Observation
Round-8 seller validator (PR #205) reported:
"Skill claims '9/9 passing' but the validator runs ~20+ scenarios across tracks I wasn't taught (governance_denied, pending_creatives_to_start, measurement_terms, inventory_list_targeting, delivery_reporting/simulate). Expectation mismatch — WARN without warning. Fresh users read '9/9 passing' then see 'WARN partial' and doubt the skill."
The seller skill produces a schema-valid agent that passes the core 9-step lifecycle. But bash scripts/skill-run.sh seller 3011 media_buy_seller now (post round-7 harness fix) correctly reports WARN: partial — 0 of 2 tracks passing because the full media_buy track runs ~40 scenarios, 13 of which the skill never teaches.
Two paths (pick one)
Option A — Scope the skill's promise. Rewrite the "Validation" section to say: "This skill teaches the 9 core lifecycle scenarios. Expect WARN: partial on the full media_buy_seller storyboard — advanced scenarios (governance denial, pending_creatives lifecycle, measurement_terms validation, inventory_list_targeting persistence, invalid_transition error shapes, delivery_reporting/simulate_and_verify) require additional stubs." Add a "Next steps" pointer for each advanced track.
Option B — Extend the skill to cover the advanced scenarios. Add minimal stub handlers for the 13 missing scenarios. Grow the skill to ~150 additional lines. Agents following the skill hit PASS not WARN.
Recommendation: Option A. The skill stays a tight "how to build a seller" teaching surface; advanced scenarios are sprawling concerns (governance flows, state-machine transitions, measurement compliance) that deserve their own skills or cookbook entries, not a mega-skill. Option B bloats the skill and conceals which patterns are core vs. optional.
Related: #205 (round-8 validation)
Observation
Round-8 seller validator (PR #205) reported:
The seller skill produces a schema-valid agent that passes the core 9-step lifecycle. But
bash scripts/skill-run.sh seller 3011 media_buy_sellernow (post round-7 harness fix) correctly reportsWARN: partial — 0 of 2 tracks passingbecause the fullmedia_buytrack runs ~40 scenarios, 13 of which the skill never teaches.Two paths (pick one)
Option A — Scope the skill's promise. Rewrite the "Validation" section to say: "This skill teaches the 9 core lifecycle scenarios. Expect
WARN: partialon the fullmedia_buy_sellerstoryboard — advanced scenarios (governance denial, pending_creatives lifecycle, measurement_terms validation, inventory_list_targeting persistence, invalid_transition error shapes, delivery_reporting/simulate_and_verify) require additional stubs." Add a "Next steps" pointer for each advanced track.Option B — Extend the skill to cover the advanced scenarios. Add minimal stub handlers for the 13 missing scenarios. Grow the skill to ~150 additional lines. Agents following the skill hit
PASSnotWARN.Recommendation: Option A. The skill stays a tight "how to build a seller" teaching surface; advanced scenarios are sprawling concerns (governance flows, state-machine transitions, measurement compliance) that deserve their own skills or cookbook entries, not a mega-skill. Option B bloats the skill and conceals which patterns are core vs. optional.
Related: #205 (round-8 validation)