Event Prospecting

Take a conference URL → get a ranked list of people the AE should talk to, with a “why reach out” rationale per person.

Required: BROWSERBASE_API_KEY env var and the browse CLI installed (npm install -g browse). Use browse cloud ... for API calls and browse open / browse get markdown for JS-heavy speaker pages.

Path rules: Always use the full literal path in all Bash commands — NOT ~ or $HOME (both trigger “shell expansion syntax” approval prompts). Resolve the home directory once and use it everywhere. When constructing subagent prompts, replace {SKILL_DIR} with the full literal path (typically /Users/jay/skills/skills/event-prospecting).

Output directory: All event prospecting output goes to ~/Desktop/{event_slug}_prospects_{YYYY-MM-DD-HHMM}/. Final deliverable is index.html (people grouped by company, ranked by company ICP), with companies.html and people.html (filterable) as alternate views, plus results.csv for cold-outbound import.

CRITICAL — Tool restrictions (applies to main agent AND all subagents):

• All web searches: use browse cloud search. NEVER use WebSearch.
• All page content extraction: use node {SKILL_DIR}/scripts/extract_page.mjs "<url>". This script fetches via browse cloud fetch --output, parses title + meta tags + visible body text, and automatically falls back to browse get markdown when fetch fails or returns thin JS-rendered content. NEVER hand-roll a browse cloud fetch | sed pipeline. NEVER use WebFetch.
• All research output: subagents write one markdown file per company OR per person to {OUTPUT_DIR}/companies/{slug}.md or {OUTPUT_DIR}/people/{slug}.md using bash heredoc. NEVER use the Write tool or python3 -c. See references/example-research.md for both file formats.
• Report compilation: use node {SKILL_DIR}/scripts/compile_report.mjs {OUTPUT_DIR} --open.
• Subagents must use ONLY the Bash tool. No other tools allowed.
• HARD TOOL-CALL CAPS: ICP triage = 1 call/company; deep research = 5 calls/company; person enrichment = 4 calls/person. See references/workflow.md for enforcement detail.

CRITICAL — Anti-hallucination rules (applies to main agent AND all subagents):

• NEVER infer product_description, industry, or a person’s role_reason from a site’s fonts, framework, design system, or typography. These are cosmetic and say nothing about what the company sells or what the person does.
• NEVER let the user’s own ICP leak into a target’s description. If you don’t know what the target does, write Unknown — do not pattern-match them onto the ICP.
• product_description MUST quote or paraphrase a specific phrase from extract_page.mjs output. If none of TITLE/META/OG/HEADINGS/BODY yield a recognizable product statement, write Unknown — homepage content not accessible and cap icp_fit_score at 3.
• A person’s hook MUST quote or paraphrase a specific finding from a browse cloud search result (podcast title, blog headline, GitHub repo, talk abstract). If no public signal exists in the last 6 months, fall back to event-context (their talk title at this event).

CRITICAL — Minimize permission prompts:

• Subagents MUST batch ALL file writes into a SINGLE Bash call using chained heredocs. One Bash call = one permission prompt.
• Batch ALL searches and ALL fetches into single Bash calls using && chaining.

Pipeline Overview

Follow these 10 steps in order. Do not skip steps or reorder.

0. Setup — output dir + clean slate
1. Load profile — read profiles/{user_slug}.json
2. Recon — detect event platform
3. Extract people — people.jsonl
4. Group by company — seed_companies.txt
5. ICP triage — fast company-level scoring (1 call/company)
6. Filter — companies with icp_fit_score >= --icp-threshold
7. Deep research — full Plan→Research→Synthesize on ICP fits
8. Enrich speakers — ask user: ICP-fit only (default) or all speakers
9. Compile report — HTML + CSV, open in browser

The user invokes the skill with a URL like /event-prospecting <URL>. Parse EVENT_URL from that invocation message. Defaults: DEPTH=deep, ICP_THRESHOLD=6. The USER_SLUG (ICP profile) is auto-resolved in Step 1 from whatever profile files exist locally — there is no built-in default profile. Do NOT ask the user to confirm the URL — they already gave you it.

Step 0: Setup Output Directory

Derive the output directory from the URL the user gave you. Do NOT hardcode any event name.

# EVENT_URL came from the invocation message (whatever the user typed after `/event-prospecting`)
EVENT_SLUG=$(node -e 'const h = new URL(process.argv[1]).hostname.replace(/^www\./,""); console.log(h.split(".")[0])' "$EVENT_URL")
TIMESTAMP=$(date +%Y-%m-%d-%H%M)
OUTPUT_DIR=/Users/jay/Desktop/${EVENT_SLUG}_prospects_${TIMESTAMP}
mkdir -p "$OUTPUT_DIR/companies" "$OUTPUT_DIR/people"

Use the full literal home path — never ~ or $HOME. Pass {OUTPUT_DIR} as the full literal path to all subagent prompts.

