Content Asset Triage

Content Asset Triage

Pre-drafting workflow for when the user hands you a folder of AI-generated creative assets and wants social content "about it." The job is to decide WHAT is postable and HOW to frame it — not to draft yet. Drafting lives in draft-post; this skill feeds it.

When to Use

• User points at a folder of generated images/art/UI/comics and wants a post or thread.
• A separate creative/agent profile produced a batch of assets for this profile to ship.
• "Which of these should we use?" / "review what it made."

When NOT to Use

• Single known asset, just needs a caption → draft-post directly.
• Quality-checking a written draft → brand-review-content.

Workflow

1. Confirm the folder before deep-diving. Users keep several project folders side by side (some shelved). If the user says "the folder the profile made," confirm the exact path or search broadly first. Do NOT latch onto the first plausible match and invest a deep pass in it — wrong folder = wasted work + a correction.
2. Read any README/index for orientation only. Then inspect candidates yourself with vision_analyze. A folder's suggested ordering reflects what was easy to generate, not what's strongest — do not trust it as a quality ranking.
3. Sort into three tiers (see references/ai-asset-triage.md for the full rubric):
   • Tier 1 — postable as-is (strong concept, clean render, no embarrassing artifacts)
   • Tier 2 — good but needs a small fix first (almost always broken/garbled TEXT)
   • Tier 3 — not showcase-ready (reads as filter/artifact, not craft)
4. Pick the strategic frame. Uneven batch → build-in-public workflow angle ("one prompt, the agent ran N skills, produced all this"), where variety is the point and rough pieces are evidence not liabilities. Uniformly strong batch → quality showcase. Default to the workflow angle for skill-demo batches. Differentiated > commoditized: When the user wants to spotlight favorites, lean into skills that produce outputs the audience HASN'T seen before (e.g., braille text-to-image, full UI generation from a prompt). Commoditized outputs (AI illustrations, comics, pixel art — things any image generator can do) get a one-line mention, not spotlight treatment. The framing: "any tool can do X, but THIS surprised me."
5. Screenshot HTML/UI assets yourself. Don't rely on the folder having PNGs of HTML files. Use headless Chrome: "/Applications/Google Chrome.app/Contents/MacOS/Google Chrome" --headless --disable-gpu --no-sandbox --window-size=1400,900 --screenshot="OUT.png" "file://SRC.html". See references/rendering-techniques.md for font/rendering pitfalls.
6. Report, then wait for the green light. Triage → recommend → (user approves) → draft → review gate. Never post straight from a triage pass.

Output Shape

Lead with the honest quality read (the tiers). Call out what the folder's own README got wrong. Give the strategic recommendation: which images lead, which frame, which need a fix first. End with options for next step — not a posted artifact.

Pitfalls

1. Don't trust the folder's README ranking. Inspect with your own eyes.
2. Don't lead with the weakest category. Skill-demo folders often open with pixel/ASCII art because it's easy to generate; those are usually Tier 3.
3. Don't frame an uneven batch as a quality showcase. That invites a quality judgment you lose. The workflow story is more honest and more defensible.
4. Don't skip the text check. Garbled/misspelled text is the #1 AI-artifact tell and the thing that gets a post screenshotted and mocked.
5. Don't post from triage. This skill is review-gated by design.
6. Don't assume PNG previews are correct. Braille/ASCII art PNGs often render as tofu boxes due to missing font glyphs. Always check the source .txt and re-render if needed. See references/rendering-techniques.md.
7. Don't over-structure the user's voice. This user writes short, casual, declarative sentences. No "First up:" / "Then:" scaffolding. No product-launch energy. Understated enthusiasm > exclamation marks. Match their actual tweet history, not a template.

Related Skills

• draft-post — execution layer; draft the actual post once triage + frame are approved.
• brand-review-content — voice/quality check on the draft before it ships.
• build-in-public — the strategy layer the workflow-angle reframe comes from.
• xurl — actually posting to X (only after approval).