Daily Briefing

Daily Briefing — Configuration Companion

When to use

Activate when the user wants to read or modify the config files that drive the daily briefing pipeline. The briefing itself is generated entirely by ~/daily-briefing/morning-briefing.sh under user cron — Hermes plays no role in producing it. This skill exists only to handle natural-language management of the policy files that the pipeline reads.

When NOT to use

• Generating a briefing on demand (it's a cron pipeline; just wait for tomorrow, or run the script manually outside Hermes).
• Answering follow-up questions about today's briefing (the briefing is in conversation context; just respond normally without a skill).

Files this skill manages

These files are plain JSON — read with Read, edit with Write (or edit-in-place tools). Always preserve 2-space indent and a trailing newline so diffs stay clean.

Operations

For any operation, the procedure is: read the relevant JSON file, mutate the in-memory dict, write it back. Always preserve the existing formatting so diffs stay clean.

Trusted / untrusted news sources (news-source-prefs.json)

The trusted list is priority-ordered: lower index = higher priority. The untrusted list is unordered.

Domain resolution (<source> → domain): Use general web knowledge: "Reuters" → reuters.com, "NYT" or "New York Times" → nytimes.com, "Bloomberg" → bloomberg.com, etc. Always normalize to lowercase, no www. prefix. If the publisher name is ambiguous (e.g. "Reuters" vs "Reuters UK" → uk.reuters.com), ask the user to clarify before writing.

Selection-rule recap (so you can explain effects of edits):

1. Stories from untrusted domains are dropped entirely.
2. Among surviving candidates, the one with the lowest index in trusted wins.
3. If no candidate's domain is in trusted, the first surviving candidate (search relevance order) wins.
4. If everything was untrusted for a topic, the topic produces no story that day.

So adding to trusted raises a source's priority; adding to untrusted blocks it; reordering inside trusted shifts which trusted source wins on overlapping coverage.

Watched news topics (news-topics.json)

The topics array is the list of keyword strings the news fetcher searches for each morning. Order doesn't affect priority — every topic gets one story slot per day if a survivor exists.

Topics are passed verbatim to the news search; quoted strings (e.g. "AI capex") work as phrase searches.

Watched stock tickers (tickers.json)

The array drives the __Markets__ block in the morning briefing. Each entry is a single object with one required field: ticker, the symbol in Yahoo Finance convention — bare for US listings, with an exchange suffix for international: SU.PA, 7203.T, 0700.HK, RIO.L.

The pipeline's ticker fetcher dispatches by suffix:

• Bare ticker (no dot) → quote API. Reliable and official; covers US listings only on the free tier.
• Suffixed ticker (has a dot) → yfinance via the local Python venv. Covers any market Yahoo Finance does, but unofficial — Yahoo can break the lib until maintainers patch.

Class shares like BRK.B should be entered as BRK-B (Yahoo's hyphen form for share classes) so they route to the quote API instead of being treated as a foreign exchange suffix.

Verify ticker ↔ company

Always web_search to ground the ticker ↔ company mapping before writing the file or naming the company in your response. This applies in both directions:

• User gives a ticker (e.g. "Add GEV"): web_search "GEV stock ticker" → confirm GE Vernova → write tickers.json with GEV → respond Added GEV (GE Vernova) to your watchlist.
• User gives a company name (e.g. "Add GE Vernova" / "Add Schneider Electric"): web_search "GE Vernova stock ticker" → confirm GEV (or SU.PA) → write tickers.json → respond Added GEV (GE Vernova) to your watchlist.

The lookup is mandatory even for tickers you think you know. Your training data may not match current ticker reality, and confidently- asserted-but-wrong company names are the canonical failure mode here. Canonical confusion sources:

• GE family: GE (General Electric, post-spinoff parent), GEV (GE Vernova, energy spin), GEHC (GE Healthcare, separate spin). All three trade independently — none inherits the others' symbol.
• Class shares: GOOG vs GOOGL, BRK-A vs BRK-B — different voting/economic rights, both real tickers.
• Multi-exchange listings: SU.PA (Schneider Electric, Euronext Paris primary) vs SBGSF (US OTC ADR). If the user names the company without specifying exchange, ask which they want before writing.

On user correction, web_search before re-writing. If the user says "you got that wrong, I meant X," do a web_search to confirm X's actual ticker before changing tickers.json. Don't pile a second wrong assertion on the first.

Confirmation format for ticker changes

After adding or replacing, respond with the format Added <TICKER> (<Company Name>) to your watchlist. — both the symbol and the verified company name. After removing, ticker-only is fine (Removed GEV from your watchlist.) since the company isn't in the file.

Calendar organizer mapping (calendar-people.json)

Emails are stored lowercased; the calendar fetcher lowercases before matching, so always lowercase on write.

Verification

Before saving any edit, confirm:

1. The file is one of news-source-prefs.json, news-topics.json, tickers.json, or calendar-people.json under ~/daily-briefing/. Editing anything else through this skill is out of scope.
2. The JSON parses after the edit (no trailing commas, no top-level key drift — only the documented top-level keys).
3. For source-prefs edits: the same domain does not appear in both trusted and untrusted. The latest action wins; remove from the other list.
4. For source-prefs edits where the user named a publisher rather than a domain: the inferred domain matches a well-known mapping, or the user has confirmed.
5. After writing, briefly summarize what changed (the operation, the target file, and the resulting priority order if relevant).

When an operation reports an error

• The config file is missing → the briefing pipeline isn't set up on this machine. Tell the user which file is absent and point them at this skill's README.md for pipeline setup.
• The file doesn't parse as JSON → report that it is malformed and name the file. Leave it as it is; do not rewrite it from memory.
• A web_search for a ticker returns nothing conclusive → say the symbol couldn't be confirmed and ask the user which company they mean. Don't write an unverified ticker.
• The user names a publisher that maps to more than one domain → ask which one before writing.

Always ask the user for guidance when there is an error; do not proactively try to resolve errors yourself.

Empty results

An empty trusted, untrusted, topics, tickers or organizers list means nothing is configured yet — say so plainly ("you aren't tracking any tickers yet") and offer to add the first entry.