Merge branch 'codex/invariant-service-seam' into codex/invariant-package-registration-gate

This commit is contained in:
Tianyi Cui
2026-07-20 19:48:51 +08:00
391 files changed
+15415 -8532

No files matched your search

+9 -8
View File
@@ -9,13 +9,14 @@ This is the implementation tier of the compaction capability — see the [interf
This backend owns the compaction policy:
- **Measurement** — the singleton `ctx.tokenMeter` prices the latest canonical logged envelope and current surface at one consumed-log revision. Post-step pressure therefore includes the actual system prompt, tools, prefix, routing, assistant completion, tool results, buffered context, and steering.
- **Retention** — compact the oldest whole surface units while preserving a recent tail and balanced tool-call/result cuts through the [`dsh-compact` boundary helpers](../compact/README.md#tool-pairing-boundaries). Turn boundaries do not protect old steps inside a runaway turn. An open indivisible tail declines until it closes; a single unit larger than the budget remains out of scope.
- **Model-free pruning** — after pressure or canonical overflow qualifies, the optional [`ctx.toolResultPrune`](../compact-tool-result-prune/README.md) service rewrites oversized tool results before range selection. Compact-basic remeasures through `ctx.tokenMeter`, skips summarization when pressure becomes safe, and otherwise summarizes the pruned surface. Below-pressure post-step checks never prune.
- **Retention** — compact the oldest whole surface units while preserving a recent tail and balanced tool-call/result cuts through the [`dsh-compact` boundary helpers](../compact/README.md#tool-pairing-boundaries). Turn boundaries do not protect old steps inside a runaway turn. An open indivisible tail declines until it closes. The optional pruner can repair an oversized closed tool unit when its text-bearing result is the removable bulk; indivisible non-tool units and non-prunable tool remainders remain out of scope.
- **Convergence** — retry head-checkpoint compaction up to `compactionRetries`; reject a summary that does not shrink its source, and throw if retries cannot return below threshold.
- **Summarization** — a direct `llm/stream` call uses the configured provider/model pair and cap, falling back to the latest logged request target and then the agent target, without running the loop-only `agent/request` seam. The input transcript preserves non-text blocks as tagged placeholders; only returned text enters the checkpoint, excluding reasoning and tool calls that would leak private reasoning or create an orphaned call.
- **Framing** — the replacement user message marks established checkpoint context with `<compacted-summary>` tags. The raw summary remains on the provenance event, and later automatic cycles merge the prior checkpoint.
- **Lifecycle** — `compactRegion()` mutates `agent.session` and records its start, summary, replacement, and end. The serial `agent/post-step` listener checks pressure after successful output and tool work are durable but before `step/end`. Canonical provider overflow is handled through `agent/request-error` after the failed step closes.
- **Overflow recovery** — below-threshold overflow bypasses normal retention and attempts one maximal balanced head reduction while leaving the newest indivisible unit. Retry is authorized only when `surface.replaceGeneration` advances; no range, no replacement, recovery failure, an exhausted cap, cancellation, or an unknown/noncanonical error preserves the original provider failure.
- **Failure handling** — an unmatched `compact/start` is an inert crash marker because no replacement landed. A region failure records an error end and leaves the surface unchanged. Operational post-step failures warn and continue; overflow-recovery failure preserves the original provider error.
- **Overflow recovery** — below-threshold overflow bypasses normal retention, prunes, then attempts one maximal balanced head reduction while leaving the newest indivisible unit. Retry is authorized whenever `surface.replaceGeneration` advances, including when pruning lands before later summary work throws. No replacement, an exhausted cap, cancellation, or an unknown/noncanonical error preserves the original provider failure.
- **Failure handling** — an unmatched `compact/start` is an inert crash marker because no summary replacement landed. A region failure records an error end; the surface remains unchanged unless pruning already landed. Operational post-step failures warn and continue, while overflow-recovery failure preserves the original provider error only when no earlier replacement advanced the surface. Cancellation remains authoritative after any progress.
The protected `summarize()` method is the sole subclass hook. A template- or remote-summarizer subclass can override it while pressure, retention, provenance, shrink validation, and shadowed-token accounting stay on `ctx.tokenMeter`. The hook returns the summary blocks together with the call envelope it used (`{ summary, provider, model, maxTokens? }`), which is logged on `compact/summary`.
