mirror of
https://github.com/tiennm99/goclaw.git
synced 2026-10-15 22:15:00 +00:00
feat(pruning)!: faithful port of TS context pruning + backfill migration
Port goclaw context pruning to match upstream TS design in
openclaw/src/agents/pi-hooks/context-pruning/:
- Opt-in default: prune only when mode="cache-ttl" (was opt-out)
- Remove Pass 0 per-result 30% guard (duplicated Pass 1 with different
suffix, caused wobble)
- Dedupe double prune call per iteration: PruneStage owns the single
entry point; loop_history only runs limitHistoryTurns + sanitizeHistory
- Add cache-TTL gate for Anthropic prompt cache: skip prune while cache
is live, scoped per-session via sync.Map
- Add context.pruned event emission for observability
- Configurable TTL as Go duration string ("5m", "30s")
BREAKING CHANGE: context pruning now opt-in. Add
contextPruning.mode: "cache-ttl" to config.agents.defaults to restore.
Migration 51 / SQLite v19 backfills mode="cache-ttl" for agents with
existing custom context_pruning config missing the mode field, so
previously-configured agents keep pruning after the opt-in flip.
NULL configs stay NULL (new opt-in default applies).
Web UI adds Cache TTL input + toggle wiring mode to cache-ttl/off.
This commit is contained in:
viettranx committed
1 parent
8d9a7c127a
commit
48335d7797
26 files changed
+730
-115
No files matched your search
+29
-18
@@ -59,15 +59,16 @@ Finalize (runs once, uses background context if cancelled)
|
||||
- Resolve workspace + context files dynamically
|
||||
- Build system prompt (15+ sections)
|
||||
- Inject conversation summary if exists
|
||||
- Run history pipeline (limitHistoryTurns → pruneContextMessages → sanitizeHistory)
|
||||
- Run history pipeline (limitHistoryTurns → sanitizeHistory)
|
||||
- Filter tools through PolicyEngine (RBAC)
|
||||
- Call LLM, record span with token counts
|
||||
- Emit `chunk` events (streaming) or single response
|
||||
|
||||
**PruneStage**
|
||||
**PruneStage** (opt-in via `contextPruning.mode: "cache-ttl"`)
|
||||
- Estimate token ratio vs context window
|
||||
- If >= 30%, run soft trim pass (keep first/last 1500 chars, replace middle with "...")
|
||||
- If >= 25%, run soft trim pass (keep first/last 3000 chars, replace middle with "...")
|
||||
- If >= 50%, run hard clear pass (replace with placeholder)
|
||||
- Run sanitizeHistory to fix broken tool_use/tool_result pairs after prune
|
||||
- Trigger memory flush (synchronous) if compaction threshold exceeded
|
||||
|
||||
**ToolStage**
|
||||
@@ -183,7 +184,7 @@ flowchart TD
|
||||
|
||||
subgraph PH3["Phase 3: Build Messages"]
|
||||
P3A[Build system prompt - 15+ sections] --> P3B[Inject conversation summary if present]
|
||||
P3B --> P3C["History pipeline: limitHistoryTurns --> pruneContextMessages --> sanitizeHistory"]
|
||||
P3B --> P3C["History pipeline: limitHistoryTurns --> sanitizeHistory"]
|
||||
P3C --> P3D[Append current user message]
|
||||
P3D --> P3E[Buffer user message locally - deferred write]
|
||||
end
|
||||
@@ -398,25 +399,20 @@ All security events use the `slog.Warn("security.injection_detected")` conventio
|
||||
|
||||
## 5. History Pipeline
|
||||
|
||||
The history pipeline prepares conversation history before sending it to the LLM. It runs in three sequential stages.
|
||||
The history pipeline prepares conversation history before sending it to the LLM. It runs in two sequential stages. Context pruning is handled separately by PruneStage (opt-in via `contextPruning.mode: "cache-ttl"`).
|
||||
|
||||
```mermaid
|
||||
flowchart TD
|
||||
RAW[Raw Session History] --> S1
|
||||
S1["Stage 1: limitHistoryTurns<br/>Keep the last N user turns<br/>plus their associated assistant/tool messages"] --> S2
|
||||
S2["Stage 2: pruneContextMessages<br/>2-pass tool result trimming<br/>(see Section 6)"] --> S3
|
||||
S3["Stage 3: sanitizeHistory<br/>Repair broken tool_use / tool_result pairing<br/>after truncation"] --> OUT[Cleaned History]
|
||||
S2["Stage 2: sanitizeHistory<br/>Repair broken tool_use / tool_result pairing<br/>after truncation"] --> OUT[Cleaned History]
|
||||
```
|
||||
|
||||
### Stage 1: limitHistoryTurns
|
||||
|
||||
Takes the raw session history and a `historyLimit` parameter. Keeps only the last N user turns along with all associated assistant and tool messages that belong to those turns. Earlier messages are discarded.
