Interface: SelfExplainOptions

Defined in: src/lib/trace-toolpack/selfExplain.ts:84

Consumer surface for .selfExplain() on the Agent builder.

Properties

delegate?

readonly optional delegate?: object

Defined in: src/lib/trace-toolpack/selfExplain.ts:92

Answer why-questions on a SEPARATE (typically cheaper) model: the skill unlocks one explain_run tool that runs a nested traceDebugAgent and returns its evidence-cited answer.

maxIterations?

readonly optional maxIterations?: number

id?

readonly optional id?: string

Defined in: src/lib/trace-toolpack/selfExplain.ts:98

Skill id (activation key for read_skill). Default 'self-explain'.

include?

readonly optional include?: SelfExplainInclude

Defined in: src/lib/trace-toolpack/selfExplain.ts:102

Which optional parts of a turn's evidence to capture. Both default true.

instruction?

readonly optional instruction?: string

Defined in: src/lib/trace-toolpack/selfExplain.ts:86

Appended to the recommended skill body (ours stays; yours adds).

maxEvents?

readonly optional maxEvents?: number

Defined in: src/lib/trace-toolpack/selfExplain.ts:109

Cap on retained events per turn (only with include.events). Default 2,000 — enough for a long tool-using turn, small enough that a server holding one binding per agent does not grow without limit. A tail that dropped events says so in inspect_tool_call.

toolpack?

readonly optional toolpack?: TraceToolpackOptions

Defined in: src/lib/trace-toolpack/selfExplain.ts:100

Bounding dials forwarded to the toolpack.

SelfExplainOptions