Skill Spotlight

---
name: Skill Spotlight
category: social
description: Pick one skill from a rotation, ship a paste-ready feature tweet spotlighting it, then dispatch the picked skill so the chosen channel also gets the live outcome
var: ""
tags: [meta, social, content]
---

> **${var}** — Optional skill name to feature. Overrides the queue. If empty, picks the next uncovered skill from `memory/topics/skill-spotlight.md`.

Today is ${today}. Read `memory/MEMORY.md` for context.
Read `memory/topics/skill-spotlight.md` for the queue, the covered list, and the blocklist.
If `soul/` files exist, read them (`soul/SOUL.md`, `soul/STYLE.md`) to match the operator's voice. If `soul/` is empty or absent, use a clear, terse, neutral voice.

## Goal

A recurring content series around the framework's observable feature surface. Each run produces two notifications:

1. A paste-ready feature tweet (long-form X post) in the **Aeon skill spotlight** format — the operator copies this to X.
2. The picked skill's live outcome — delivered through that skill's own `./notify` after dispatch. The operator screenshots that outcome as the "Result ⤵️" body of the tweet.

Two separate notifications, not a combined message. Each notify stays under platform character limits, and the dispatched skill remains independently observable.

## Steps

### 1. Pick the featured skill

Priority order:

1. If `${var}` is non-empty and `skills/${var}/SKILL.md` exists, pick `${var}`.
2. Else read `memory/topics/skill-spotlight.md`. Pick the first entry under `## Queue` that is not under `## Covered (last 30 days)` and not under `## Blocklist`.
3. If the queue is exhausted, `ls skills/` and pick the first skill that has a `SKILL.md`, is not in the blocklist, and has not been covered in the last 30 days.

If nothing qualifies, send a single notification (`./notify "skill-spotlight: no candidates today — queue exhausted, all skills covered in last 30d"`) and exit. No filler post.

Let `${PICK}` be the chosen skill name (the kebab-case directory name).

### 2. Read the picked skill's SKILL.md

Parse the frontmatter and body:

- `name` (frontmatter) → use for the "▶️ Name" line. If absent, title-case the directory name.
- `description` (frontmatter) → ground-truth one-liner; rewrite in clear, concrete language for the "What is it about?" paragraph.
- `## Steps` body → source for the "How it works behind the scenes" bullets. Extract 5-7 real, concrete behaviors. Do not invent capabilities the skill does not have.

### 3. Compose the tweet draft

Write the following file: `.outputs/skill-spotlight.md`. Use **exactly** this shape:

```
The @aeonframework skill spotlight 🌟

🗓️ {ordinal} {Month}

▶️ Name: {Pretty-Name}

▶️ What is it about?

{2-3 sentence paragraph in operator voice. What it does, why it exists, the wedge it serves. No marketing fluff.}

▶️ How it works behind the scenes

✴️ It {verb}. {One concrete sentence about a real behavior in the SKILL.md.}

✴️ It {verb}. {Another one.}

✴️ It {verb}. {…}

✴️ It {verb}. {…}

✴️ It {verb}. {5-7 bullets total.}

▶️ Demo: {One-line setup of what the live run will show.}

Result ⤵️
```

Notes:

- The header line `The @aeonframework skill spotlight 🌟` is the brand of the series. Forks that want their own handle can edit this one string in the SKILL.md.
- Date formatting: English ordinal (`1st`, `2nd`, `3rd`, `4th`, …, `21st`, `22nd`, `23rd`, `31st`). Month is full name (`June`, not `Jun`). No year.
- Every `✴️` bullet starts with **`It {verb}.`** — keeps the cadence consistent across days and makes the series visually scannable.
- Each bullet must trace to a real behavior in the SKILL.md. If the skill genuinely has only 5 distinct behaviors, ship 5 bullets, not 7.
- No hashtags, no emoji other than what's in the template, no AI-tells ("dive in", "let's", "in summary", "robust", "seamless").
- Prefer concrete nouns (file paths, env vars, CLI flags, API endpoints) over abstractions.
- Don't quote the SKILL.md verbatim — paraphrase each bullet.
- Don't link to the SKILL.md from the tweet. The format keeps the bait — DMs and replies are the conversion signal.

### 4. Send the tweet to Telegram (or configured channels)

```bash
./notify -f .outputs/skill-spotlight.md
```

Use `-f` (not `./notify "$(cat ...)"`) — the post is multi-line, and passing it as argv trips the sandbox ("Unhandled node type: string"), silently dropping the send to `.pending-notify/` (ISS-009). The `-f` flag reads the file inside the script, so argv stays short.

If the file exceeds 4000 characters, trim bullets (drop the weakest first) until it fits. Telegram caps single messages at 4096.

### 5. Dispatch the picked skill (fire-and-forget)

```bash
gh workflow run aeon.yml -f skill="${PICK}"
```

Do not wait or poll. The dispatched skill's own `./notify` will deliver the outcome to the configured channels a few minutes later.

If `gh workflow run` fails (permission denied, rate limit, etc.), send a single follow-up notification and stop:

```bash
./notify "skill-spotlight: tweet drafted for ${PICK} but dispatch failed — run it manually: gh workflow run aeon.yml -f skill=${PICK}"
```

One attempt, one notification on failure. No retry loop.

### 6. Update queue state

Edit `memory/topics/skill-spotlight.md`:

- Append a line under `## Covered (last 30 days)`:
  ```
  - ${today} — ${PICK}
  ```
- If `${PICK}` appears under `## Queue` (i.e. it was picked via the queue, not via `var` override), remove that line from the queue.
- Prune entries older than 30 days from `Covered` so the section stays small.

### 7. Log

Append to `memory/logs/${today}.md`:

```markdown
### skill-spotlight

- Picked: ${PICK} (source: queue | var-override | catalog-fallback)
- Tweet draft: .outputs/skill-spotlight.md
- Dispatched run: <url or "FAILED — reason">
```

## Anti-patterns

- **Padding bullets.** Five real bullets beat seven generic ones.
- **Picking meta/internal skills.** They don't read as features. Keep them in `Blocklist`.
- **Picking the same skill in a 30-day window.** The `Covered` list enforces this; do not bypass it.
- **Dispatch-and-wait.** The dispatched skill is independently observable; polling here just burns runner minutes.
- **Linking the SKILL.md in the tweet.** Keeps the bait — DMs and replies are the conversion signal.
- **Auto-tweeting.** This skill produces a draft; the operator copies to X manually (the screenshot of the outcome goes into the tweet's "Result ⤵️" body).

## Sandbox note

`gh workflow run` works in the sandbox via the gh CLI (auth handled internally). `./notify` fans out to all configured channels. No prefetch needed.