@@ -50,7 +51,7 @@ export function apply(ctx: Context): void {
}
```
Loading the plugin registers `ctx.compact`. With `auto: true` (the default) it compacts automatically under token pressure; a consumer (a future `/compact` tool) can also call `ctx.compact.compactIfNeeded(...)` or `ctx.compact.compactRegion(...)` directly.
Loading the plugin registers `ctx.compact`. Add [`dsh-compact-tool-result-prune`](../compact-tool-result-prune/README.md) as a sibling before this plugin to enable the optional model-free pass. With `auto: true` (the default) it compacts automatically under token pressure; a consumer (a future `/compact` tool) can also call `ctx.compact.compactIfNeeded(...)` or `ctx.compact.compactRegion(...)` directly.
## Model Experience
@@ -58,7 +59,7 @@ Loading the plugin registers `ctx.compact`. With `auto: true` (the default) it c
#### What the model sees
After a successful step crosses the threshold, the next request receives the checkpoint preamble below, a blank line, `<compacted-summary>`, the data-dependent summary, and `</compacted-summary>`. Overflow recovery rebuilds the immediate retry from that replacement. This one checkpoint replaces the selected older range and is followed by the retained recent units.
After a successful step crosses the threshold, oversized tool results are first rewritten when the optional pruner is loaded. If summarization remains necessary, the next request receives the checkpoint preamble below, a blank line, `<compacted-summary>`, the data-dependent summary, and `</compacted-summary>`. Overflow recovery rebuilds the immediate retry from whatever replacement advanced the surface. A checkpoint replaces the selected older range and is followed by the retained recent units.
##### Conversation checkpoint preamble
@@ -68,7 +69,7 @@ This is an automatically generated checkpoint condensing an earlier span of the
#### Token effect
The replacement reduces future input history rather than appending a second copy. The summary remains until a later compaction replaces it; one oversized indivisible unit can still exceed the budget.
Model-free pruning can avoid the auxiliary call entirely; otherwise it reduces that call's transcript before the summary replaces an older range. The replacement reduces future input history rather than appending a second copy. A summary remains until a later compaction replaces it, while an indivisible non-tool unit can still exceed the budget.
#### KV Cache effect
@@ -144,7 +145,7 @@ Prefix-stable for auxiliary calls while this instruction and the summarizer rout
- **Meter accuracy follows the fixed heuristic** — missing reusable provider usage falls back to character count plus structural overhead rather than exact tokenization.
- **Overflow classification is adapter-maintained** — provider wording can change; both DeepSeek adapters normalize currently recognized context-limit failures to `CONTEXT_WINDOW_EXCEEDED`.
- **Single-unit and envelope-only overflow remain outside surface compaction** — recovery cannot split one indivisible message/tool unit or shrink system/tools/prefix.
- **Some indivisible-unit and envelope-only overflow remains outside surface compaction** — recovery cannot shrink system/tools/prefix, split an indivisible non-tool node, or repair a tool unit whose non-prunable remainder still exceeds the window. The optional pruner can shrink text-bearing tool-result bulk inside an otherwise indivisible pair.
- **`compactRegion` requires an open turn** — a manual call on a fully-closed session throws ("no open turn") rather than compacting.
- **Summarization failure fails closed with full, over-budget history** — including truncation at the summarization `maxTokens`, which hidden reasoning tokens can consume; the auto path logs a warning and proceeds.
- **Summarization failure preserves the latest durable surface** — before any replacement, the auto path logs a warning and proceeds with full over-budget history. If pruning already landed, a later summarization failure proceeds from that durable pruned surface. Summarization truncation at `maxTokens`, which hidden reasoning tokens can consume, follows the same rule.
- **The summarization call has no transcript-snapshot coverage** — `dsh-llm-replay` derives calls from `assistant/chunk` events, so this chunk-less direct `ctx.llm.stream()` call cannot replay (named deferred replay infrastructure in [the seam Agent Note](../../../.agents/notes/implemented/feature/2026-06-18-compaction-capability-seam.md)).
