buying-signal-detector

---
name: buying-signal-detector
description: Score and stack dated public buying signals and return who to contact this week with a ready message; use when choosing this week's contacts.
---

# Contact the companies ready to buy this week

Ranked accounts with stacked live signals: reason, owner, deadline and a message leading with their problem.

## Claude does

Inputs: {COMPANIES}, {OFFER}, {BUYER_ROLE}, {ICP}, {TODAY}; optional {WINS} (what preceded past wins), {PROOF} (named customer results).

1. Build {POINTS}: 5-10 types as points/expiry/window days after the event. Defaults: demo 100/7/0-7, pricing enquiry 80/7/0-7, new executive 75/90/14-45, funding 45/60/8-56, job posting 40/30/0-30, tech change ?/45/7-14, expansion ?/60/0-60, complaint ?/30/0-30; ? is unpublished: set 20-45, say so. {WINS}: keep only types preceding wins.
2. Agree {BANDS} with the user first: score range, owner, deadline. Defaults: 150+ closer 1h, 100-149 SDR 24h, 50-99 nurture 72h, 20-49 marketing, 0-19 monitor.
3. Score ICP fit 1-5 before research; fit 1-2 goes to {NO_SIGNAL} as "fit too low".
4. Research the rest (LinkedIn Jobs, Crunchbase, press, BuiltWith, company site). Job posts: pain, not title; a named tool means migration; 5+ roles per department is a surge, 1-3 nurture.
5. One row per event: Company, Fit, Type, Evidence (quoted), Source URL, Date, lead yes/no. No date or URL, no row; never guess. Layoffs, litigation, misfortune: lead no. Rows: {SIGNALS}; the rest: {NO_SIGNAL}.

## Then Codex does

Codex gets the rows and both tables for the arithmetic and writing: ageing, expiry, multipliers, stacking, tiers, bands, contact dates, messages, three files.

```
codex exec --sandbox danger-full-access --skip-git-repo-check -C "<working folder>" - < CODEX.md
```

First copy CODEX.md into the working folder with {OFFER}, {BUYER_ROLE}, {TODAY}, {POINTS}, {BANDS}, {PROOF}, {SIGNALS}, {NO_SIGNAL} filled in.

## Claude checks

1. Every supplied row present and unchanged; nothing added or invented.
2. Recompute three account scores; funding windows open from day 8, new-executive windows inside days 14-45.
3. Messages only on two or more live signals or one live enquiry; none for one posting or raise, none on lead-no rows.
4. No "I saw", "I noticed", "congrat", greeting or sign-off; every message 75-125 words; every ranked account shows reason, owner, deadline.

Any failure: send Codex the rows and reason, rerun, recheck; never patch by hand.

## Rules

- A signal is a dated, sourced public event, never a hunch. Free public sources only. Never lead with layoffs, litigation or misfortune.
- One signal is a guess; fit gates the signal. No trigger, no window: unchanged accounts close under 1%.
- Each type has its own expiry and window. Skip week one after a raise. Contact within 48 hours; half the value goes by day 7.
- Never open with "I saw", "I noticed" or congratulations; lead with the implied problem.
- Every band needs an owner and deadline before detection starts. Keep 5-10 types, print the reason beside every score, recalibrate monthly from wins.
- Vendor reply rates and "74% when first in" are unverified.

## If Codex is not installed

Claude does the Codex half and says so.

## Built from

- ColdIQ Signal Sourcer: https://github.com/Cold-IQ/ColdIQ-s-GTM-Skills/blob/main/master-skills/signal-sourcer/SKILL.md (269 stars, api.github.com)
- Clay buying-signals guide: https://www.clay.com/guides/how-to-identify-buying-signals (May 2026)
- louisblythe trigger-event-detection: https://github.com/louisblythe/Sales-Skills/blob/main/skills/trigger-event-detection/SKILL.md (143 stars, api.github.com)
- Common Room prompts: https://www.commonroom.io/prompts/ (signal platform)
- Elias, SHiFT!: https://shiftselling.com/files/thebook/SHiFT-The_Trigger_Event_Selling_Book-Preview_Chapter.pdf (2010 book)
- GTME.cc Signal Engine: https://gtme.cc/playbooks/signal-based-icp (July 2026)

# Contact the companies ready to buy this week - Codex task

Work alone from this file: no web search, no added signals, no changed facts.

## You are given

- {OFFER}: what is sold, which problem it removes.
- {BUYER_ROLE}: the message's reader.
- {TODAY}: YYYY-MM-DD.
- {POINTS}: per type, points, expiry days, window open and close days after the event.
- {BANDS}: score range, owner, deadline.
- {PROOF}: named customer results, may be empty.
- {SIGNALS}: one per event: Company, Fit 1-5, Type, Evidence (quoted), Source URL, Date, lead yes/no.
- {NO_SIGNAL}: companies with nothing found.

## Produce

Folder `buying-signals/` with three files.

1. `signals.csv` (one per signal): Company, Fit, Type, Evidence, Source URL, Date, Age, Status, Multiplier, Points, Score, Window opens, Window closes.
   - Age = {TODAY} minus Date; Status = expired if Age exceeds expiry, note only if lead no, else live.
   - Multiplier by Age: 0-1 days 1.5, 2-7 1.2, 8-14 1.0, 15-30 0.7, 31+ 0.3.
   - Score = Points*Multiplier*Fit/5 (expired and note-only rows 0); windows = Date plus window days.
2. `accounts.csv` (one per company): Company, Fit, Live signals, Tier, Score, Band, Owner, Deadline, Reason, Contact from, Contact by.
   - Score = sum of live scores; Band, Owner, Deadline from {BANDS}.
   - Tier A: two or more live signals, or one live demo/pricing enquiry; B: one; C: none.
   - Reason: one sentence naming each live event and its age, never the number alone.
   - Contact from = {TODAY} if a live window is open, else the earliest future opening; Contact by = two days later.
3. `report.md`: "Contact this week" (Tier A by score, row then message), "Nurture" (Tier B), "Expired or note only" (rows and why), "No signal" ({NO_SIGNAL} unchanged).

Message (Tier A only, 75-125 words, to {BUYER_ROLE}), four parts: the problem the live events imply, never naming the event; one sentence on how {OFFER} removes it; a {PROOF} result, or without {PROOF} a question on how they handle it now; a light close (new executive: ask about their first 90 days). No greeting or sign-off.

## Rules

- Copy Company, Evidence, Source URL and Date exactly; never shorten, merge, or invent a customer, number or name; a row missing Date or URL is left out and reported.
- No "I saw", "I noticed", "congratulations" or "reaching out"; never mention a note-only event.

## Return

1. Counts: rows received, live, expired, note only, left out (with reasons); accounts per tier; messages.
2. Top three accounts: each live signal's Age, Multiplier, Fit, Score, then the sum and band.
3. Each message's word count and a search for "I saw", "I noticed", "congrat", "Hi ", "Best," (expect none).
4. Paths of the three files.