Marketing Team Tracker

Marketing team tracker

Keep a living roster of the marketing teams worth stealing from, and harvest their best plays on demand. Each play is classified against five repeatable patterns, given a one-line "steal this", and filed in an append-only swipe file your content, ideation, and social skills can pull from. Knowledge type: temporal-signal-brief (per .claude/rules/ontology.md); maturity: emergent — each scan is fresh and time-bound, briefs are not locked.

Seeded from Tom Orbach's "Marketing Ideas" roundup (the 9 teams he follows + the five patterns they share), cite-only — same precedent as .claude/rules/linkedin-cold-dm-doctrine.md. The roster + swipe file are ours; the seed is attributed in the premium reference.

Cousin to /signal-scan — that skill answers "what happened with one topic in N days." This one carries a standing roster of marketing teams, classifies what it finds against a marketing-play taxonomy, and accumulates a swipe file across runs. Distinct from /marketing-ideas (ideation from a brief — the tracker feeds it raw plays) and /competitor-research (a deep static dossier on a threat, not an inspiration source).

When to run

Invoke for: track marketing teams, marketing inspiration scan, marketing swipe file, what's {company} doing in marketing, GTME Pulse sourcing, pre-content-calendar inspiration, add {company} to the tracker.

Do NOT invoke for:

• A deep competitor threat dossier → /competitor-research (static, 13-dimension, adversarial).
• Per-account buying signals / timing triggers → /niche-signal (outbound, "when to reach out").
• Ideation from a brief → /marketing-ideas (this skill feeds it real plays to riff on).
• One topic's last-N-days news → /signal-scan (topic-agnostic, no standing roster).

Brain-first (mandatory): before any external call, run the .claude/rules/brain-first-lookup.md ladder — /recall {team} + grep the swipe file. A play you logged last cycle may already answer the question. Annotate the brief if you went external after a miss.

Modes + flags

Inputs

The roster is the input — no required upstream skill. Recommended: /signal-scan (its resolved entities for a team transfer straight into the roster).

Read first, every run (brain-first):

• projects/research/marketing-team-tracker/roster/canonical-roster.md — the tracked teams + per-team surfaces + last_scanned.
• projects/research/marketing-team-tracker/swipe-file/{team}.md — what's already logged (so you don't re-surface it).
• projects/research/marketing-team-tracker/marketing-team-strategy.md — the cross-cycle scratchpad (decisions locked, things to stop doing, open questions). Per .claude/rules/iterative-strategy-scratchpad.md.

The scan loop

Six steps. Schemas in the premium reference. Surface→tool mapping in the premium reference.

1. Read state (brain-first). Load the roster, the per-team swipe files, and the scratchpad. Apply locked decisions; suppress named anti-patterns. Only scan for the gap since last_scanned.
2. Resolve each team's surfaces. From the roster: the named operator's LinkedIn + the company page + blog/changelog + X handle. Resolve before searching — this is the step that turns a brand-name keyword dump into actual plays (borrowed from /signal-scan). If a surface isn't in the roster, resolve it once and write it back.
3. Scan since last_scanned. Per team, sweep the resolved surfaces for marketing moves (campaigns, launches, stunts, positioning shifts, content series, community plays) shipped inside the --days window. Engagement-ranked, recency-decayed. Free discovery before metered extraction (crawl-cost-discipline.md); gate paid Apify (apify-credits.md).
4. Extract → classify → steal-this. For each new play: name it, classify it against the five patterns (the premium reference), write a one-line "steal this" takeaway, and tag confidence + source URL + access date. No quote, no metric you can't cite → mark [Not available], never invent (per evidence-bound-outputs.md).
5. Cross-team cluster. Note where teams converge on the same pattern this cycle ("three teams ran a pick-a-fight play this month") — that convergence is the signal worth surfacing to Pulse.
6. Emit + persist. Write the dated brief; append new plays to each swipe-file/{team}.md; bump each team's last_scanned; update the scratchpad with anything that should carry forward. Apply the thin-input guard before you ship.

The refresh discipline mirrors /competitor-research: read the existing swipe file first, compare don't append-blind, date-stamp a "Recent changes" line, no-change = no entry.

The five patterns

Every play classifies into one (occasionally two). Full rubric + worked examples in the premium reference.

Thin-input guard

If a scan surfaces fewer than 3 fresh plays worth stealing across the whole roster, say so — ship a short "thin cycle" note naming what was checked, rather than padding the brief with weak plays. Marketing teams don't ship a stealable move every 30 days; an honest dry note beats a manufactured one (pattern from /niche-signal's thin-input rule).

Credit gate

Default surfaces — Exa news + Firecrawl blog/LinkedIn-page/changelog — incur no Apify spend. --x runs through an Apify actor and is off by default: when on, follow .claude/rules/apify-credits.md — fetch-actor-details (free) first, estimate cost, gate before call-actor (soft <$5, hard ≥$5). Probe one team before fanning out across the roster (goal-driven-loops.md).

Quality gate (binary, before declaring done)

• Brain-first ladder run before any external call; brief annotated if it went external after a miss.
• Each team's surfaces resolved (operator + company page + blog + X handle named in the roster, or "none found").
• Every play carries pattern + steal-this + source URL + access date; no invented engagement numbers (mark [Not available]).
• Date window respected — nothing older than --days N as a fresh play (older context flagged as background).
• Thin-input guard applied — < 3 fresh plays ships an honest dry note, not padding.
• Cross-team pattern watch present when 2+ teams converge.
• Scratchpad + last_scanned updated; new surfaces written back to the roster.
• --x / paid calls: cost estimated + gated.

Composition