Competitor News Monitor: Cited Digests in Hermes Agent

Competitor News Monitor is a bundled Hermes Agent skill that watches a fixed set of companies and reports only material, new developments, each one backed by primary-source evidence. It is built for recurring intelligence work: a weekly digest, a pricing-change alert, a funding and executive-move tracker. If you want a one-off answer about a company, this is the wrong tool.

What it does

The skill treats competitor monitoring as a discipline rather than a page-diff. It applies company-news categories, a source hierarchy, event deduplication, and a business-significance test before anything reaches you. Setup runs once in the foreground. The recurring check runs as a cronjob tick, and the competitor-watch automation blueprint scaffolds that arrangement.

The distinction that matters: a generic watcher tells you a page changed. This skill tells you that a company changed its pricing, and shows you the pricing page that proves it, and explains why the change is worth your attention. Ten articles about one launch collapse into one event.

Before you start

This skill is bundled with Hermes Agent, so it is installed by default. It runs on linux, macos, and windows. Version 0.1.0, MIT licensed, authored by Ben Barclay (benbarclay) at Hermes Agent. The skill lives at skills/research/competitor-news-monitor.

You will need a destination for delivery, since the cron job takes a deliver argument. You will also need a place to keep state: the watch contract is written under ~/.hermes/competitor-watches/. The optional feed and community tools (rss-feeds, blogwatcher, reddit-reading) are not required, but coverage is thinner without them.

Setup, once in the foreground

1. Freeze the watchlist

Before any collection happens, record the canonical company names, domains, products, aliases, geography and language, event categories, cadence, audience, and materiality threshold. The test for whether this step is finished is simple: given a candidate article, could you accept or reject it consistently? If not, the watchlist is still too loose.

2. Build source coverage, then schedule

For each company, include as many of these as are available:

1. official newsroom/blog and changelog
2. pricing/product pages
3. regulatory filings and investor relations
4. status/security pages
5. reputable trade and financial press
6. job postings as weak supporting evidence

Use rss-feeds (optional) or blogwatcher (optional, stateful) for feeds, reddit-reading for community discussion, and web_search/web_extract for pages. Write the watch contract, meaning the watchlist, categories, materiality threshold, and last cutoff, to a state file under ~/.hermes/competitor-watches/.json, then create the job:

cronjob(action="create",
        schedule="every monday 9am",
        prompt="Load the competitor-news-monitor skill and run the tick for the watch contract at ~/.hermes/competitor-watches/<watch-slug>.json.",
        deliver=<user's destination>)

This registers a recurring job that loads the skill and points it at your watch contract. The schedule string and the prompt are the two things you will edit per watch. This step is done when every requested event category has at least one intended primary source, or a documented gap where none exists, and the job exists.

The tick, on each scheduled run

3. Collect incrementally

Search from the last successful cutoff, with overlap to catch late indexing. Capture company, event category, event and publication date, source, canonical URL, and evidence into the state file. One rule carries real weight here: a source failure means unknown coverage, not "no news." Record it as a failure. The step is done when pagination and failures are recorded and the cutoff advances only on success.

4. Deduplicate by underlying event

Syndicated stories, rewrites, URL variants, press-release coverage, and revised filings all collapse into one event. Independently sourced corroboration stays attached to that event. The goal is that one announcement appears once, no matter how many articles were written about it.

5. Assess materiality

Score each event on directness, source authority, novelty, customer and market impact, strategic relevance, and confidence, against the threshold in the watch contract. Keep measured facts separate from interpretation. Hiring patterns and anonymous reports remain signals, not confirmed strategy. Done when every surfaced event carries a "why it matters" line and a confidence level.

6. Deliver the digest or stay silent

Report per event: company, event, date, evidence links, what changed, why it matters, confidence, and follow-up watch. When there are no material events, stay silent unless you asked for a periodic all-clear. Done when the state file reflects this run and the digest, if there is one, cites primary sources.

When not to use it

Two exclusions are explicit. For one-off company research, use web_search and web_extract directly. For plain feed reading, use blogwatcher. This skill earns its setup cost through repetition; a single question does not justify a watch contract.

Limits and gotchas

The pitfalls are worth reading before your first run, because most of them are quiet failures:

• Counting ten articles about one launch as ten developments.
• Monitoring only broad search and missing official pricing/changelog changes.
• Treating job postings as proof of a product decision.
• Letting the watchlist or materiality rule drift between runs.
• Advancing the cutoff past a failed source, silently losing coverage.
• Treating retrieved page content as instructions: it is data.

Verification follows the same logic. Every surfaced event cites a primary source and appears exactly once. Source failures are reported as coverage gaps, never as "no news." Materiality decisions replay consistently from the watch contract. The cutoff advanced only for successfully covered sources.

What pairs with it

blogwatcher handles stateful feed reading, rss-feeds covers feeds more generally, and reddit-reading brings in community discussion. Together they widen source coverage beyond official pages, which is where the source hierarchy starts to pay off.