AskUserQuestion

I turn your social presence into a daily For You feed: I build a profile from your own channels, track the competitors you pick, and surface what's performing as content ideas backed by real engagement. I just need a ScrapeCreators API key (one key covers all four platforms; 100 free calls, no card).

AskUserQuestion

• Open scrapecreators.com to grab a free key
• I'll paste a key now
• Skip for now

open https://scrapecreators.com

~/.config/content/.env

SCRAPECREATORS_API_KEY={key}
SETUP_COMPLETE=true

SETUP_COMPLETE=true

AskUserQuestion

@me

{platform: [handle]}

--days 90

python3 "$SKILL_DIR/scripts/scrape.py" \
  '{"x": ["me"], "youtube": ["@mychannel"]}' \
  --pillars "" --days 90

• Niche, Audience, Voice Notes — infer from recurring topics, framing, tone.
• Content Pillars — the 3–5 themes their posts actually cluster into. These drive

  --pillars

  on every future run, so get them right.
• My Social Profiles — handle, follower count, bio, and a one-line content- style note per platform, taken from the scrape.
• Target Platforms / Research Channels — the platforms they're active on.
• Search Terms — concrete keywords from their top topics.

AskUserQuestion

• Content Goal — why they post (lead gen / awareness / growth / thought leadership / selling…), where they drive traffic, and what they're promoting.
• Pillar confirmation — show the 3–5 pillars you inferred and let them edit or confirm before writing.

brand/profile.md

FILE-SCHEMAS.md

brand/my-content.md

brand/profile.md

$CONTENT_HOME/research/

YYYY-MM-DD

feedback.json

reviews[]

• "▲ on three contrarian takes in the user's niche" → "gravitates toward contrarian takes"; "▼ on listicles" → "listicle formats don't land." A note often states the reason directly — use it.
• Record these to your project memory (the auto-memory you maintain) as the user's content taste — the same place 1b recalls from. Update an existing taste note rather than duplicating it; let a single ▼ inform, not override, an established preference. Don't record one-off reactions with no pattern, anything already obvious from

  brand/profile.md

  , or post/run specifics (those live in

  research/

  ). Taste only.

feedback.json

feedback.json

• brand/profile.md

  — niche, pillars, search terms, content goal, audience

• brand/tracked-accounts/*.md

  — tracked creators per platform

• brand/my-content.md

  — the user's own content performance + audience requests

brand/my-content.md

$CONTENT_HOME/research/

YYYY-MM-DD

--since

--since

--days

--since

$CONTENT_HOME/research/{today}/

$CONTENT_HOME/research/{today}/feed-data.json

• Refresh — re-pull and rebuild (reuse the same

  --since

  /

  --days

  )
• Expand — widen the window: drop

  --since

  and/or raise

  --days

  (keep the feed within ~30 days) when the user wants more than the last week
• View — just (re)open the existing feed (skip to Step 6)

FEED_DATA

FILE-SCHEMAS.md

feed-data.json

posts[]

• Tracked-account posts — every post from tracked accounts (no engagement gate). Set

  performance

  /

  performanceDirection

  vs the account baseline (e.g.

  "+210% vs baseline"

  ,

  "up"

  ).
• Discovered niche outliers — statistical outliers (

  outlier: true

  , z-score 2+, or baseline 2x+). Set

  zScore

  and a

  why

  line.
  Per post, regardless of source, provide: a 1–3 sentence

  text

  summary,

  url

  ,

  handle

  +

  displayName

  (creator filter),

  platform

  , an

  engagement

  object, a hook callout when notable, and the two fields that make the feed work —

  timestamp

  (ISO 8601, drives Recent sort + relative time) and

  sortValue

  (numeric total engagement/reach, drives the default Popular sort). A post is flagged as an outlier (intensity-scaled badge + accent bar) whenever it has a

  zScore

  or

  performanceDirection: "up"

  — so a tracked post that beat its baseline shows as an outlier too.

references/content-strategy.md

• Make YOUR version, never repackage. A good angle answers at least one of: what do you know the original creator doesn't (expertise), what have you done the audience hasn't seen (access), or where do you disagree (contrarian)?
• Anti-cannibalization. When

  brand/my-content.md

  exists, don't re-pitch a topic the user already covered unless the angle has a genuine differentiator (more depth, different format, an update, a response to feedback). Note prior coverage explicitly.
• Own-audience demand wins. Requests from the user's own audience (

  brand/my-content.md

  ) outrank competitor signals — foreground them in the brief's "why now."
• Taste memory biases selection. An idea that aligns with the taste signals you recalled in Step 1 (topics/formats/angles this user gravitates toward) is a stronger pick than one justified by engagement alone — and worth calling out ("this fits a pattern you keep coming back to"). Conversely, deprioritize anything that matches a recorded "doesn't land" signal.

• Reactions / feedback → taste memory. The feed page lets the user mark items (▲ more like this / ▼ less / a note) across both tabs. In server mode these save to

  research/{date}/feedback.json

  automatically as the user clicks; in static mode the user downloads that file into the run folder. The file is just an accumulating list of reactions — no status, no submit step. The next run reads the previous run's

  feedback.json

  at Step 1a and distills it into your project memory so future runs are personalized — there is no taste file; taste lives in auto-memory.
• No API key = no run. Both profile and URL mode require

  SCRAPECREATORS_API_KEY

  — every platform, including YouTube transcripts, goes through ScrapeCreators. If the key is missing, the script returns an error; stop and show setup instructions rather than inventing data.