@@ -33,8 +33,14 @@
"@deepseek-ai/dsh-llm": "^0.0.1",
"@deepseek-ai/dsh-session": "^0.0.1",
"@deepseek-ai/dsh-token-meter": "^0.0.1",
"@deepseek-ai/dsh-compact-tool-result-prune": "^0.0.1",
"cordis": "^4.0.0-rc.7"
},
"peerDependenciesMeta": {
"@deepseek-ai/dsh-compact-tool-result-prune": {
"optional": true
}
},
"dependencies": {
"schemastery": "^3.18.0"
},
@@ -49,6 +55,7 @@
"@deepseek-ai/dsh-llm": "workspace:^",
"@deepseek-ai/dsh-session": "workspace:^",
"@deepseek-ai/dsh-token-meter": "workspace:^",
"@deepseek-ai/dsh-compact-tool-result-prune": "workspace:^",
"@deepseek-ai/dsh-tools": "workspace:^",
"cordis": "^4.0.0-rc.7"
}
+41 -14
View File
@@ -12,6 +12,8 @@ import type { Session } from '@deepseek-ai/dsh-session'
import { CONTEXT_WINDOW_EXCEEDED_CODE, assertNever } from '@deepseek-ai/dsh-llm'
import type { ContentBlock } from '@deepseek-ai/dsh-llm'
import type { Agent } from '@deepseek-ai/dsh-agent'
// Type-only: makes the optional sibling service available to `ctx.get()`.
import type {} from '@deepseek-ai/dsh-compact-tool-result-prune'
import { resolveConfig } from './config.ts'
import { compactSurfaceRegion, selectCompactableRange } from './region.ts'
import { summarizeWithLlm } from './summarizer.ts'
@@ -98,22 +100,35 @@ export class BasicCompactService extends CompactService {
|| retryAttempt >= this.config.maxOverflowRetries
|| signal.aborted) return next()
let generation: number
const generation = agent.session.surface.replaceGeneration
let result: CompactionResult | null
try {
generation = agent.session.surface.replaceGeneration
result = await this.compactIfNeeded(agent, 'context-overflow', signal)
} catch (recoveryError: unknown) {
const message = recoveryError instanceof Error ? recoveryError.message : String(recoveryError)
// A model-free prune can land before later summary work fails. That
// durable reduction is sufficient retry proof; do not discard it just
// because the optional second phase threw. Cancellation still wins.
// eslint-disable-next-line @typescript-eslint/no-unnecessary-condition -- signal can abort while recovery is awaited.
if (!signal.aborted && agent.session.surface.replaceGeneration > generation) {
ctx.logger.warn(
`context-overflow compaction failed after durable surface progress: ${message}; `
+ 'retrying from the replacement surface',
)
return { action: 'retry' }
}
ctx.logger.warn(
`context-overflow compaction failed: ${message}; preserving the original request error`,
// eslint-disable-next-line @typescript-eslint/no-unnecessary-condition -- signal can abort while recovery is awaited.
`context-overflow compaction failed: ${message}; ${signal.aborted
? 'cancellation prevents retry'
: 'preserving the original request error'}`,
)
return next()
}
// eslint-disable-next-line @typescript-eslint/no-unnecessary-condition -- signal can abort while compaction is awaited.
if (signal.aborted || result === null
if (signal.aborted
|| agent.session.surface.replaceGeneration <= generation) return next()
logResult(result, 'context overflow recovery')
if (result !== null) logResult(result, 'context overflow recovery')
return { action: 'retry' }
})
}
@@ -142,7 +157,7 @@ export class BasicCompactService extends CompactService {
* @param agent - agent whose latest durable routed request is measured.
* @param trigger - normal post-step pressure or context-overflow recovery.
* @param signal - live turn cancellation signal forwarded to summarization.
* @returns the latest compaction result, or `null` when no check/work applies.
* @returns the latest summary compaction result, or `null` when no summary ran.
