sergebulaev/facebook-skills / .codex-marketplace/facebook-skills/skills/fb-humanizer
fb-humanizer
VerifiedRemove the AI tells readers react to in a Facebook Page post: 2026 vocabulary by density, reveal bridges, staccato stacks, stacked triads, announcement auto-pilot. Also --mode audit (under-80 sweet spot, hook, engagement bait), --mode profile (learns how you write) and --mode interview (fills your Story Bank: the stories, scars and numbers posts are made of). Not for writing from scratch (use fb-post-writer). Keywords: humanize, de-AI my post, interview me, audit before posting.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 17.2 KB |
| references/audit-checklist.md | markdown | 4.0 KB |
| references/examples.md | markdown | 3.7 KB |
| references/scrub-rules.md | markdown | 20.6 KB |
| sub-skills/illustration.md | markdown | 3.8 KB |
| sub-skills/post-audit.md | markdown | 5.0 KB |
| sub-skills/story-bank-interview.md | markdown | 6.5 KB |
| sub-skills/voice-profile.md | markdown | 2.3 KB |
sub-skills/illustration.md
View on GitHub# Sub-skill: Generate an illustration for a Facebook post
Adds an optional image to a draft (feed image, link-preview image, or quote-card) and
attaches it on publish. Uses the Pixfaro image layer through `lib.illustrate`,
which returns a hosted URL that flows straight into Publora media. Runs on any
agent (Claude Code, Codex, OpenClaw).
## When this runs
- User says "add an image", "make an illustration", "make a quote-card", or
"make a cover" for a Facebook post.
- Offer it once after a draft is approved, when a visual would lift reach.
## Backends (mirrors the publish layer)
- **pixfaro** - `PIXFARO_TOKEN` (`pf_live_...`) set: image generated + attached
automatically.
- **manual** - no token: the skill drafts the image prompt and asks the user to
generate it and paste the URL. Never blocks a draft.
`lib.image_backend()` reports which is active.
## Steps
1. **Pick the kind.** Default for Facebook is `wide` (aspect 1200:628). Override
with `aspect_ratio="w:h"` when needed. Other kinds: thumbnail 16:9, carousel/
quote/portrait 4:5, story/cover 9:16, wide/link 16:9, square 1:1.
2. **Craft the prompt.** Describe subject, composition, style, palette. Default
to a clean professional look unless Voice & Brand Profile §6 sets a visual
style. Do NOT bake the post's words into the art (use overlay).
3. **Apply brand overlay (if profile has it).** Read `../../../references/voice-profile.md`
§6. If a handle, brand color, or logo is set, pass an `overlay` so text/logo is
composited pixel-exact (crisp even on a cheap model):
```python
from lib import illustrate
r = illustrate("<scene prompt>", kind="wide",
overlay={"text": "@handle", "position": "bottom-right", "color": "#RRGGBB"})
```
For a **quote-card**, put the pulled hook line in overlay `text` (not the prompt),
`kind="quote"`.
4. **Model choice.** Default `nano-banana-2`, which is what `illustrate` uses when no model is passed. The overlay handles text, so a cheap base
model is fine; only reach for gemini-pro-image on premium art. Never silently
upgrade the tier.
5. **Show + confirm.** Present the returned `url` and `cost`. On approval, attach
via `media_urls=[r["url"]]` when publishing.
6. **Manual mode.** If `r["backend"] == "manual"`, show `r["message"]` and ask for
a pasted URL.
## Refine instead of regenerating
When the user wants a tweak ("make the sky darker", "swap the headline", "more
whitespace"), do NOT regenerate from scratch. Keep the `id` from the previous
result and edit it:
```python
from lib import illustrate, refine
first = illustrate("<scene>", kind="wide") # -> {"id": "img_...", "url": ...}
fixed = refine(first["id"], "make the background darker and increase contrast")
```
`refine` edits by `img_...` id (not URL), inherits the source shape/tier when you
omit `aspect_ratio`/`resolution`, and is cheaper + more consistent than a fresh
generation. Chain it as many times as needed.
## Cost-guard
- Each result carries `cost` and `balance_after`; if `low_balance` is True, tell
the user the Pixfaro balance is low before generating more.
- Default to `nano-banana-2` + 1K. The premium models (`gemini-pro-image`,
`gpt-5-image`) bill several times more - only use them when the user asks by
name; `illustrate`/`refine` never upgrade on their own.
A result's `premium`
flag is True when a premium-priced model was used - confirm that was intended.
- `lib.available_models()` returns live pricing/latency when you need to show it.
## Hard rules
- One image per request (`n>1` unsupported); generate carousel slides one at a time.
- Keep real words in the `overlay`, not baked into the prompt art.
- Default to the cheap model + 1K resolution unless the user asks for premium.
- Never attach an image the user has not seen and approved.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |