signal-monitor

name	signal-monitor
version	1.0.0
description	Tracks buying-signal triggers (open job postings, new ad launches, tech-stack changes, recent posts) across a target account list and produces a weekly trigger digest with prioritized outreach hooks. Designed to be scheduled. Uses linkedin_job_search, ad_search, web_tech_stack, post_keyword_search, and profile_activities. Different from competitive-intel (which is one-shot per competitor) and account-research (which is a deep one-pager) — this is recurring, multi-account, and trigger-focused. Proactively invoke on phrases like "monitor these accounts", "buying signals", "hiring intent", "trigger digest", "watch for new ads", "alert me when X starts hiring".
benefits-from	["richapi-gtm","account-research","competitive-intel","scheduled-workflow","evidence-score"]
allowed-tools	["Bash","Read","Write","Edit","AskUserQuestion"]
triggers	["monitor these accounts","watch this list for buying signals","hiring intent","trigger digest","alert me when <company> starts hiring","watch for new ads from","account signal monitoring","ABM trigger system","intent signals"]

signal-monitor

Multi-account, multi-signal, recurring trigger monitoring. Designed to run on a schedule and surface only the changes since the last sweep — so the user gets a short trigger digest, not a wall of static facts.

Preamble — run first

~/.claude/skills/richapi-gtm-skills/bin/richapi-skills-preflight

If CATALOG_OK: no → stop and run richapi-skills-sync.

When to use this skill

Signal	Use
One-shot deep brief on ONE company	account-research
Snapshot competitive comparison (you vs them)	competitive-intel
Recurring trigger sweep across N accounts	this skill
"What changed this week?" framing	this skill
Monitor a single rep, post, or news event	use raw MCP tools — overkill for one row

The defining feature is recurrence + delta. If the user says "do it once," route to account-research or competitive-intel. If they say "every week" / "every Monday" / "tell me when," it's this skill.

Phase 0 — define the watchlist + signal set

Two inputs required. Ask for both upfront if unclear.

a. The watchlist

Either:

• A list of company names / domains / LinkedIn URLs (1-100 ideal; >100 needs strong cost confirm)
• A saved CSV the user references
• An ICP description that resolves to a list (route through build-prospect-list FIRST, then come back)

State the count and confirm: "Monitoring {{N}} accounts. OK?"

b. The signal set

Pick the subset the user actually cares about. Each is independently costed and individually toggleable:

Signal	Tool	Default cost / account / sweep
Hiring — new open roles	linkedin_job_search (scoped per company)	~0.4–2.0 (depends on hiring volume)
Ad launches — new LinkedIn ads	ad_search (advertiser=company, dateFrom=last_sweep)	~0.4–2.0
Tech-stack changes — new pixels/SaaS	web_tech_stack (compare snapshot)	~1.0
Content velocity — new posts	post_keyword_search or profile_activities on key execs	~0.5–2.0
Funding / news — press releases	search_bing or web_scrape of newsroom	~0.5

Default starter set: hiring + ads (cheapest, highest signal-to-noise for B2B SaaS). Add tech-stack only when product-led signals matter. Add content/news on explicit ask.

State per-sweep cost estimate before the first run:

"Per-sweep cost: ~{{N * sum_signal_costs}} credits ({{N}} accounts × {{signals}}). Weekly = ~{{4 * cost}} credits/month. OK?"

Phase 1 — establish baseline (first run only)

If this is the first sweep for this watchlist, the entire surface looks "new" — that's noise, not signal. Two strategies:

1. Cold baseline (recommended): on first run, capture state but mark everything as "baseline — not a trigger." User sees an empty digest with a note: "Baseline established. Real triggers start next sweep."
2. Lookback baseline: on first run, scope each signal to the last 7d (dateFrom: 7d ago) and treat anything in that window as "recent triggers worth knowing." Use this when the user wants signal now.

Ask which they want. Default to lookback if they're impatient.

Persist the baseline state to a local file (e.g. ~/.richapi-skills/signal-monitor/<watchlist-id>.json). Each sweep writes a new snapshot; the digest is the diff vs the previous snapshot.

Phase 2 — sweep

Per-account, per-signal, in this order (cheapest → most expensive to keep cost-debt visible):

2a. Hiring sweep — linkedin_job_search

Scope each call to the company:

linkedin_job_search(
  company: "<account>",
  datePosted: "past_week",   # or "past_month" for monthly cadence
  size: 25                   # cap per account
)

Diff vs baseline: surface new jobIds only. For each new role, optionally call linkedin_job_detail (0.2/call) to get the description — but ONLY for roles whose title matches a high-signal pattern the user defined (e.g. "VP Sales", "Head of RevOps"). Don't pull details for every junior role.