*/
override async compactIfNeeded(
agent: Agent,
@@ -152,22 +167,34 @@ export class BasicCompactService extends CompactService {
const model = routedModel(agent.session)
if (model === undefined) return null
const meter = this.ctx.tokenMeter
const threshold = Math.floor(meter.contextWindow * this.config.thresholdRatio)
let measurement = meter.measure(agent.session)
switch (trigger) {
case 'context-overflow': {
const measurement = meter.measure(agent.session)
const range = selectCompactableRange(agent.session, measurement, 0)
if (range === null) return null
return this.compactRegion(range.start, range.end, agent, signal)
}
case 'context-overflow':
break
case 'pressure':
if (measurement.totalTokens < threshold) return null
break
/* v8 ignore next -- closed-union exhaustiveness guard */
default:
assertNever(trigger, 'compaction trigger')
}
const threshold = Math.floor(meter.contextWindow * this.config.thresholdRatio)
let measurement = meter.measure(agent.session)
// Pruning is optional so compact-basic remains independently composable.
// Once either trigger qualifies, land the model-free pass before choosing
// a summary range, then remeasure through the singleton replay fold.
const prune = this.ctx.get('toolResultPrune')
if (prune !== undefined) {
prune.pruneSession(agent.session)
measurement = meter.measure(agent.session)
}
if (trigger === 'context-overflow') {
const range = selectCompactableRange(agent.session, measurement, 0)
if (range === null) return null
return this.compactRegion(range.start, range.end, agent, signal)
}
if (measurement.totalTokens < threshold) return null
let result: CompactionResult | null = null
@@ -11,6 +11,7 @@ import type { ContentBlock, GenerateOptions, StreamChunk } from '@deepseek-ai/ds
import { Session, SessionId } from '@deepseek-ai/dsh-session'
import TokenMeterService from '@deepseek-ai/dsh-token-meter'
import { agentEvents, type Agent } from '@deepseek-ai/dsh-agent'
import ToolResultPruneService from '@deepseek-ai/dsh-compact-tool-result-prune'
const SIGNAL = new AbortController().signal
const MODEL = 'test-model'
@@ -97,6 +98,43 @@ function toolConversation(): Session {
return session
}
/** One closed routed tool step followed by an open turn for rewrite events. */
function oversizedToolResult(chars = 3_000, withCompactablePrompt = false): Session {
const session = new Session(SessionId(`oversized-tool-${chars}`))
const callId = CallId('oversized')
session.append('turn/start', { turn: 1, trigger: { kind: 'message', source: { kind: 'user' } } })
if (withCompactablePrompt) {
session.append('user/message', {
content: [{ type: 'text', text: 'older history '.repeat(200) }],
source: { kind: 'user' },
}, { surfaceOp: 'append' })
}
session.append('step/start', { turn: 1, step: 1 })
session.append('request/header', {
header: { config: { provider: MODEL, model: MODEL } },
reason: 'initial',
})
session.append('assistant/message', {
turn: 1,
step: 1,
content: [{ type: 'tool-call', id: callId, name: 'bash', arguments: '{}' }],
provenance: { provider: MODEL, model: MODEL },
}, { surfaceOp: 'append' })
session.append('tool/call', { turn: 1, step: 1, callId, name: 'bash', arguments: '{}' })
session.append('tool/result', {
turn: 1,
step: 1,
callId,
content: [{ type: 'text', text: 'X'.repeat(chars) }],
isError: false,
meta: { presentation: 'preserved' },
}, { surfaceOp: 'append' })
session.append('step/end', { turn: 1, step: 1 })
session.append('turn/end', { turn: 1, reason: { kind: 'completed' } })
session.append('turn/start', { turn: 2, trigger: { kind: 'message', source: { kind: 'user' } } })
return session
}
class TestCompactService extends BasicCompactService {
summary: ContentBlock[] = [{ type: 'text', text: 'small checkpoint' }]
summaryProvider = 'summary-provider'
@@ -416,6 +454,78 @@ describe('pressure measurement and retention', () => {
})
})
describe('optional model-free tool-result pruning', () => {
