blog-ideas

name	blog-ideas
description	Daily scheduled brainstorm (also runnable on demand via /blog-ideas) that reviews the week-so-far across ingested activity, the daily journal, and the blog archive, and maintains a living weekly page of blog-post ideas.
effort	strong
required-skills	["vault"]
user-invocable	true
context	inline

Blog-idea brainstorm

Once a day, review the week so far across everything I've been reading, writing, and thinking, and maintain a single living page of blog-post ideas for the week. The page accumulates and sharpens across the week; you rewrite it in place each run. Write only within agent/ (the living page is the only thing you write); read broadly across the vault as Phase 2 enumerates — journals/, blog/, and agent/pages/ are all in scope for reading.

Phase 1: Orient

1. Call blog_ideas_week to get this week's week key, days_so_far, and the page_path. NEVER compute the ISO week yourself.
2. vault_read the page_path. If it exists, this is the living page you'll refine — it's also your within-week dedup memory; note its updated: date (the watermark for incremental reads in Phase 2). If it doesn't exist yet, this is the week's first run and you'll create it.
3. Call blog_ideas_week with offset_weeks=-1 and offset_weeks=-2 and vault_read those pages if present — just enough to notice themes I keep circling week to week.
4. vault_read agent/pages/blog-ideas/parking-lot.md if it exists — the rolling set of seeds from the last ~month that haven't ripened. Consider these alongside this week's material: a parked seed that connects to something new can finally graduate into a headliner.

Phase 2: Gather

The goal is to read everything from this week, not to search for the words "blog ideas." vault_recent is the right tool — it lists what changed by date. Do NOT use vault_search to enumerate: it requires a content query and only returns pages containing that exact text, so it misses almost everything (it's for the dedup check in step 4 only).

1. Everything that changed this week — call vault_recent with days=<days_so_far> (from blog_ideas_week). It returns every vault page modified this week, newest first, each with its modified date and source_type: across all my ingest + distilled sources (agent/pages/…: bookmarks, mastodon, youtube, github, podcasts, music, kindle, plus topical pages) and my Obsidian daily journal (journals/…, source_type=user). Ignore agent/pages/notifications/… (system noise) and the agent/pages/blog-ideas/… pages themselves.
2. vault_read the entries worth considering. ALL of it is fair game: ingested bookmarks/posts/podcasts/videos/repos, distilled topical pages (source_type=page), and my daily-journal notes (source_type=user — TIL bullets, session notes, offhand musings; *(blog candidate)* is a strong signal but not the only material). These pages are already distilled summaries — read them directly, don't re-fetch source URLs.
   • Read incrementally. On the week's first run (no existing page), read everything dated on/after monday. On later runs, only vault_read entries modified after the page's updated: date — the living page already captures everything earlier this week, so trust it and merge the new findings in. This keeps daily runs cheap instead of re-reading the whole week every time.
3. Stalled drafts — vault_list folder=blog/drafts and vault_read any whose title relates to this week's themes. When fresh material supplies the missing piece for an abandoned draft, that's a high-value idea.
4. Published archive (on demand) — here, and only here, use vault_search with a topic query against blog/daily and blog/weeknotes to check a candidate: skip near-duplicates of things I've already written; frame genuine extensions as follow-ups; keep angles in my actual voice.

Keep reads focused: enumerate broadly, but only vault_read this week's entries plus the handful of drafts/posts tied to an idea you're actively developing.

Phase 3: Synthesize

• Treat journal notes and ingested activity as EQUAL peers — an idea may spring from a striking bookmark with no journal note behind it, or from a half-thought in the journal.
• Connect across sources: a journal note + a bookmark + a stalled draft can be one idea.
• Hold a high bar for headliners; everything else is a seed.

Phase 4: Write the living page

vault_write the page_path with this structure (rewrite in place — refine and merge existing entries, add new ones only if they clear the bar, demote faded ones; do NOT just append):

---
tags: [blog-ideas, brainstorm]
summary: Blog post ideas for the week of <monday> (<week>)
week: <week>
updated: <today's date>
---

# Blog ideas — week of <monday>

## Headliners
### <hook title>
- **angle:** why this, why now
- **sources:** [[page]], journal <date>, connects to draft "<name>"
- **shape:** outline bullets / the missing piece
- **status:** new | sharpened <date> | follow-up to <published post>

## Seeds
- <one-liner> — source hint

## Recurring
- <theme> — circled ~N weeks running; maybe it's ripe

Keep 3–5 headliners, ordered strongest first — headliner #1 is the week's lead (a downstream consumer like the weeknotes composer treats it that way). Surface cross-week repeats under Recurring as signal — do NOT re-pitch them cold in Headliners.

Keep this page structure stable and parseable: the section headings (Headliners / Seeds / Recurring) and the headliner field set (hook title, angle, sources, shape, status) are a contract other skills read — don't rename or drop them.

Phase 5: Tend the parking lot

The parking lot (agent/pages/blog-ideas/parking-lot.md, under agent/ so you can write it) is how slow-burning themes survive — ideas that develop across notes and bookmarks over a week or two before they're clearly worth writing. A single week rarely makes a trend obvious; this is where evidence accumulates. After updating the weekly page, vault_write the parking lot:

• Reinforce: if this week's material adds evidence to a parked theme, append the new sources and bump its reinforced: date. Accumulating evidence is the signal it's ripening.
• Graduate: when a parked theme has gathered enough to be worth writing, promote it to a Headliner on this week's page and remove it from the lot.
• Add: park this week's genuinely-new seeds that didn't become headliners, each tagged seen: <today>.
• Prune: drop entries that haven't gained traction in ~1 month (≈4 weeks since seen/reinforced) — if it hasn't ripened in a month, let it go. If one is still actively accumulating at the month mark, keep it and flag it as a persistent slow-burn rather than dropping it silently.

Entry shape (keep the page a simple list under a short header):

- <theme / seed one-liner> — sources: [[a]], [[b]], journal <date>
  (seen <YYYY-MM-DD>, reinforced <YYYY-MM-DD>)

Finishing up

End with a short narrative summary that doubles as the newsletter's raw material: the headliner titles, a one-line hook each, and a link to the full [[<page_path without the .md>]]. If there was nothing new worth surfacing this run, begin your summary with HEARTBEAT_OK on its own line followed by a brief quiet-cycle note.