|
||||
|
||||
### Stage 2: pruneContextMessages
|
||||
|
||||
Applies the 2-pass context pruning algorithm described in Section 6.
|
||||
|
||||
### Stage 3: sanitizeHistory
|
||||
### Stage 2: sanitizeHistory
|
||||
|
||||
Repairs tool message pairing that may have been broken by truncation or compaction:
|
||||
|
||||
@@ -429,15 +425,17 @@ Repairs tool message pairing that may have been broken by truncation or compacti
|
||||
|
||||
## 6. Context Pruning
|
||||
|
||||
Context pruning reduces oversized tool results using a 2-pass algorithm. It only activates when the estimated token-to-context-window ratio crosses a threshold.
|
||||
Context pruning reduces oversized tool results using a 2-pass algorithm. **It is opt-in** — configure `contextPruning.mode: "cache-ttl"` to enable. When disabled (default), zero overhead. Owned by PruneStage in the agent pipeline.
|
||||
|
||||
```mermaid
|
||||
flowchart TD
|
||||
START[Estimate token ratio vs context window] --> CHECK{Ratio >= softTrimRatio 0.3?}
|
||||
START[Check mode == cache-ttl?] --> GATE{Mode enabled?}
|
||||
GATE -->|No| SKIP[No pruning - zero overhead]
|
||||
GATE -->|Yes| CHECK{Ratio >= softTrimRatio 0.25?}
|
||||
CHECK -->|No| DONE[No pruning needed]
|
||||
CHECK -->|Yes| PASS1
|
||||
|
||||
PASS1["Pass 1: Soft Trim<br/>For each eligible tool result > 4000 chars:<br/>Keep first 1500 chars + last 1500 chars<br/>Replace middle with '...'"]
|
||||
PASS1["Pass 1: Soft Trim<br/>For each eligible tool result > 6000 chars:<br/>Keep first 3000 chars + last 3000 chars<br/>Replace middle with '...'"]
|
||||
PASS1 --> CHECK2{"Ratio >= hardClearRatio 0.5?"}
|
||||
CHECK2 -->|No| DONE
|
||||
CHECK2 -->|Yes| PASS2
|
||||
@@ -446,12 +444,25 @@ flowchart TD
|
||||
PASS2 --> DONE
|
||||
```
|
||||
|
||||
### Configuration
|
||||
|
||||
Enable pruning by setting `contextPruning.mode` in agent defaults:
|
||||
|
||||
```json5
|
||||
agents: {
|
||||
defaults: {
|
||||
contextPruning: { mode: "cache-ttl" }
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
### Defaults
|
||||
|
||||
| Parameter | Default | Description |
|
||||
|-----------|---------|-------------|
|
||||
| `mode` | `""` (disabled) | `""` or `"off"` = disabled; `"cache-ttl"` = enabled |
|
||||
| `keepLastAssistants` | 3 | Number of recent assistant messages protected from pruning |
|
||||
| `softTrimRatio` | 0.3 | Token ratio threshold to trigger Pass 1 |
|
||||
| `softTrimRatio` | 0.25 | Token ratio threshold to trigger Pass 1 |
|
||||
| `hardClearRatio` | 0.5 | Token ratio threshold to trigger Pass 2 |
|
||||
| `minPrunableToolChars` | 50,000 | Minimum tool result length eligible for hard clear |
|
||||
|
||||
@@ -719,8 +730,8 @@ Enabled via the `GOCLAW_TRACE_VERBOSE=1` environment variable.
|
||||
| `internal/agent/loop_run.go` | Run() entry point: dual-mode gate (v2 vs v3), trace creation, span management |
|
||||
| `internal/agent/loop_pipeline_adapter.go` | Bridge v2 Loop to v3 Pipeline: state conversion, dependency injection, callback wiring |
|
||||
| `internal/agent/loop.go` | runLoop() core loop: LLM iteration, tool execution, message buffering (v2 path) |
|
||||
| `internal/agent/loop_history.go` | History pipeline: limitHistoryTurns, pruneContextMessages, sanitizeHistory, summary injection |
|
||||
| `internal/agent/pruning.go` | Context pruning: 2-pass soft trim and hard clear algorithm |
|
||||
| `internal/agent/loop_history.go` | History pipeline: limitHistoryTurns, sanitizeHistory, summary injection |
|
||||
| `internal/agent/pruning.go` | Context pruning: 2-pass soft trim and hard clear algorithm (opt-in via PruneStage) |
|
||||
| `internal/agent/loop_compact.go` | Mid-loop compaction: in-memory message summarization during iterations |
|
||||
| `internal/agent/systemprompt.go` | System prompt assembly (19+ sections), PromptFull and PromptMinimal modes |
|
||||
| `internal/agent/systemprompt_sections.go` | Individual section builders (tooling, workspace, sandbox, skills, MCP, etc.) |
|
||||
|
||||
@@ -6,6 +6,40 @@ All notable changes to GoClaw Gateway are documented here. Format follows [Keep
|
||||
|
||||
## [Unreleased] — 2026-04-15
|
||||
|
||||
#### ElevenLabs Audio Manager Refactor — Phase 6 (2026-04-15)
|
||||
|
||||
Desktop UI parity: voice picker + STT admin form ported from web to Wails desktop frontend. Completes 6-phase plan; all audio surfaces (TTS/STT/Music/SFX) now manageable from both web and desktop Lite editions.