const pruneConfig = { thresholdChars: 100, headChars: 20, tailChars: 10 }
it('does not prune a below-pressure session opportunistically', async () => {
const ctx = createContext(10_000)
const prune = new ToolResultPruneService(ctx, pruneConfig)
const compact = new TestCompactService(ctx, {
auto: false,
thresholdRatio: 0.8,
retainTokens: 100,
})
const session = oversizedToolResult()
const pruneSession = vi.spyOn(prune, 'pruneSession')
expect(await compactIfNeeded(compact, session)).toBeNull()
expect(pruneSession).not.toHaveBeenCalled()
expect(compact.calls).toHaveLength(0)
expect(session.surface.replaceGeneration).toBe(0)
})
it('skips LLM summarization when pruning alone clears pressure', async () => {
const ctx = createContext(1_000)
void new ToolResultPruneService(ctx, pruneConfig)
const compact = new TestCompactService(ctx, {
auto: false,
thresholdRatio: 0.5,
retainTokens: 50,
})
const session = oversizedToolResult()
expect(ctx.tokenMeter.measure(session).totalTokens).toBeGreaterThanOrEqual(500)
expect(await compactIfNeeded(compact, session)).toBeNull()
expect(ctx.tokenMeter.measure(session).totalTokens).toBeLessThan(500)
expect(compact.calls).toHaveLength(0)
expect(session.surface.replaceGeneration).toBe(1)
})
it('summarizes the pruned surface when pruning is insufficient', async () => {
const ctx = createContext(2_000)
void new ToolResultPruneService(ctx, pruneConfig)
const compact = new TestCompactService(ctx, {
auto: false,
thresholdRatio: 0.5,
retainTokens: 50,
})
const session = toolConversation()
expect(await compactIfNeeded(compact, session)).not.toBeNull()
expect(compact.calls).toHaveLength(1)
expect(compact.calls[0]!.text).toContain('tool result middle pruned')
expect(compact.calls[0]!.text).not.toContain('result 1 '.repeat(300))
})
it('retains the original compact-basic behavior without the optional plugin', async () => {
const ctx = createContext(2_000)
const compact = new TestCompactService(ctx, {
auto: false,
thresholdRatio: 0.5,
retainTokens: 50,
})
const session = oversizedToolResult(3_000, true)
expect(await compactIfNeeded(compact, session)).not.toBeNull()
expect(compact.calls).toHaveLength(1)
const original = session.events.find(event => event.type === 'tool/result')
expect(original?.type === 'tool/result' && original.data.content[0])
.toEqual({ type: 'text', text: 'X'.repeat(3_000) })
expect(session.events.filter(event =>
event.type === 'tool/result' && event.surfaceOp !== 'append')).toHaveLength(0)
})
})
describe('compaction region transaction', () => {
it('lands a framed, replayable checkpoint with exact pricing provenance', async () => {
const compact = service()
@@ -878,6 +988,89 @@ describe('automatic listener and loader composition', () => {
expect(session.surface.nodes).toContain(retainedSeq)
})
it('authorizes overflow retry when pruning alone advances an indivisible surface', async () => {
const ctx = createContext(10_000)
void new ToolResultPruneService(ctx, {
thresholdChars: 100,
headChars: 20,
tailChars: 10,
})
const compact = new TestCompactService(ctx, {
thresholdRatio: 1,
retainTokens: 900,
})
const session = oversizedToolResult()
expect(await recover(ctx, agent(session, MODEL), overflow())).toEqual({ action: 'retry' })
expect(session.surface.replaceGeneration).toBe(1)
expect(session.events.some(event => event.type === 'compact/summary')).toBe(false)
expect(compact.calls).toHaveLength(0)
})
it('continues overflow recovery with summarization on the pruned surface', async () => {
const ctx = createContext(10_000)
void new ToolResultPruneService(ctx, {
thresholdChars: 100,
headChars: 20,
tailChars: 10,
})
const compact = new TestCompactService(ctx, {
thresholdRatio: 1,
retainTokens: 900,
})
const session = toolConversation()
expect(await recover(ctx, agent(session, MODEL), overflow())).toEqual({ action: 'retry' })
expect(session.events.some(event => event.type === 'compact/summary')).toBe(true)
