dsh-force-compact

Aggressive, local-first context compaction for DeepSeek Harness agents.

A DSH Cordis function plugin that keeps the agent's working context lean by design: serve Qwen3.8‑27B on a self-hosted llama.cpp with a modest context, and the plugin shrinks the conversation itself — a large-window feel with no API cost and no data egress.

中文

Why

• Self-hosted inference — the agent talks to a local OpenAI-compatible llama.cpp server through the standard DeepSeek adapter; no separate adapter needed.
• Low context, high signal — instead of fighting a small cap, the plugin shrinks the conversation, so the agent reasons over a tight prompt while keeping deep memory in the compressed head.
• Think off for compactions, passthrough everywhere else — disableThinking: true (default) turns thinking off on this plugin's own compaction summarization call only; every other model request rides the machine's configuration unchanged.
• Private & free — no per-token billing, no egress.

What it does

Two compaction engines coexist behind one facade (resolveCompaction), transparent to callers:

No toggling — official wins when reachable, builtin takes over otherwise.

Trigger points

• Per-request guard (agent/pre-step) — reads the session's projected context tokens (the exact number the harness renders bottom-right). At autoThresholdTokens it rejects the outgoing request and compacts the head instead, retaining the latest retainLatestTokens verbatim. Below the threshold the request proceeds.
• Turn end / idle (agent/status → idle) — when the agent quiesces, optionally compacts via compactNow (gate: turnEndForceCompactionEnabled).
• Manual /force-compact — immediate compactNow when idle; when busy it queues a process-local flag consumed at the next model step. Loaded lazily — see "Command availability" under Install.
• session/flush — the awaited durability checkpoint.

Every path funnels into the single "compaction result landed in the session" boundary — the same point where the LiveUI signal fires.

Decision key is projectedTokens (provider-anchored, same figure as the UI corner), so the plugin never drifts from what you see; the threshold-aware shrink gate skips summarizer calls that provably cannot pull the session below the threshold (kills the low-threshold dead loop).

The builtin transaction bills shadowedTokenCount from the same tokenMeter.measure per-node prices the official engine uses, so the meter's collapse protocol settles the drop correctly — the bottom-right counter goes down after compaction.

Thinking control: scoped to compactions

Since the 2026-08 semantics revision, disableThinking controls one thing: whether this plugin's own summarization call (engine/builtin.js → engine/summarizer.js → ctx.llm.stream) carries reasoningEffort:'off'. Everything else is untouched:

When the target is a llama.cpp / OpenAI-compatible endpoint, the thinking: { type: 'disabled' } field the adapter emits is silently ignored there — so the summarizer ALSO stamps the llama.cpp-native top-level reasoning_effort: "none", gated on the exact same condition. One options object carries BOTH fields:

Each family tolerates-and-ignores the foreign key, so emitting both is harmless. The field is stamped in src/engine/summarizer.js (immediately before llm.stream(options)), NOT in the llm/stream waterfall — a prior draft injected there but proved ineffective structurally (middle-layer returns are discarded; in-place seed mutation crashes the host); see the src/hooks/wire-rewrite.js module header for the write-up. That hook now serves only the LiveUI watermark role.

Need thinking off on business calls too? Set your provider's reasoningEffort at the request-header level — the plugin deliberately stays out of that decision.

Observability: per-attempt audit lines

Every summarization attempt logs two lines (visible at the default debug: true) — the durable proof of the scoping decision and its wire fields, without capturing traffic:

[force-compact] <sessionId>: compaction thinking-policy — settings.disableThinking=true → extra.reasoningEffort='off' (this summarization call carries thinking-OFF)
[force-compact] <sessionId>: summarization wire-fields → <provider>/<model>: reasoningEffort='off' + reasoning_effort="none" (llama.cpp-native wire field)

• Line 1 (engine/builtin.js) records where disableThinking is read and routed into the call options; with the setting off it records machine default.
• Line 2 (engine/summarizer.js) records both wire fields exactly as they leave the options object, plus resolved provider/model; unstamped fields are labeled (absent…).

Empirically grounded: probed against a local llama.cpp endpoint, a baseline request returned populated reasoning_content (the model thinks by default), while the same request with top-level reasoning_effort:"none" returned none at all — the field genuinely disables thinking there, and business calls (which omit it) keep thinking.

LiveUI status

A tiny host→client messenger (the liveUi settings field mirrored live to the browser) pins a badge beside the turn:

• Red "compressing" — just before a compaction commits (screen text is Chinese);
• Green "done" — the instant a compaction lands; 3 s later a fresh random working line takes over;
• Blue "working" — otherwise a rotating playful one-liner;
• Cleared at conversation end — when the agent goes idle (the turn is fully done), an empty text (isImportant) is pushed: the badge text is wiped and the phase color removed, restoring the official look. Replaces the former conversation-START forced working-pair override (removed 2026-09).

Publishers are fail-safe: a messenger glitch can never disturb the actual compaction.

