Wizard schedules emit prose messages where the playbook-call grammar requires a call — 12 entries across 8 wizards + trinity/onboard

[vybe]

The wizards teach every agent they generate how to write a schedule — and they currently teach it wrong. Twelve schedules: entries across eight wizards plus trinity/onboard's schema example carry a prose message: where protocols/playbook-call.md requires a one-line playbook call. Every agent created by these wizards inherits the violation at birth.

Why it matters beyond style

A prose message names no playbook, so:

1. It restates the process instead of calling it. "Sweep tracked competitors for changes — pricing, product, messaging, hiring, news — and report anything material" is a second copy of /monitor's procedure, living in an unversioned scheduler field, free to drift from the SKILL.md that owns it. Changing the playbook no longer changes the schedule.
2. It blinds the new autonomy-mode gate. /audit-wizards v1.3 (issue Promote --autonomous to a bundle-wide (marketplace-wide) run-mode convention for gated skills on crons #6) flags any automation: gated skill named in a schedules: block that declares no --autonomous mode. It resolves "named in" from the playbook-call grammar. A prose message resolves to nothing, so a gated skill scheduled that way is invisible to the gate — the exact /video-intake bug class the gate exists to catch. The check reports these files as a coverage NOTE rather than passing them silently.

Inventory

Class A — a real playbook exists; mechanical rewrite (9)

file	current prose message	→ call
trinity/onboard	"Run /weekly-report and post the summary"	/weekly-report
create-agent/kb-agent	"Run a coherence pass over the knowledge graph…"	/coherence-sweep
create-agent/kb-agent	"Ingest any new source documents… update the graph."	/refresh-index
create-agent/receptionist	"Review the overnight inbox and routing queue…"	/process-inbox
create-agent/prospector	"Refresh research on tracked target accounts…"	/research-company
create-agent/chief-of-staff	"Prepare today's briefing — calendar, top priorities…"	/daily-briefing
create-agent/chief-of-staff	"Produce the weekly digest — what shipped…"	/weekly-digest
create-agent/recon	"Sweep tracked competitors for changes…"	/monitor
create-agent/recon	"Refresh competitor battlecards from the week's findings."	/battlecard

Class B — the schedule calls work that has no playbook at all (2)

file	message	gap
create-agent/webmaster	"Check deployed sites — build status, broken links, performance regressions, and SSL/cert expiry; report issues."	the wizard generates only /create-website, /onboarding, /update-dashboard. There is no site-health playbook to call.
create-agent/ghostwriter	"Draft this week's content calendar across platforms, in the configured brand voice."	generates /write, /set-voice, /library, /repurpose, /hooks. There is no content-plan playbook.

These two are the interesting ones: the prose was load-bearing precisely because it was standing in for a playbook nobody wrote. The schedule promises a capability the generated agent does not have — so today it fires and improvises. Fixing it means authoring the playbook (/site-health, /content-plan) or dropping the schedule; a message rewrite alone cannot fix it.

Class C — one message, several playbooks (1)

file	message	issue
create-agent/doctor	"Review newly ingested health documents, update lab-value trends, and flag anything worth raising with a doctor."	conflates /ingest-documents + /lab-trends + /visit-prep. The grammar is one call per schedule, so this needs either a splitting into separate entries or a weekly-review playbook that composes them.

Already conformant: trinity/create-dashboard (message: "/update-dashboard").

Fix

Class A is a mechanical sweep. Classes B and C need a design call first — which is why they are called out separately rather than folded into the sweep.

Ref: canon/protocols/playbook-call.md, ent#271 decision 11. The same conformance pass was run on trinity-pm's own ten schedule messages (trinity-pm 67597aa); this is the marketplace half, and it matters more because these are templates.

[vybe]

Class A swept — 9 of 12 conformed (898ede4).

file	→ call	version
trinity/onboard	/weekly-report	5.3
kb-agent	/coherence-sweep, /refresh-index	1.10
receptionist	/process-inbox	1.8
prospector	/research-company	1.9
chief-of-staff	/daily-briefing, /weekly-digest	1.9
recon	/monitor, /battlecard	1.8

Plugins: create-agent 1.13.1, trinity 2.7.3. Each schedule's human-readable intent survives in its purpose: field, which is where the schema puts it — nothing was lost, it moved to the key that is actually a human note. onboard's example comment now states the grammar inline, since that example is what every reader copies.

Verified by the gate itself rather than by inspection: /audit-wizards v1.3's coverage NOTE went from 9 files to 3 — exactly the Class B/C set below.

Still open — the 3 that need a design call, not a rewrite:

• Class B (2) — webmaster site-health and ghostwriter content-calendar. The prose was load-bearing: it stood in for a playbook nobody wrote. Both schedules promise the generated agent a capability it does not have, so today they fire and improvise. Needs either /site-health and /content-plan authored in those wizards, or the schedules dropped.
• Class C (1) — doctor's weekly review conflates /ingest-documents + /lab-trends + /visit-prep. Either split it into separate entries or author a weekly-review playbook that composes them.

Keeping this issue open for those three.