expect(compact.calls).toHaveLength(1)
expect(compact.calls[0]!.text).toContain('tool result middle pruned')
})
it('retries from a durable prune when later overflow summarization throws', async () => {
const ctx = createContext(10_000)
const warnings: string[] = []
ctx.logger.warn = ((message: string) => void warnings.push(message)) as typeof ctx.logger.warn
void new ToolResultPruneService(ctx, {
thresholdChars: 100,
headChars: 20,
tailChars: 10,
})
const compact = new TestCompactService(ctx, {
thresholdRatio: 1,
retainTokens: 900,
})
compact.error = new Error('summary unavailable after prune')
const session = oversizedToolResult(3_000, true)
expect(await recover(ctx, agent(session, MODEL), overflow())).toEqual({ action: 'retry' })
expect(session.surface.replaceGeneration).toBe(1)
expect(session.events.filter(event => event.type === 'tool/result')).toHaveLength(2)
expect(session.events.findLast(event => event.type === 'compact/end')?.data)
.toMatchObject({ error: 'summary unavailable after prune' })
expect(warnings).toContainEqual(expect.stringContaining('retrying from the replacement surface'))
})
it('lets cancellation win when summary throws after a durable prune', async () => {
const ctx = createContext(10_000)
const controller = new AbortController()
void new ToolResultPruneService(ctx, {
thresholdChars: 100,
headChars: 20,
tailChars: 10,
})
const compact = new TestCompactService(ctx, {
thresholdRatio: 1,
retainTokens: 900,
})
compact.mutateDuringSummary = () => { controller.abort('cancelled during summary') }
compact.error = new Error('summary cancelled after prune')
const session = oversizedToolResult(3_000, true)
expect(await recover(ctx, agent(session, MODEL), overflow(), 0, controller.signal))
.toEqual({ action: 'fail' })
expect(session.surface.replaceGeneration).toBe(1)
})
it('preserves the newest whole tool-call/result pair during forced overflow compaction', async () => {
const ctx = createContext()
void new TestCompactService(ctx, {
@@ -9,6 +9,7 @@ import Include from '@cordisjs/plugin-include'
import LlmService from '@deepseek-ai/dsh-llm'
import TokenMeterService from '@deepseek-ai/dsh-token-meter'
import BasicCompactService from '@deepseek-ai/dsh-compact-basic'
import ToolResultPruneService from '@deepseek-ai/dsh-compact-tool-result-prune'
let root: string | undefined
let context: Context | undefined
@@ -32,6 +33,7 @@ async function loadYaml(lines: readonly string[]): Promise<Context> {
const modules = new Map<string, unknown>([
['@deepseek-ai/dsh-llm', LlmService],
['@deepseek-ai/dsh-token-meter', TokenMeterService],
['@deepseek-ai/dsh-compact-tool-result-prune', ToolResultPruneService],
['@deepseek-ai/dsh-compact-basic', BasicCompactService],
])
context.loader.internal = {
@@ -50,12 +52,17 @@ async function loadYaml(lines: readonly string[]): Promise<Context> {
}
describe('real Loader composition', () => {
it('loads the flat token-meter and compact-basic YAML shape', async () => {
it('loads the shipped token-meter, pruning, and compact-basic YAML order', async () => {
const loaded = await loadYaml([
"- name: '@deepseek-ai/dsh-llm'",
"- name: '@deepseek-ai/dsh-token-meter'",
' config:',
' contextWindow: 4096',
"- name: '@deepseek-ai/dsh-compact-tool-result-prune'",
' config:',
' thresholdChars: 100',
' headChars: 20',
' tailChars: 10',
"- name: '@deepseek-ai/dsh-compact-basic'",
' config:',
' thresholdRatio: 0.5',
@@ -68,6 +75,7 @@ describe('real Loader composition', () => {
.map(entry => entry.options.name)
expect(unloaded).toEqual([])
expect(loaded.tokenMeter.contextWindow).toBe(4096)
expect(loaded.get('toolResultPrune')).toBeInstanceOf(ToolResultPruneService)
expect(loaded.get('compact')).toBeInstanceOf(BasicCompactService)
expect((loaded.compact as BasicCompactService).config).toMatchObject({
thresholdRatio: 0.5,
@@ -32,6 +32,9 @@
},
{
"path": "../../support/invariants"
},
{
"path": "../compact-tool-result-prune"
}
]
}