How it works

agent/request(payload, next)              # every model request
    return await next()                  # pure pass-through (thinking-off scopes
                                          # ONLY to the plugin's own summarizer)

agent/pre-step(payload, next)             # before each model step
    projectedTokens >= autoThresholdTokens?
        no  -> next()                     # let the request proceed
        yes -> compactRegion(head-before-retainLatestTokens, signal)
               return { kind: "reject" }  # no model request this step

agent/status({ agent, status })          # lifecycle transition
    status === "idle" && turnEndForceCompactionEnabled?
        -> compactNow(agent, freshSignal) # turn-end compaction

session/flush(session)                   # durability checkpoint
    select region -> project messages -> preview + shrink gate
    -> compaction.compactRegion(start, end, agent, signal)

Supporting modules:

• src/hooks/guard.js — agent/request pure pass-through + pre-step threshold gate + process-local force flag (thinkingDisabled survives only as a legacy predicate).
• src/hooks/command.js — the /force-compact command (lazily registered).
• src/hooks/idle.js — turn-end forced compaction.
• src/hooks/wire-rewrite.js — the llm/stream LiveUI watermark hook (no wire manipulation; historical note in the module header).
• src/engine/region.js — head/tail-anchored region selection (with the official pairing ledger).
• src/engine/summarizer.js — the one-shot LLM summarizer, fully aligned with official compaction-basic (target resolution, prefix-cache alignment, purpose:'compaction' tag, fail-closed finish classification, usage capture).
• src/engine/builtin.js — the builtin persistent transaction (official compaction/* vocab).
• src/engine/checkpoint.js — preview + shrink gate + delegation to the compaction service.
• src/core/projected.js — provider-anchored projectedTokens.
• src/core/ui-signal.js — the LiveUI messenger.

Install

As an installable package (recommended):

# from npm (published):
npm install @falling-ts/dsh-force-compact
# from git:
dsh plugin --profile web add github:falling-ts/dsh-force-compact
# from a local checkout:
dsh plugin --profile web add ./dsh-force-compact

Or, from a local checkout, as a --patch overlay without installing:

dsh web --patch dsh-force-compact/cordis.patch.yml

The plugin is loaded iff ~/.dsh/logs/dsh-force-compact.log gains:

[force-compact] debug logging enabled — writing [force-compact] lines to <absolute path>

Command availability — /force-compact loads lazily

The commands service arrives with the agent-presets plane, after the plugin's boot-time apply, so registration happens at the first guarded-listener activation (agent/request / agent/pre-step / agent/status / session/flush), settling permanently on the first success. Practical effect: after (re)starting the instance, a fresh session's / picker does NOT show /force-compact until that session makes its first model request — send any one message, then the command is registered process-wide.

• Success: [force-compact] /force-compact command registered (deferred)
• commands permanently absent: one … still UNREGISTERED 10 min … warn explains the empty picker. Until registered, the rest of the plugin works — degradation, not an install failure.

Verify a compaction happened:

idle compaction (builtin) shadowed N nodes (~M tokens)
builtin compaction OK — replaced span seq[A..B] (N nodes, ~K tokens) with a P-char checkpoint
compaction thinking-policy — settings.disableThinking=true → extra.reasoningEffort='off' (…)
summarization wire-fields → <provider>/<model>: reasoningEffort='off' + reasoning_effort="none" (…)

(The last two lines are the per-attempt audit pair described under "Observability".)

Settings

$DSH_HOME/settings.yaml, namespace falling-ts-force-compact:

Example — an aggressive local profile:

falling-ts-force-compact:
  disableThinking: true
  autoThresholdTokens: 40000   # compact sooner ⇒ keep the live prompt small
  retainLatestTokens: 8000
  turnEndForceCompactionEnabled: true

Without the settings service the plugin falls back to the same defaults and still compacts — the namespace is optional, never a hard dependency.

Tuning for low-context llama.cpp

Keep autoThresholdTokens comfortably below the served context: the live prompt stays small and latency flat, while the agent keeps deep memory through the compressed head. Pressure is measured in projected tokens (provider-anchored), so the threshold maps predictably onto the UI figure.

Behavior notes

• Runtime dependency: the compaction service (preset plane agent-presets:compaction-basic), read live via ctx.get('compaction'); unreachable → the builtin engine takes over (or the request proceeds).
• Optional dependencies: settings / tokenMeter / commands / llm / agents are read via ctx.get(...) with guards — a missing one degrades gracefully.
• Per-request settings read: parameters are read every model request, so edits take effect on the next request without a restart.
• Signals: agent/* Waterfalls forward the current turn's signal; the session/flush checkpoint and the agent/status idle listener each mint a fresh AbortController.
• Persistence: durable output is the compaction/* bracket events + a surfaceOp:replace user/message checkpoint, replay-safe across builds.
• Client half: web/client.js adds the settings section "Force Compact" (localized labels), live-editable without restart (uSES-safe mirror).
• One intentional timer: the 3 s publishDone fallback (presentation-only, documented deviation). Otherwise the plugin is pure listeners + a process-local Map force flag.

Screenshots

Settings page — the Force Compact section; all nine fields above are editable live without a restart.

Conversation page — the LiveUI signal paints three states (red: compressing / green: done / blue: working); the green banner fades after about 3 s back to a random working line; at conversation end the badge is cleared (empty text) back to the official look.

License

MIT (see LICENSE).