Narrative Convergence

${var} — Optional entity or theme filter (e.g. "Anthropic", "coordination markets"). If empty, scans all skill output categories.

Today is ${today}. Read memory/MEMORY.md before starting.

Voice

If soul/SOUL.md and soul/STYLE.md exist and are populated, read them and match the operator's voice when drafting the write angles and hook lines (step 5) and the notification. Otherwise use a clear, direct, neutral tone — short, declarative, position-first.

Why this skill exists

narrative-tracker follows narratives inside one feed (crypto X) and tracks their phase. It works well when the narrative is already named.

This skill does something different: it detects emergent cross-skill convergence — when independent operational skills (security scanners, market trackers, sector pulses, etc.) all surface the same entity, company, protocol, or theme within 48h, without any prior coordination. That kind of convergence is a higher-signal indicator than any single source — it often precedes a breakout narrative. Example: a security skill flags a company's automated-vulnerability work, a social digest catches that same company announcing a major deal, and a market tracker notes a related fraud-prevention win — three independent skills, one entity, in 48h. That bleedthrough is the signal. This skill catches it automatically.

Config

The signal-category map is operator-editable and lives in memory/topics/signal-categories.md. If the file doesn't exist, create the seed below and continue. The categories are what let the skill measure cross-category diversity (the core of the convergence score) — edit them to match the skills you actually run.

# Signal Categories

## Housekeeping (excluded — no external signals)
heartbeat, memory-flush, aeon-doctor, aeon-update, auto-merge,
skill-health, skill-repair, self-improve, operator-scorecard,
shiplog, fleet-control, fork-fleet, narrative-convergence

## Signal categories (skill → category)
| Category | Skills |
|----------|--------|
| market | token-pick, token-movers, defi-overview, monitor-polymarket, pm-manipulation, unlock-monitor, picks-tracker |
| social | fetch-tweets, narrative-tracker, mention-radar, write-tweet, last30 |
| ecosystem | github-trending, github-monitor, bd-radar, competitor-monitor |
| onchain | onchain-monitor, investigation-report, tx-explain, base-mcp |
| security | vuln-scanner, vuln-tracker, sc-audit, inbox-triage |
| research | article, digest, you-web-search, glim-mcp, idea-pipeline, idea-forge |
| opportunity | hunter-22, feedback-builder, taskmarket-delegate |

Steps

1. Identify which outputs to read

List output/.chains/*.md with the Glob tool. Exclude the Housekeeping skills from signal-categories.md — they carry no external signal.

Map each remaining output file to its category using the table in signal-categories.md. Any signal skill not listed in the table goes into an other category (so newly-added skills still count toward convergence, just without a named lane).

If ${var} is set, note it as a filter hint but still read all outputs — apply filtering at the scoring step.

2. Read each signal skill's output

For each signal skill output file that exists:

1. Read the file (or first 600 chars if large — enough to get entities and theme).
2. Extract: named entities (companies, protocols, people, tokens, projects) and key themes (e.g. "DNS rebinding", "coordination markets", "compute commoditization").
3. Note the skill name and category.

Build an entity/theme map:

{
  "<Entity>": [{ skill: "vuln-scanner", category: "security" }, { skill: "fetch-tweets", category: "social" }],
  "<theme>": [{ skill: "monitor-polymarket", category: "market" }, ...],
  ...
}

Also read memory logs from the last 2 days (Glob memory/logs/*.md, take the 2 most recent). From each log, extract entities/themes mentioned in specific skill run entries and add them to the map with their source skill. Every skill appends a log entry, so the signal map can be reconstructed from logs alone when output/.chains/ is sparse.