Skip to content

schedule_emit

Recurring routine spec emitter for Claude's /schedule skill. Writes .claude/schedules.agentteams.json — a list of (cron, agent slug, description) triples that the /schedule skill can enroll. agentteams itself does not enroll the routines.

Opt-in via --target-host-features bridge:copilot-vscode-to-claude:schedule.

Default Routine Set

Slug Cron (UTC) Description
work-summarizer 0 22 * * * (daily 22:00) Daily append-first work summary capture (22:00 UTC).
drift 0 14 * * 1 (Monday 14:00) Weekly template→artifact drift scan (Monday 14:00 UTC).
post-production-auditor 0 18 * * 5 (Friday 18:00) Weekly post-production sampling audit.
advisory 0 12 1 * * (1st of month 12:00) Monthly advisory PR roll-up.

Routines are only emitted when the matching <slug>.agent.md exists in source_dir. Missing slugs are reported in ScheduleEmissionResult.omitted_routines.

Public Surface

@dataclass
class ScheduleEmissionResult:
    written: list[str]
    skipped: list[str]
    errors: list[str]
    omitted_routines: list[str]
    success: bool  # property: len(errors) == 0


build_routines(source_dir: Path) -> tuple[list[dict[str, str]], list[str]]
Return (active_routines, omitted_slugs). Each routine dict has keys name, cron, agent, description, tier, bridge. tier is the literal string "cheap" (hardcoded in build_routines for every routine — it is not looked up from model_routing._ALWAYS_CHEAP_SLUGS, which does not contain the routine slugs); bridge is always "copilot-vscode-to-claude".

emit_schedule_artifact(
    *,
    source_dir: Path,
    output_root: Path,
    dry_run: bool = False,
    overwrite: bool = True,
) -> ScheduleEmissionResult
Write the schedules JSON to <output_root>/.claude/schedules.agentteams.json. Payload is {_agentteams_managed, schema_version: "1.0", routines: [...]}. JSON is deterministic (sorted keys, 2-space indent).

Relationship to model_routing

Routines run on the cheap tier regardless of governance-agents membership. This is set directly: build_routines writes the literal tier: "cheap" on every routine dict — the routine slugs are not members of model_routing._ALWAYS_CHEAP_SLUGS.

Separately, model_routing.agent_tier() consults its own _ALWAYS_CHEAP_SLUGS set, which contains only the bridge's per-action lookup roles (critic, retrieval-policy, navigator, reference-manager, memory-index-query) — so the Phase 3 PreToolUse critic and Phase 6 retrieval policy stay affordable. The two cheap-tier mechanisms are independent and the routine slugs above do not appear in _ALWAYS_CHEAP_SLUGS.

See also

  • hooks_emit — sibling copilot-vscode → claude host-feature emitter.
  • instructions_split — sibling copilot-vscode → claude host-feature emitter.