Albert/add video marketing

[albertsalgueda]

No description provided.

[github-actions]

Claude finished @albertsalgueda's task in 2m 26s —— View job

---

PR Review: video-marketing skills

• Read changed files and diff
• Check version sync, JSON manifests, cross-skill references
• Review new video-marketing workflows and references
• Post review feedback

Solid, well-structured addition. The new video-ad surface is coherent: routing is disambiguated across social-video-short (continuous) / ugc-video-ad (scripted) / cinematic-product-ad (no-dialogue commercial) / launch-teaser (designed film), and each workflow points at the right sibling. I verified the things most likely to break CI and cross-skill navigation.

Verified clean:

• Version sync — VERSION, krea-generate/VERSION, package.json, all three plugin manifests, marketplace.json, README badge, and all four SKILL.md frontmatter are aligned at 0.6.1. validate.yml's version-sync check should pass.
• JSON manifests — all still valid JSON; only the version string changed.
• Cross-references resolve — new refs (ugc-scripts.md, video-ad-qa.md, video-ad-post.md) and every relative link I checked exist: ../../krea-generate/workflows/lora-train-and-use.md, ../../../krea-marketing/workflows/cinematic-product-ad.md from seedance-2.md, meta-ads-mcp.md, and the marketing-creative-anatomy.md "Video Hook Families" section (quiet confession / product proof / pattern interrupt) that ugc-scripts.md maps onto.
• ffmpeg snippets — drawtext, concat-demuxer, and loudnorm/amix examples are syntactically sound and use placeholder tokens consistently.

Findings (2, both in evals/scenarios.md)

1. Scenario count header is off by one (line 3) — header says "45 scenarios" but the file now has 46 (base 40 + new 41–46). Confirmed via grep -c '^### [0-9]'.
2. Scenario 46 uses an undocumented category guardrail (line 414) — the documented enum is routing | refusal | cost | vision | polling | edge_case (scenarios.md line 5 + README). run.sh reads the category as free text so it won't crash, but it buckets the scenario under an unexpected label. Use refusal/edge_case or extend the enum in both places.

Minor (non-blocking, no change required)

• 70-score threshold wording: ugc-video-ad.md step 10 / Banned treat 70 as the delivery floor, while video-ad-qa.md's table splits below-70 into "55–69 rework before delivering" and "<55 do not deliver as-is." The Banned line ("do not deliver below 70 without telling the user the score") reads as may deliver 55–69 with disclosure, which slightly softens the table. Harmless, but you may want one canonical rule.

Neither finding blocks merge; #1 and #2 are quick doc fixes. Nice work on the live-verified prompt architecture and the start-image-is-the-quality-ceiling gating — those read as genuinely actionable.
· albert/add-video-marketing

[github-actions]

Claude finished @albertsalgueda's task in 1m 32s —— View job

---

PR Review: video-marketing skills (re-review)

• Read new commits and diff vs origin/main
• Verify prior review findings were addressed
• Re-check version sync, cross-references, and new workflow prose
• Post review feedback

Re-reviewed after e6cb4bb Address eval scenario review comments. Both prior findings are fixed and I found no new blockers.

Prior findings — resolved:

1. ✅ Scenario count — evals/scenarios.md:3 now reads 46 scenarios; grep -c '^### [0-9]' returns 46. Matches.
2. ✅ Undocumented guardrail category — scenario 46 is now Category: refusal. No guardrail references remain anywhere in evals/ or krea-marketing/.

Re-verified clean:

• Version sync — VERSION, krea-generate/VERSION, package.json, all three plugin manifests, marketplace.json, and all four SKILL.md frontmatters are aligned at 0.6.1; README badge matches. validate.yml version-sync should pass.
• Cross-references resolve — every relative link in the new/changed files checks out: ugc-video-ad.md → ugc-scripts.md, ugc-social-video.md, video-ad-qa.md, video-ad-post.md, marketing-creative-anatomy.md, ../../krea-generate/references/{cost-preflight,progress-reporting}.md; cinematic-product-ad.md → ../../krea-generate/references/models/seedance-2.md and back from seedance-2.md:643. Sibling routing (social-video-short ↔ ugc-video-ad ↔ cinematic-product-ad ↔ launch-teaser) is disambiguated symmetrically in every trigger block.
• Routing table — krea-marketing/SKILL.md:67-68 and full-ad-campaign.md both point scripted UGC → ugc-video-ad.md and no-dialogue commercial → cinematic-product-ad.md. Consistent.

Minor (now largely reconciled, non-blocking):

• The 70-score threshold wording tightened: ugc-video-ad.md:32 step 10 now says "Below 70: apply the per-criterion fix and rework before delivering," which lines up with video-ad-qa.md:34-36 (70–84 deliver / 55–69 rework / <55 discard). The only residual softness is the Banned line ugc-video-ad.md:50 ("do not deliver below a 70 … without telling the user the score"), which still reads as may deliver 55–69 with disclosure. Harmless given step 10 is the canonical gate; you could align the Banned line to "rework before delivering" if you want one voice.

