Rule 3: reject trailing "/schedule X" as closing-flourish variant

[EsatanGW]

Summary

Extends docs/output-craft-discipline.md Rule 3 to explicitly reject trailing "/schedule X" offers as a closing-flourish variant. Agents should not offer to schedule next steps they could simply execute in the same turn; scheduling is only appropriate when gated on external factors (releases, methodology cadences, pending user decisions, upstream artifacts). Updates consumer mirrors in AGENTS.md §10 and CHANGELOG.md; adds ROADMAP closure entry documenting the full initiative lifecycle.

Scope classification

• Full — multiple surfaces, new normative rule in operating contract, cross-cutting consumer updates per CLAUDE.md §5.

Surfaces touched

• Information surface (new normative rule on agent output discipline; affects validation of agent prose output)

Rationale:

• New Rule 3 sub-clause adds a previously-unstated discriminator for closing-flourish detection, binding on agent output validation and review.

Source of Truth impact

• SoT file: docs/output-craft-discipline.md Rule 3 §Must not do (new bullet)
• Consumer updates in this PR: AGENTS.md §10 Rule 3 (summary line extended), CHANGELOG.md [Unreleased] (Added section), ROADMAP.md (initiative closure with phase log)
• Known consumers NOT updated and why: Thin-bridges (CLAUDE.md, GEMINI.md, .windsurfrules, .cursor/rules/*.mdc, agents/*.md) inherit automatically per docs/file-role-map.md; docs/ai-operating-contract.md §7 already points to output-craft as application target, so cross-reference topology remains intact.

Evidence

docs/output-craft-discipline.md Rule 3 §Must not do — new bullet:

- Offer to schedule next steps the agent could simply do. A trailing "I'll schedule X" or "/schedule Y" is a closing flourish in scheduling clothing. The discriminator is whether the next step is **gated on something the agent does not control** — an external release, a methodology-defined cadence (e.g. [`harness-evolution-discipline.md`](harness-evolution-discipline.md)'s per-material-model-release sweep), a user decision still pending, an upstream artifact that has not landed. If the next step is gated externally, schedule it; if it is gated only on the agent doing the work, doing the work is the only honest option. "Let me queue a follow-up" when the follow-up is one tool call away is the trailing offer's slow form — same shape, same fingerprint, same Rule 3 violation.

AGENTS.md §10 Rule 3 — summary line extended:

- **Summaries are caveats and next steps, not recap.** The diff and the manifest are the record; the agent's contribution at the summary slot is what those records do not say — caveats (residual risk, gotchas, what was assumed), next steps (what is still pending, what to confirm). Do not restate completed actions; do not open with preamble; do not close with flourish; do not offer to schedule what the agent could simply do — schedule only what is gated on something outside the agent's control (external release, methodology-defined cadence, user decision still pending, upstream artifact).

CHANGELOG.md [Unreleased] — new Added entry:

- **`docs/output-craft-discipline.md` Rule 3 §Must not do — new bullet rejecting trailing "/schedule X" offers as a closing-flourish variant, plus consumer mirror in `AGENTS.md §10` Rule 3.** [detailed explanation of discriminator and rationale]

ROADMAP.md — initiative closure:

• Initiative no-trailing-schedule-offer moved to Closed initiatives block
• Full phase-gate table (P0–P7) with all gates marked ✅ passed
• Phase log explaining version bump (minor 1.36.0), mode choice (Full), and scope boundaries

Breaking-change level

• L1 — additive + deprecation. New normative rule

https://claude.ai/code/session_01YaxQEUWuh1v61e7SBKodzp