Star Momentum

SKILL.md preview

---
name: Star Momentum
description: Project the date a watched repo crosses its next star milestone — alert only when projected date lands inside the Show HN dispatch window (7-14 days out, Tue/Wed/Thu)
var: ""
tags: [meta, growth]
---
> **${var}** — Optional. Pass `dry-run` to skip the notification (article still writes, log still appends). Pass a positive integer to override the auto-picked milestone (e.g. `var=500`). Empty = audit every watched repo and auto-pick the next un-crossed ladder rung.

Today is ${today}. Convert the last 14 days of `repo-pulse` star-count data into a projected date for the next milestone crossing, and surface a single decision-ready alert: when should the operator dispatch `show-hn` so the launch lands at the milestone moment?

The skill answers a question `star-milestone` and `repo-pulse` cannot. `star-milestone` celebrates a crossing **after** it happens. `repo-pulse` reports today's deltas. Neither tells the operator "the next milestone is on a Wednesday 9 days from now — that's the launch slot." Without that lead-time signal, the milestone passes reactively, and a dispatch-ready Show HN draft sits unused while the moment slips by.

## Why this exists

`show-hn` shipped May 1 (PR #151) as a `workflow_dispatch` skill. Its launch checklist requires Tue–Thu morning timing. Today there is no signal that tells the operator "the milestone you wanted to anchor the launch around is 9 days out, on a Wednesday — dispatch now so the post is ready Tuesday." This skill provides that signal — and only that signal. It is silent on every other day. The dispatch decision still belongs to the operator; this skill just makes the timing legible.

## Config

No new secrets. No new env vars. No new state file beyond `memory/topics/star-momentum-state.json` for prior-run dedup. No outbound HTTP — pure local file I/O over `memory/logs/` and `memory/topics/`.

Reads:
- `memory/watched-repos.md` — repos to track. Skip lines containing `aeon-agent` or ending in `-aeon` (agent repos, not project repos).
- `memory/logs/YYYY-MM-DD.md` for the last 14 days — extract the `**owner/repo**: stargazers_count=N, forks_count=M` lines that `repo-pulse` writes under its `## Repo Pulse` blocks.
- Optional fallback: `articles/repo-pulse-*.md` if any fork writes them — same regex applies. Logs are the source of truth on the canonical instance.
- `memory/topics/star-momentum-state.json` — prior-run dedup state.

Writes:
- `articles/star-momentum-${today}.md` — the per-repo projection report (always written, even when no alert fires).
- `memory/topics/star-momentum-state.json` — last-alert timestamp per `(repo, target_milestone)` pair.
- `memory/logs/${today}.md` — log block.

## Milestone ladder

```
50, 100, 150, 200, 250, 300, 400, 500, 750, 1000, 1500, 2000, 3000, 5000, 7500, 10000, 15000, 25000, 50000, 100000
```

Same ladder as `star-milestone` so the two skills agree on which numbers count as round-number moments worth marking.

## Steps

### 1. Parse var

- If `${var}` matches `^dry-run` → `MODE=dry-run`. Strip the prefix; remainder is treated as a milestone override.
- Otherwise `MODE=execute`.
- If the remaining var is a positive integer → `OVERRIDE_MILESTONE=$var`. Otherwise `OVERRIDE_MILESTONE=auto`.
- If the remaining var is non-empty and non-numeric → log `STAR_MOMENTUM_BAD_VAR: ${var}` and exit (no notify, no article).

### 2. Load repos

```bash
mkdir -p memory/topics articles
[ -f memory/topics/star-momentum-state.json ] || echo '{"last_run_at":null,"alerts":{}}' > memory/topics/star-momentum-state.json

REPOS=$(grep -oE '^- [a-zA-Z0-9._-]+/[a-zA-Z0-9._-]+' memory/watched-repos.md \
  | sed 's/^- //' \
  | grep -vE '(aeon-agent|-aeon$)' || true)
```

If `REPOS` is empty, log `STAR_MOMENTUM_NO_REPOS` and exit cleanly without notifying.

### 3. Per-repo: build the 14-day stargazer series

For each repo in `REPOS`:

```bash
SERIES=""
for D in $(seq 13 -1 0); do
  DATE=$(date -u -d "${today} - ${D} days" +%Y-%m-%d 2>/dev/nu

…