mindful

name	mindful
description	Proactively read the latest Daily Harness Analysis digest and connect the few ideas relevant to what the user is working on now to the task at hand, each with a ready way to try it.
license	Apache-2.0
metadata	{"author":"nmhjklnm","version":"1.0"}

mindful

Read your Daily Harness Analysis digest for the user and proactively apply it to what they're doing. The point is not to print the news — it's to notice the one or two ideas, tools, or papers in the latest issues that could improve the task at hand, and say concretely how to use them.

Set your site URL first. This skill talks to your deployed digest. Replace $SITE below with the base URL you configured as SITE_DOMAIN in the repo .env (e.g. https://digest.example.com, or http://localhost:8000 when previewing locally). Everything below assumes $SITE resolves to it.

Be proactive — this is the whole job

Don't wait to be told what to look for, and don't dump a list:

1. Figure out the current work first. Before fetching anything, form a 1–2 line "current focus" from the live context: the cwd, recent/open files, the git diff, what the user has been asking in this session. (If there is genuinely no signal, fall back to a plain digest of the latest issue — see "代读 mode" below — rather than asking.)
2. Connect, don't catalog. For each idea you surface, lead with why it's relevant to their work and how they'd apply it here — then the action. "You're building a memory store; DeltaMem's residual-tree dedup is the same problem — git clone …" beats "Here are today's memory papers."
3. Be selective. 2–5 items, best first. If nothing is strongly relevant, say so in one line and surface the single most interesting thing instead of padding.

How to fetch (progressive disclosure — don't over-pull)

The site is static; you are the search.

1. GET $SITE/ideas/index.json — a small manifest: latest, all dates, topics (id + label + count), kinds, and a date_url template. No item bodies; it just tells you what exists.
2. GET the date slice(s) you need via date_url, default latest: $SITE/ideas/date/<date>.json → that issue's full items, each {title, topic, topic_label, kind, summary, source_url, use, page_url}. Widen to a date range only when the user asks for history or a keyword/topic needs it (fetch a few days, not everything).
3. Filter locally — default date = latest, default topic = all. Match the user's focus / args against title + summary + topic + kind.

Args (compose freely)

• /mindful — latest issue, all topics, ranked by relevance to current work.
• /mindful <keyword> — keyword filter (latest by default; fetch more dates for history).
• /mindful <topic> — map the word to a topics[].id/label from the manifest, filter to it.
• /mindful <date | range> — fetch those date slice(s).
• mix them: /mindful memory this week, /mindful safety repo.

Output

• One-line lead: e.g. 3 items today relate to your <X>: (or, if weak, nothing strongly relevant, but this one's worth a look:).
• Per item, tight: title — one line on why it fits their work · use (the ready command, e.g. git clone … / read <url>) · source_url.
• That's it. This output replaces the user reading the site themselves.

代读 mode (no clear current task)

If there's no usable signal about what they're working on, just proactively digest the latest issue: the 3–5 most notable items with a one-line take and each item's use. Don't ask "what are you working on?" — read it for them.

Rules

• Read-only. Surface the use command; never auto-git clone/install.
• Don't fetch every date slice "to be safe" — manifest first, then the minimum dates. Default is just latest.
• Always include each item's real source_url so the user can go deeper.