Output per account:

• Count of new openings
• Top 3 new role titles
• Inferred trigger type (e.g. "scaling sales = budget for tools", "hiring CRO = strategic shift")

2b. Ad sweep — ad_search

ad_search(
  advertiser: "<account>",
  dateFrom: "<last_sweep_date>",
  size: 50
)

Diff: surface ad IDs not seen in prior snapshot. Run ad_details (0.5/call) on at most the top 2 new ads per account. Extract: creative angle, CTA, audience signals.

Trigger meaning: a company starting LinkedIn ads = active GTM motion + budget + measurable target audience. High intent for vendor outreach.

2c. Tech-stack diff — web_tech_stack

web_tech_stack(domain: "<account>")

Diff vs baseline list of detected tools. Surface:

• Added: new SaaS in the stack → competitive displacement opportunity OR adjacency-sell opportunity (depending on what the user sells).
• Removed: ripped out tool → unhappy customer of that vendor; warm prospect for alternatives.

Trigger meaning depends entirely on what the user sells. Capture context once at watchlist creation: "What tools, when added/removed, would you want to know about?"

2d. Content velocity — post_keyword_search / profile_activities

For each account, optionally tracked exec(s) (capture LinkedIn URLs at watchlist creation). Run profile_activities (1/call) per exec; surface posts since last sweep. Skip silently if they posted nothing — don't pad the digest with low-signal "no activity" rows.

For company-page posts, use post_keyword_search(fromCompany: "<account>") instead.

Phase 3 — score and rank triggers

Not every trigger deserves equal attention. Use a simple weight schema (defaults; user can override):

Trigger	Weight
New CRO/CMO/VP-Sales hire	10
3+ new sales roles in one sweep	8
New LinkedIn ad campaign launched	6
Tech-stack add of a tool the user explicitly tracks	5–10
Tech-stack remove of a tool the user explicitly tracks	5–10
Single new exec post	2
Single new junior role	1

Per account, sum the trigger weights. Sort the digest by total score, descending. Cut the digest at the top N (default 20) — never show 100 rows.

Phase 4 — produce the digest

Single deliverable per sweep. Markdown, scannable, max 1 page. Format:

# Trigger digest — {{watchlist_name}} — {{date}}

## Top movers ({{count}} accounts changed)

### {{account_1}} — score 18
- 🟢 Hiring 4 sales roles this week (incl. "VP of Revenue Operations")
- 🟢 Launched 2 new LinkedIn ads — angle: "compliance-first observability"
- 🟡 Added Salesforce.com to stack (was on HubSpot — 14 days ago)
→ Suggested play: warm outbound to new VP RevOps; reference compliance angle. [evidence-score: 87/100]

### {{account_2}} — score 12
- ...

## Quiet accounts ({{count}})

{{list, no detail — user can drill in if curious}}

## Sweep cost: {{credits}} credits | Next sweep: {{date}}

Always end with cost spent + next sweep date.

Phase 5 — persist + schedule

Write the new snapshot to ~/.richapi-skills/signal-monitor/<watchlist-id>.json (overwrites prior). Optionally also write a dated digest archive (<watchlist-id>/digests/<date>.md) so the user can compare digests over time.

If the user hasn't yet scheduled this, strongly offer it at the end:

"Worth $X/month if I run this every Monday and email you the digest? Set it up via scheduled-workflow."

This skill's value is 80% in the recurrence. One-shot signal monitoring is just account-research with extra steps.

Cost ceiling

Watchlist size	Default sweep cost (hiring + ads only)	Hard cost-gate
≤25 accounts	~25 credits / sweep	50 credits
26–100 accounts	~100 credits / sweep	250 credits — confirm
101–500 accounts	~500+ credits / sweep	always confirm; recommend bucketing into sub-watchlists
>500 accounts	—	refuse without explicit "I know" from the user

Add ~1 credit / account for tech-stack and ~2 credits / account for content if those signals are enabled.

Anti-patterns

• ❌ Running every signal on every account by default. Pick a subset; let the user opt in to more.
• ❌ Skipping the baseline phase and surfacing 500 "new" rows on day 1.
• ❌ Calling ad_details on every ad in every sweep. Cap to top-N per account.
• ❌ Surfacing "no activity" rows. The digest is for movers — silence is fine.
• ❌ Re-running on the same accounts more than weekly without strong signal-density justification. Diminishing returns + cost burn.
• ❌ Treating this as a research deliverable. It's a trigger feed — short, ranked, action-oriented.

Follow-ups to suggest per-trigger

• 🆕 hire of decision-maker → pre-meeting-briefing for the new exec, or enrich-and-verify to source their email.
• 🆕 ad launch → competitive-intel for the full creative + targeting analysis.
• 🆕 tech-stack add → account-research to see if any other ICP signals corroborate.
• 🆕 content surge → sequence-builder with the new post as a personalization anchor.

Offer to schedule this

This skill exists to be scheduled. If the user runs it once and doesn't schedule, ask once at session end:

"Signal monitoring's value compounds when it runs every week. Want me to set up a Monday-morning digest? → scheduled-workflow"

One offer. Drop if declined.