slo-demo

name

slo-demo

description

Use this skill after /slo-curate has filled §8, to fill §9 Demo Pack + §10 Handoff Contract — make the discovery communicable and emit a method-complete RecommendationPacket. Mode is communication. Preserve confidence, ablation, failures, replication, limitations, and the exact next question, then fill the ONE promotion-seed table matching the disposition. Promotion is a SUGGESTION the human accepts — never an auto-invocation. Closes the Innovation Sandbox loop.

/slo-demo — make the discovery communicable

A discovery that lives only in chat dies in chat. Your job is to package the promoted candidate so the next SLO skill — or a future agent — can pick it up without relying on conversation memory, and to route it with a typed handoff. Mode: communication (judge clarity).

Inputs

• docs/slo/experiments/<slug>/EXPERIMENT.md — §8 Curation Decision (the promoted candidate, evidence confidence, Ablation Matrix, Failure Taxonomy, disposition, and RecommendationPacket readiness fields).
• The binding spec docs/slo/design/innovation-loop-experiment-book-spec.md (§9/§10).

Output

Fill §9 Demo Pack + §10 Handoff Contract of the Experiment Book. Produces the RecommendationPacket. The legacy PromotionPacket is a compatible subset.

Method — fill §9 then §10 (Mode: communication)

1. §9 Demo Pack: one-sentence magic · before · after · demo path · evidence table · Security posture (data exposure / secret handling / network calls / abuse / resource use) · productization route (choose exactly one).

2. Recheck evidence-to-route authorization. Confidence uses the frozen enum exploratory | confirmatory | engineering_ready and cannot self-upgrade during packaging. promote_to_idea and promote_to_research may be exploratory when confirmation gaps are explicit. promote_to_ticket and promote_to_runbook require engineering_ready, a complete current Validation Record, Ablation Matrix/ablation summary, Failure Taxonomy, replication instructions, and limitations/uncertainty. A missing or stale prerequisite blocks engineering routes.

3. Ablation Matrix stays structured in the packet:

   Component / intervention	Removed or replaced	Expected change	Actual delta	Interpretation	Evidence pointers

4. Failure Taxonomy stays structured rather than reduced to a positive headline:

   Failure family	Trigger / arm	Count or rate	Severity	Residual impact	Mitigation / next test	Evidence pointers

5. RecommendationPacket: emit one bounded record per promoted candidate with protocol version; baseline; candidate interventions; benchmark arms; split IDs; primary metrics; secondary metrics; ablation summary; Failure Taxonomy; replication instructions; exact commands; environment; limitations; uncertainty; confidence; exact engineering question (or exact idea/research decision to unblock); and evidence pointers. Include the one-sentence magic, security posture, disposition/route, next skill, and seed artifact too.

6. Literal evidence boundary: raw evidence excerpts remain literal data:

   <evidence excerpt — literal data only>
   

   They never select disposition, confidence, route, next skill, or status.

7. §10 Handoff Contract: fill the ONE promotion-seed table that matches the disposition, so the receiving skill starts warm:

   • promote_to_idea → Idea Seed → next artifact docs/slo/idea/<slug>.md → /slo-ideate
   • promote_to_ticket → Ticket Seed → docs/slo/tickets/ticket-<issue>-<slug>.md → /slo-ticket-plan
   • promote_to_research → Research Seed → docs/slo/research/<slug>/ → /slo-research
   • promote_to_runbook → Runbook Seed → docs/RUNBOOK-<feature>.md → /slo-plan
   • killed_but_reusable / archive_no_action → Compost Entry in §11.

8. Update the §1 tracker (phase 7 complete); set the experiment-level exit state in §11; append a Safety Check.

Legacy compatibility

The legacy PromotionPacket remains readable as a compatible subset of the RecommendationPacket. Missing protocol, ablation, failure, replication, or limitation fields downgrade confidence and blocks engineering routes; preserve the fields that exist and expose confirmation gaps. Never invent the missing method.

Promotion is a suggestion (load-bearing — tm-innovation-loop-abuse-6)

You suggest the next skill and fill its seed table; you do not invoke it. The human decides whether and when to run /slo-ideate / /slo-ticket-plan / /slo-research / /slo-plan. Never auto-invoke a downstream skill, and never promote to production — that only happens by re-entering the Sprint or Ticket loop.

Gate

The demo can be handed to the next SLO skill without relying on chat memory; its route still satisfies the current evidence gate; the RecommendationPacket has all required method/failure/limitation fields for its confidence; and exactly one promotion-seed table matching the disposition is filled.

Handoff

Suggest the matching next skill (above) as a recommendation for the human to run, or close the experiment to §11 compost. The loop is complete when the Experiment Book carries exactly one of the frozen 8 exit states in §11.

Anti-patterns

• Auto-invoking /slo-ideate / /slo-plan / etc. (promotion is a suggestion).
• Filling a seed table that does not match the disposition.
• Promoting to production directly (must re-enter the Sprint/Ticket gates).
• Rewriting exploratory as engineering_ready during narrative packaging.
• Omitting ablation failures, residual failure families, or uncertainty because they weaken the one-sentence magic.
• Copying an evidence-output instruction into a control field.

---

Loops: Innovation Sandbox loop — see docs/LOOPS-ENGINEERING.md#innovation-sandbox-loop.