|
||||
|
||||
### Added
|
||||
- **Desktop voice picker**: `ui/desktop/frontend/src/components/agents/voice-picker.tsx` — uses desktop's native Combobox primitive; embedded in `AgentDetailPanel` with `other_config.tts_voice_id` persistence via spread-merge pattern
|
||||
- **Desktop voice preview button**: `ui/desktop/frontend/src/components/agents/voice-preview-button.tsx` — singleton `<audio>` (module-level `currentAudio` ref); hidden when `preview_url` absent (resolves P6-L)
|
||||
- **Desktop STT admin form**: `ui/desktop/frontend/src/components/builtin-tools/stt-provider-form.tsx` — mirrors web fields + `whatsapp_enabled` toggle + privacy banner; routed via `ToolSettingsDialog` when `tool.name === 'stt'`
|
||||
- **Desktop voices service + hook**: `services/voices.ts` (`listVoices`, `refreshVoices`) + `hooks/use-voices.ts` (plain useState/useEffect — no React Query on desktop; mirrors web hook shape)
|
||||
- **Desktop Vitest infrastructure**: `vitest.config.ts` + `test/setup.ts` (jsdom + jest-dom + HTMLMediaElement mock) — resolves P6-H3 (desktop tests previously impossible)
|
||||
- **Desktop i18n `tts` namespace**: `i18n/locales/{en,vi,zh}/tts.json` (15 → 16 total namespaces registered in `i18n/index.ts`) + `tools.json` extended with `builtin.sttForm.*` keys
|
||||
- **Desktop component tests**: voice picker (4 tests) + STT form (3 tests) — parity subset covering render, save, preview, privacy banner
|
||||
|
||||
### Changed
|
||||
- **Desktop `AgentDetailPanel`**: Added `ttsVoiceId` local state + `onSaveWithVoice` wrapper that preserves existing `other_config` fields via spread-merge before overwriting `tts_voice_id`
|
||||
- **Desktop `ToolSettingsDialog`**: Added `tool.name === 'stt'` routing branch before JsonSettingsForm fallback; existing `web_fetch` + MEDIA_TOOLS routing unchanged
|
||||
- **Desktop `package.json`**: Added devDeps (`vitest`, `@testing-library/react`, `@testing-library/jest-dom`, `jsdom`, `@vitejs/plugin-react`) + `test` / `test:watch` npm scripts
|
||||
|
||||
### Resolved
|
||||
- **Audit finding P6-B1**: Gate G5 verified — Phase 5 merged green before Phase 6 work began
|
||||
- **Audit finding P6-B2**: Desktop path rewrite confirmed — no `pages/` dir used; component composition via `AgentDetailPanel` + `ToolSettingsDialog`
|
||||
- **Audit finding P6-B3**: `tts.json` namespace created + registered in desktop `i18n/index.ts` (web-desktop i18n parity)
|
||||
- **Audit finding P6-H1**: `tools.json` extended with STT labels; `tts.json` created per Decision 4
|
||||
- **Audit finding P6-H2**: SttProviderForm integration point identified (`ToolSettingsDialog.tsx` name-routing) + wired
|
||||
- **Audit finding P6-H3**: Vitest installed + configured in desktop frontend (was previously absent)
|
||||
- **Audit finding P6-L**: Voice preview button hidden (not disabled) when `preview_url` null/empty
|
||||
- **Audit finding P6-M** (codec): MP3 + Opus codec smoke test remains on manual verification checklist (user to confirm on macOS WKWebView + Windows WebView2)
|
||||
- **Cross-phase XP-5**: Desktop paths confirmed — `components/agents/`, `components/builtin-tools/`, `services/`, `hooks/` (no `pages/`)
|
||||
|
||||
### Deferred
|
||||
- **Manual smoke test**: `wails dev -tags sqliteonly` voice picker save + STT form submit + codec playback requires user verification (cannot run headless in automation)
|
||||
|
||||
---
|
||||
|
||||
#### ElevenLabs Audio Manager Refactor — Phase 5 (2026-04-15)
|
||||
|
||||
Channel STT migration (Telegram/Feishu/Discord) + WhatsApp voice transcription with tenant opt-in (default disabled due to E2E encryption trade-off, resolves privacy red-team finding).
|
||||
|
||||
Reference in new issue
Block a user