Coder/deepseek-harness
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(tools): preserve canonical output boundaries

Browse Source
This commit is contained in:
Tianyi Cui
2026-07-21 18:03:01 +08:00
parent 8f3aca4128
commit e1633fbc3f
33 changed files with 269 additions and 103 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
packages/core/agent-loop/src/tool-calls.ts
View File

@@ -248,7 +248,7 @@ function appendToolResult(
callId: block.id,
content: result.content,
isError: result.isError,
...result.error ? { error: result.error } : {},
...result.error?.info ? { error: result.error.info } : {},
// The tool's private presentation payload (e.g. a result-time diff),
// persisted so a UI bridge reproduces the card on replay.
...result.meta !== undefined ? { meta: result.meta } : {},

2
packages/core/agent-loop/tests/cancel.spec.ts
View File

@@ -397,7 +397,7 @@ describe('Agent.cancel()', () => {
expect(result?.type === 'tool/result' ? result.data : undefined).toMatchObject({
callId: 'c1',
isError: true,
error: { info: { name: 'AbortError', code: 'ABORTED' } },
error: { name: 'AbortError', code: 'ABORTED' },
})
send(agent, 'continue safely')

4
packages/core/agent-loop/tests/contract-regressions.spec.ts
View File

@@ -259,7 +259,7 @@ describe('abort during tool execution ends the turn', () => {
case 'assistant/message': order.push('assistant/message'); break
case 'tool/call': order.push(`tool/call:${event.data.callId}`); break
case 'tool/result': {
const outcome = event.data.error?.info?.code === 'ABORTED' ? 'synthetic-aborted' : 'real'
const outcome = event.data.error?.code === 'ABORTED' ? 'synthetic-aborted' : 'real'
order.push(`tool/result:${event.data.callId}:${outcome}`)
break
}
@@ -308,7 +308,7 @@ describe('abort during tool execution ends the turn', () => {
expect(results[1]!.data).toMatchObject({
callId: CallId('c2'),
isError: true,
error: { info: { name: 'AbortError', code: 'ABORTED' } },
error: { name: 'AbortError', code: 'ABORTED' },
})
})

16
packages/core/agent-loop/tests/tool-calls.spec.ts
View File

@@ -479,18 +479,12 @@ describe('tool-call scheduler: abort handling', () => {
{
callId: CallId('c1'),
isError: true,
error: {
message: 'tool call skipped because the step was aborted before execution',
info: { name: 'AbortError', code: 'ABORTED' },
},
error: { name: 'AbortError', code: 'ABORTED' },
},
{
callId: CallId('c2'),
isError: true,
error: {
message: 'tool call skipped because the step was aborted before execution',
info: { name: 'AbortError', code: 'ABORTED' },
},
error: { name: 'AbortError', code: 'ABORTED' },
},
])
})
@@ -523,7 +517,7 @@ describe('tool-call scheduler: abort handling', () => {
expect(events(agent).filter(e => e.type === 'tool/result').map(e => e.data.callId))
.toEqual([CallId('c1'), CallId('c2')])
expect(events(agent).filter(e => e.type === 'tool/result').at(-1)?.data)
.toMatchObject({ callId: CallId('c2'), isError: true, error: { info: { name: 'AbortError', code: 'ABORTED' } } })
.toMatchObject({ callId: CallId('c2'), isError: true, error: { name: 'AbortError', code: 'ABORTED' } })
})
it('stops replenishing after abort, commits started results, and drains accepted additional contexts', async () => {
@@ -555,7 +549,7 @@ describe('tool-call scheduler: abort handling', () => {
expect(events(agent).filter(e => e.type === 'tool/result').slice(-2).map(e => ({
callId: e.data.callId,
isError: e.data.isError,
errorInfo: e.data.error?.info,
errorInfo: e.data.error,
})))
.toEqual([
{ callId: CallId('c3'), isError: true, errorInfo: { name: 'AbortError', code: 'ABORTED' } },
@@ -601,6 +595,6 @@ describe('tool-call scheduler: abort handling', () => {
expect(events(agent).filter(e => e.type === 'tool/call').map(e => e.data.callId))
.toEqual([CallId('c1'), CallId('c2'), CallId('c3')])
expect(events(agent).filter(e => e.type === 'tool/result').at(-1)?.data)
.toMatchObject({ callId: CallId('c3'), isError: true, error: { info: { name: 'AbortError', code: 'ABORTED' } } })
.toMatchObject({ callId: CallId('c3'), isError: true, error: { name: 'AbortError', code: 'ABORTED' } })
})
})

2
packages/core/session/README.md
View File

@@ -58,7 +58,7 @@ Durable values need one accepted representation, not a check followed by a secon
`context/message` renders its `content` verbatim as a user-role message, and may attach JSON `meta` for replayable plugin state; metadata remains durable but is excluded from `deriveMessages()`.
`tool/result` persists the model-facing content, canonical failure detail, and optional presentation metadata. A tool's successful canonical `value` is deliberately execution-local and never enters the session event, so replay reconstructs the Native/model presentation but cannot recover intermediate programmatic values. This does not change `SESSION_FORMAT_VERSION`: the persisted projection remains authoritative.
`tool/result` persists the model-facing content, optional internal failure identity, and optional presentation metadata. A tool's successful canonical `value` and human-readable canonical failure message remain execution-local; rendered error content is the replay-authoritative message. This preserves the existing event shape and does not change `SESSION_FORMAT_VERSION`.
### Session event vocabulary (`types.ts`)