Nice, well-structured addition — the live-verified prompt architecture and the start-image-is-the-quality-ceiling gating in cinematic-product-ad.md are genuinely actionable. Looks ready to merge.
· albert/add-video-marketing

[github-actions]

Claude finished @albertsalgueda's task in 47s —— View job

---

PR Review: video-marketing skills (re-review after f79bb10)

• Read latest commit diff (Organize new eval scenario groups)
• Re-verify version sync, JSON manifests, cross-references
• Re-check eval scenario category enum + counts
• Post review feedback

Re-reviewed after f79bb10 Organize new eval scenario groups. This commit only adds two section headers to evals/scenarios.md to group the new scenarios. No new blockers; both prior findings remain fixed.

New commit — verified clean:

• Group counts add up to 46. The new headers ## Scripted UGC video ads 0.6.0 (4) (scenarios 41–44) and ## Cinematic product ads 0.6.1 (2) (scenarios 45–46) are correct. Summing every ## …(N) group header: 5+3+3+3+3+3+2+9+8+1+4+2 = 46, which matches both the file header (46 scenarios) and grep -c '^### [0-9]' (46).
• Categories 41–46 all use the documented enum — routing, edge_case, refusal, vision, routing, refusal. No stray guardrail remains. run.sh will bucket every scenario under an expected label.
• New headers are inside the scenario body, above ## Format spec for the runner (line 430), so they don't disrupt the runner's ^### [0-9] scenario parsing.

Still clean from prior passes (unchanged by this commit):

• Version sync — VERSION, krea-generate/VERSION, package.json, all three plugin manifests, marketplace.json, README badge, and all four SKILL.md frontmatters aligned at 0.6.1.
• Cross-references resolve — ugc-video-ad.md, cinematic-product-ad.md, and the new references (ugc-scripts.md, video-ad-qa.md, video-ad-post.md) all resolve, including the seedance-2.md ↔ cinematic-product-ad.md round-trip.
• Sibling routing — social-video-short ↔ ugc-video-ad ↔ cinematic-product-ad ↔ launch-teaser disambiguated symmetrically.

Minor (non-blocking, unchanged): the ugc-video-ad.md:50 Banned line ("do not deliver below a 70 … without telling the user the score") still reads as may deliver 55–69 with disclosure, slightly softer than the video-ad-qa.md table's "55–69 rework before delivering." Step 10 is the canonical gate, so this is harmless — align the Banned line to "rework before delivering" only if you want one voice.

Looks ready to merge.
· albert/add-video-marketing

[github-actions]

Claude finished @albertsalgueda's task in 1m 16s —— View job

---

PR Review: video-marketing skills (re-review after f79bb10 + 1966cde)

• Read latest commit diffs (Organize new eval scenario groups, Tighten UGC virality delivery gate wording)
• Re-verify version sync, JSON manifests, cross-references
• Re-check eval scenario category enum + counts
• Post review feedback

Re-reviewed the two commits since the last pass. Both are clean, and they resolve the last outstanding minor note. No new findings — this is ready to merge.

1966cde — 70-score wording, now canonical. The residual softness I flagged earlier is gone. ugc-video-ad.md:50 Banned line now reads "Do not deliver below a 70 virality score; tell the user the score, apply the weak-criterion fixes, and re-score before delivery," which matches step 10 and the video-ad-qa.md:34-36 table (70–84 deliver / 55–69 rework / <55 discard). One voice across all three touchpoints now.

f79bb10 — eval scenario grouping, counts all correct. The six new scenarios are split into version-labeled group headers, and every group count is accurate:

• Scripted UGC video ads 0.6.0 (4) → scenarios 41–44 ✅
• Cinematic product ads 0.6.1 (2) → scenarios 45–46 ✅
• All 12 group headers sum to 46, and scenarios.md:3 reads 46 scenarios. grep -c '^### [0-9]' → 46. Matches.

Runner safety confirmed. The new ## … group headers and the surrounding --- separators are inert to run.sh: the awk parser only opens a scenario on /^### [0-9]+\./ and only closes it on the next scenario or ## Format spec. Intervening group headers match no rule, so they can't split or corrupt a scenario block. Categories on 41–46 all stay inside the documented enum (routing | edge_case | refusal | vision) — the earlier guardrail issue remains resolved.

Re-verified clean (unchanged by these commits, spot-checked):

• Version sync — VERSION, krea-generate/VERSION, package.json, all three plugin manifests, marketplace.json, and all four SKILL.md frontmatters aligned at 0.6.1. validate.yml version-sync should pass.
• Cross-references — the ugc/cinematic workflow ↔ reference links verified in the prior pass still resolve; no paths changed in these commits.

Nice, tidy landing — the scenario grouping makes the eval suite much easier to scan, and the delivery-gate wording is now unambiguous.
· albert/add-video-marketing