable rustc — it failed with E0658. Rewrite as a plain match guard with a nested `if let` inside the arm body so the language-picker hotkeys compile on every supported rustc. Workspace also now declares `rust-version = "1.88"` to match the codebase's actual reliance on `let_chains` in if/while conditions, so users on too-old toolchains see a clear cargo error instead of a confusing rustc one. `AGENTS.md` and `CLAUDE.md` gain a "stable Rust only" section documenting the trap and how to rewrite around it. Also annotate the deferred `TuiPrefs` (#657) and `handoff::THRESHOLDS` (#667) APIs with `#[allow(dead_code)]` so CI's `-D warnings` flag stays green while the call sites are staged for v0.8.13. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

47 fmt drifts had accumulated from the squash-merged community PRs on this branch (#653, #654, #655, #645, #658, #668, #659, #661, #660, #667, #656). Pure formatting — no behavioural changes — applied via `cargo fmt --all` to satisfy CI's `cargo fmt --all -- --check` gate. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…mbown#664) THRESHOLDS and threshold_message() are reserved for the planned context-ratio tracking + system-prompt injection. Add #[allow(dead_code)] and doc comments so the compiler does not warn and reviewers understand these are future-wiring, not stale leftovers. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

merchloubna70-dot · 2026-05-05T07:19:08Z

Added #[allow(dead_code)] with doc comments to THRESHOLDS and threshold_message() in handoff.rs — these are reserved for the planned context-ratio tracking + injection, not stale leftovers.

…wn#667)

…terminal_colors (Hmbown#671) Neither field had any code path that read it. Shipping config knobs that do nothing trains users to mistrust config. Remove until the implementation exists.

…wn#667)

…terminal_colors (Hmbown#671) Neither field had any code path that read it. Shipping config knobs that do nothing trains users to mistrust config. Remove until the implementation exists.

The match guard at tui/ui.rs:1603 used `&& let Some(...) = ...` inside an `if` guard, which requires the `if_let_guard` nightly feature on Rust < 1.94. Reported by an external user attempting `cargo install deepseek-tui` on stable rustc — it failed with E0658. Rewrite as a plain match guard with a nested `if let` inside the arm body so the language-picker hotkeys compile on every supported rustc. Workspace also now declares `rust-version = "1.88"` to match the codebase's actual reliance on `let_chains` in if/while conditions, so users on too-old toolchains see a clear cargo error instead of a confusing rustc one. `AGENTS.md` and `CLAUDE.md` gain a "stable Rust only" section documenting the trap and how to rewrite around it. Also annotate the deferred `TuiPrefs` (Hmbown#657) and `handoff::THRESHOLDS` (Hmbown#667) APIs with `#[allow(dead_code)]` so CI's `-D warnings` flag stays green while the call sites are staged for v0.8.13. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

47 fmt drifts had accumulated from the squash-merged community PRs on this branch (Hmbown#653, Hmbown#654, Hmbown#655, Hmbown#645, Hmbown#658, Hmbown#668, Hmbown#659, Hmbown#661, Hmbown#660, Hmbown#667, Hmbown#656). Pure formatting — no behavioural changes — applied via `cargo fmt --all` to satisfy CI's `cargo fmt --all -- --check` gate. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Hmbown · 2026-05-07T03:03:30Z

Shipped on main in 2cdf4a256 (feat(engine): context-limit handoff replaces routine compaction, #667), with cleanup follow-up c0e27485a removing the unused prefer_handoff config field. Auto-close didn't fire — closing manually.

macworkers and others added 3 commits May 4, 2026 12:39

fix: remove base.txt change — file is not loaded by runtime

3cff070

base.txt is not referenced via include_str! in prompts.rs. Only base.md is loaded (BASE_PROMPT). Remove the redundant change to base.txt as noted by Gemini Code Assist review.

feat(engine): context-limit handoff thresholds (closes Hmbown#664)

401b563

Hmbown mentioned this pull request May 5, 2026

release: v0.8.12 — 20 community PRs + cargo-install fix + workspace-bleed hotfix #704

Merged

7 tasks

kibuniverse pushed a commit to kibuniverse/DeepSeek-TUI that referenced this pull request May 5, 2026

feat(engine): context-limit handoff replaces routine compaction (Hmbo…

2cdf4a2

…wn#667)

MMMarcinho pushed a commit to MMMarcinho/DeepSeek-TUI that referenced this pull request May 6, 2026

feat(engine): context-limit handoff replaces routine compaction (Hmbo…

e584c63

…wn#667)

Hmbown closed this May 7, 2026

linzhiqin2003 mentioned this pull request May 10, 2026

chore: remove orphaned handoff.rs (context-limit threshold table) #1387

Closed

3 tasks

zhnnky329 mentioned this pull request May 14, 2026

perf(cache): snapshot system prompt at session creation, wire warmup, and share prefix across sub-agents #1640

Closed

This was referenced Jun 12, 2026

v0.8.60: Make headless subagent lanes inspectable through Whaleflow trace handles #3180

Closed

v0.8.60: Group background tasks into workflows with phase summaries and collapsed Bash runs #3082

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(engine): context-limit handoff replaces routine compaction (closes #664)#667

feat(engine): context-limit handoff replaces routine compaction (closes #664)#667
merchloubna70-dot wants to merge 6 commits into
Hmbown:mainfrom
merchloubna70-dot:feat/context-handoff-664

merchloubna70-dot commented May 5, 2026

Uh oh!

gemini-code-assist Bot commented May 5, 2026

Uh oh!

Hmbown commented May 5, 2026

Uh oh!

Hmbown commented May 5, 2026

Uh oh!

merchloubna70-dot commented May 5, 2026

Uh oh!

merchloubna70-dot commented May 5, 2026

Uh oh!

merchloubna70-dot commented May 5, 2026

Uh oh!

Hmbown commented May 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

merchloubna70-dot commented May 5, 2026

Uh oh!

gemini-code-assist Bot commented May 5, 2026

Uh oh!

Hmbown commented May 5, 2026

Review notes from v0.8.12 integration

prefer_handoff is dead code

Uh oh!

Hmbown commented May 5, 2026

v0.8.12: removed dead prefer_handoff field

Uh oh!

merchloubna70-dot commented May 5, 2026

Uh oh!

merchloubna70-dot commented May 5, 2026

Uh oh!

merchloubna70-dot commented May 5, 2026

Uh oh!

Hmbown commented May 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

`prefer_handoff` is dead code

v0.8.12: removed dead `prefer_handoff` field