5
packages/core/session/src/repair.ts
View File

@@ -92,10 +92,7 @@ export function interruptedTurnClosers(events: readonly SessionEvent[]): Session
callId,
content: [{ type: 'text', text: 'Tool call interrupted by a crash; no result was recorded.' }],
isError: true,
error: {
message: 'Tool call interrupted by a crash; no result was recorded.',
info: { name: 'InterruptedError', code: 'interrupted' },
},
error: { name: 'InterruptedError', code: 'interrupted' },
},
surfaceOp: 'append',
...callSeq !== undefined ? { sourceEventSeqs: [callSeq] } : {},

15
packages/core/session/src/types.ts
View File

@@ -243,12 +243,13 @@ export interface SessionEventMap {
*/
'tool/call': { turn: number; step: number; callId: CallId; name: string; arguments: string }
/**
* A completed tool call's model-facing result, canonical failure detail, and
* optional tool-private `meta` presentation payload. `meta` is opaque to the
* core (the producing tool owns its shape and reads it back in `presentResult`)
* but MUST be JSON-serializable: `Session.append` runtime-validates all event
* data with `isJsonValue`, so a non-serializable `meta` is rejected at the
* source, and the durable log reproduces the identical card on replay. Absent
* A completed tool call's model-facing result, optional internal failure
* identity, and optional tool-private `meta` presentation payload. `meta` is
* opaque to the core (the producing tool owns its shape and reads it back in
* `presentResult`) but MUST be JSON-serializable: `Session.append`
* runtime-validates all event data with `isJsonValue`, so a non-serializable
* `meta` is rejected at the source, and the durable log reproduces the
* identical card on replay. Absent
* unless the tool attaches one (e.g. `dsh-tool-fs` carries its result-time
* contextual diff here).
*/
@@ -258,7 +259,7 @@ export interface SessionEventMap {
callId: CallId
content: ContentBlock[]
isError: boolean
error?: { message: string; info?: { name: string; code: string } }
error?: { name: string; code: string }
meta?: JsonValue
}
/** Steering content injected between steps of a running turn. */

2
packages/core/session/tests/repair.spec.ts
View File

@@ -64,7 +64,7 @@ describe('interruptedTurnClosers', () => {
expect(closers.map(e => e.seq)).toEqual([3, 4, 5])
const result = closers[0]!
expect(result.type === 'tool/result' && result.data).toMatchObject({
turn: 2, step: 1, callId: CallId('call-1'), isError: true, error: { info: { code: 'interrupted' } },
turn: 2, step: 1, callId: CallId('call-1'), isError: true, error: { code: 'interrupted' },
})
})

40
packages/core/tools/src/index.ts
View File

@@ -349,6 +349,25 @@ export class ToolOutputError extends HarnessError {
}
}
/** Convert one projector exception into the canonical invalid-output failure. */
function projectionError(toolName: string, projector: 'render' | 'presentationMeta', error: unknown): ToolOutputError {
return new ToolOutputError(toolName, [`output.${projector} failed: ${errorMessage(error)}`])
}
/** Snapshot one projector result before later durable-result materialization. */
function snapshotProjection<T>(toolName: string, projector: 'render' | 'presentationMeta', candidate: T): T {
try {
const detached = snapshotJsonValue(candidate)
if (detached === undefined) {
throw new ToolOutputError(toolName, [`output.${projector} returned non-lossless JSON`])
}
return detached
} catch (error: unknown) {
if (error instanceof ToolOutputError) throw error
throw projectionError(toolName, projector, error)
}
}
/** Successful canonical tool execution, including its Native/model projection. */
export interface ToolExecutionSuccess {
readonly isError: false
@@ -1156,10 +1175,23 @@ export class ToolRegistry extends Service {
const violations = validateJsonSchemaValue(tool.output.schema, detached, 'value')
if (violations.length > 0) throw new ToolOutputError(tool.name, violations)
const value = deepFreeze(detached as JsonValue)
const content = tool.output.render(exec.arguments, value)
const meta = exec.parent === undefined && tool.output.presentationMeta !== undefined
? tool.output.presentationMeta(exec.arguments, value)
: undefined
let rendered: ContentBlock[]
try {
rendered = tool.output.render(exec.arguments, value)
} catch (error: unknown) {
throw projectionError(tool.name, 'render', error)
}
const content = snapshotProjection(tool.name, 'render', rendered)
let meta: JsonValue | undefined
if (exec.parent === undefined && tool.output.presentationMeta !== undefined) {
let projected: JsonValue
try {
projected = tool.output.presentationMeta(exec.arguments, value)
} catch (error: unknown) {
throw projectionError(tool.name, 'presentationMeta', error)
}
meta = snapshotProjection(tool.name, 'presentationMeta', projected)
}
return this.markCanonical(this.materializeFinalResult({
isError: false,
value,

9
packages/core/tools/tests/tools.spec.ts
View File

@@ -147,6 +147,7 @@ describe('ToolRegistry', () => {
})
expect(result.isError).toBe(true)
expect(result.content[0]?.type === 'text' && result.content[0].text).toContain('Error:')
expect(result.error).toMatchObject({ info: { name: 'ToolOutputError', code: 'INVALID_TOOL_OUTPUT' } })
expect(observedError).toBe(true)
})
@@ -209,10 +210,10 @@ describe('ToolRegistry', () => {
}))
const result = await ctx.tools.execute({ callId: CallId(projector), name: `throwing-${projector}`, arguments: {} })
expect(result).toMatchObject({
isError: true,
error: { message: projector === 'render' ? 'renderer exploded' : 'metadata exploded' },
})
expect(result.isError).toBe(true)
expect(result.error?.message)
.toContain(projector === 'render' ? 'renderer exploded' : 'metadata exploded')
expect(result.error?.info).toEqual({ name: 'ToolOutputError', code: 'INVALID_TOOL_OUTPUT' })
expect('value' in result).toBe(false)
})