agentctx

type ExecMark ¶ added in v0.43.0

type ExecMark struct {
	// Agent is the variant-scoped pseudo-identity written to the event's
	// agent field, e.g. "pamela#dryrun" or "pamela#eval:candidate". The "#"
	// is rejected by resource-name validation, so it can never collide with a
	// real agent, and exact-match agent queries exclude these events for free.
	Agent string
	// Source is the event source, e.g. "dryrun" or "eval". Exact-match
	// filterable, and excludable via the audit exclude_source filter.
	Source string
	// Summary requests reduced emission (lifecycle and errors only) instead of
	// the default full live-turn semantics.
	Summary bool
}

ExecMark labels every audit event produced by a turn that ran under a non-live execution policy (a dry-run preview or an eval sample), so the record stays complete while the view can stay quiet.

It travels in the context rather than as a parameter because audit events are emitted from a dozen places in the turn — including helpers that never see the policy — and a single overlay at the emit site is the only way to mark all of them without threading the policy everywhere.

func ExecMarkFrom(ctx context.Context) *ExecMark

type SkillSummary ¶ added in v0.31.8

type SkillSummary struct {
	Name         string
	Description  string
	Body         string // skill body (markdown); truncated at render time
	IsScheduled  bool
	ScheduleName string
}

SkillSummary carries lightweight metadata about the skill driving the current session. Used by the supervisor to understand *why* a tool call is being made, especially for scheduled skill invocations.

func SkillContext(ctx context.Context) *SkillSummary