Step 1: Load User Profile

The profile defines the ICP that ICP triage and deep research score against. Load from {SKILL_DIR}/profiles/{user_slug}.json (interchangeable across all GTM skills — same shape as company-research). example.json is a template, not a real profile — never use it.

DO NOT look outside {SKILL_DIR}/profiles/ for profiles — never reach into other skills’ directories. If a profile is needed elsewhere, the user copies it explicitly.

Resolution order:

1. If the user invoked with --user-company <slug>, use that slug.
2. Else, list profiles/*.json excluding example.json. If exactly one profile exists, use it (and tell the user which one). If multiple exist, ask the user (plain chat) which one.
3. If zero profiles exist, fail loudly and instruct the user to create one (copy profiles/example.json to profiles/<your_slug>.json and fill it in, or run the company-research skill which builds one automatically).

PROFILES=$(ls {SKILL_DIR}/profiles/*.json 2>/dev/null | xargs -n1 basename | sed 's/\.json$//' | grep -v '^example$')
COUNT=$(echo "$PROFILES" | grep

The profile yields: company, product, icp_description, existing_customers. These get embedded verbatim in every subagent prompt downstream.

Step 2: Recon

Detect the event platform and extraction strategy. One command:

node {SKILL_DIR}/scripts/recon.mjs {EVENT_URL} {OUTPUT_DIR}

Writes {OUTPUT_DIR}/recon.json with platform, strategy, and (for Next.js) nextDataPaths. See references/event-platforms.md for the platform catalog and detection priority.

Expected outcomes:

• Stripe Sessions class (Next.js): platform: "next-data", 1-3 paths
• Sessionize: platform: "sessionize"
• Lu.ma / Eventbrite: platform: "luma" | "eventbrite"
• Anything else: platform: "custom", strategy: "markdown" (best-effort fallback)

Step 3: Extract People

node {SKILL_DIR}/scripts/extract_event.mjs {OUTPUT_DIR} --user-company {USER_SLUG}

Reads recon.json, dispatches to the platform-specific extractor, writes people.jsonl (one speaker per line) and seed_companies.txt (deduped companies).

The --user-company flag also drops the host-org’s own employees (a Stripe-hosted event drops Stripe employees) and the user’s own employees from the speaker list — those aren’t prospects.

Sanity-check the output:

wc -l {OUTPUT_DIR}/people.jsonl {OUTPUT_DIR}/seed_companies.txt
head -3 {OUTPUT_DIR}/people.jsonl

If people.jsonl is empty or under ~10 lines, recon picked the wrong platform — see references/event-platforms.md and re-run with adjusted strategy.

Step 4: Group by Company

extract_event.mjs emits seed_companies.txt already (one company per line, deduped, sorted). This step is informational — verify the count looks reasonable before fanning out:

wc -l {OUTPUT_DIR}/seed_companies.txt

Expected: roughly 0.4-0.6× the speaker count (most events have ~2 speakers per company on average, some companies send 5+, many send 1).

Step 5: ICP Triage

Fast pass — one tool call per company, no deep research. Score every company in seed_companies.txt against the user’s ICP and write a thin triage stub to companies/{slug}.md. Companies with icp_fit_score >= --icp-threshold (default 6) advance to Step 7’s deep research; the rest stay as triage stubs.

Dispatch pattern: split seed_companies.txt into batches of ~10 and fan out N subagents in a SINGLE Agent batch (multiple Agent tool calls in one message). Each subagent runs the prompt from references/workflow.md → “ICP Triage” section. Hard cap: 1 tool call per company (just extract_page.mjs on the homepage), enforced via the # browse call N/1 comment pattern.

# Build batch files: each batch line is "name|guessed_homepage|slug".
# extract_event.mjs only emits company NAMES (no URLs), so we slugify and guess
# https://{slug-without-spaces}.com as the canonical homepage. The triage subagent
# is allowed to write product_description: "Unknown — homepage content not accessible"
# and cap score at 3 if the guessed URL 404s — that's the documented fallback in
# workflow.md (rule 3 of the ICP Triage prompt). Burning a real browse cloud search to
# discover the URL would bust the 1-call-per-company HARD CAP.
node -e '
const fs = require("fs");
const slugify = (s) => (s || "").toLowerCase().replace(/[^a-z0-9]+/g, "-").replace(/^-+|-+$/g, "");

Then in a single message, dispatch one Agent call per batch (up to 6 in parallel; subsequent waves after the first returns). Each Agent gets the prompt from references/workflow.md → “ICP Triage” with these substitutions before sending:

• {SKILL_DIR} → full literal skill path (e.g. /Users/jay/skills/skills/event-prospecting)
• {OUTPUT_DIR} → full literal output path
• {USER_COMPANY}, {USER_PRODUCT}, {ICP_DESCRIPTION} → from the loaded profile
• {EVENT_NAME} → recon.json .title
• {COMPANY_LIST} → contents of the batch file (e.g. cat {OUTPUT_DIR}/_batch_triage_aa)
• {TOTAL} → number of lines in this batch (substitute into # browse call N/{TOTAL})

Agent dispatch (skeleton, repeat per batch in one message):

Agent(
  description: "ICP triage batch aa",
  prompt: <ICP Triage prompt from workflow.md with all placeholders substituted>,
  subagent_type: "general-purpose"
)
Agent(
  description: "ICP triage batch ab",
  prompt: <same prompt template, COMPANY_LIST swapped to batch ab>,
  subagent_type: "general-purpose"
)
... up to 6 per message

After all subagents return, verify every company in seed_companies.txt has a corresponding companies/{slug}.md:

ls {OUTPUT_DIR}/companies/*.md | wc -l
# Should equal `wc -l {OUTPUT_DIR}/seed_companies.txt`

Clean up the batch files: rm {OUTPUT_DIR}/_batch_triage_*.

Step 6: Filter by ICP Threshold

Read each companies/*.md frontmatter, keep those with icp_fit_score >= 6 (or whatever --icp-threshold is). Write the surviving company slugs to {OUTPUT_DIR}/icp_fits.txt:

THRESHOLD=6   # from --icp-threshold flag
for f in {OUTPUT_DIR}/companies/*.md; do
  score=$(awk '/^icp_fit_score:/{print $2; exit}' "$f")
  if [ -n "$score" ] && [ "$score" -ge "$THRESHOLD" ]; then

Expected: 20-40% of seed_companies.txt. If the survival rate is < 10%, the threshold may be too high or the ICP description too narrow — surface a warning to the user.

Step 7: Deep Research

Full Plan→Research→Synthesize on ICP-fit companies only. Hard cap: 5 tool calls per company (homepage extract + 2-3 sub-question searches + 1-2 supplementary fetches). Subagents OVERWRITE the existing companies/{slug}.md triage stub with the richer deep-research version (frontmatter triage_only: false).

Dispatch pattern: split icp_fits.txt into batches of ~5 (deep mode default) and fan out one Agent per batch in a SINGLE message (up to 6 Agents per message). Each Agent gets the prompt from references/workflow.md → “Deep Research” with these substitutions:

• {SKILL_DIR}, {OUTPUT_DIR}, {USER_COMPANY}, {USER_PRODUCT}, {ICP_DESCRIPTION}
• {EVENT_NAME} (from recon.json .title), {EVENT_CONTEXT} (track / topic, manually inferred from the event homepage)
• {COMPANY_LIST} → contents of the batch file (each line slug|website)

# Build {company-slug|website} pairs by reading frontmatter from each triage stub
while read slug; do
  website=$(awk '/^website:/{print $2; exit}' {OUTPUT_DIR}/companies/${slug}.md)
  echo "${slug}|${website}"
done < {OUTPUT_DIR}/icp_fits.txt > {OUTPUT_DIR}/_deep_targets.txt

# Split into ~5-company batches (deep mode)
split -l

Agent dispatch (skeleton, repeat per batch in one message):

Agent(
  description: "Deep research batch aa",
  prompt: <Deep Research prompt from workflow.md with all placeholders substituted; COMPANY_LIST = cat _batch_deep_aa>,
  subagent_type: "general-purpose"
)
Agent(
  description: "Deep research batch ab",
  prompt: <same template, COMPANY_LIST = cat _batch_deep_ab>,
  subagent_type: "general-purpose"
)
... up to 6 per message; second wave after the first returns

After all subagents return, verify the deep-research files exist and have triage_only: false:

grep -l "triage_only: false" {OUTPUT_DIR}/companies/*.md | wc -l
# Should equal wc -l icp_fits.txt

Step 8: Enrich Speakers

Per person: harvest LinkedIn URL, recent activity (podcast / blog / talk / GitHub / X), and write people/{slug}.md. Hard cap: 4 tool calls per person, three lanes:

1. browse cloud search "{name} {company} linkedin" (always)
2. browse cloud search "{name} podcast OR talk OR blog 2026" (deep+)
3. browse cloud search "{name} github" (deeper)
4. browse cloud search "{name} site:x.com OR site:twitter.com" (deeper, best-effort)

Quick mode: skip Step 8 entirely. Deep mode: lanes 1-2. Deeper mode: lanes 1-4.

Step 8a — Ask the user: scope of enrichment

Before dispatching, compute the two candidate counts and ask the user to choose. The default is ICP-fit only (faster, cheaper, what most users want); enriching every speaker is opt-in because cost scales linearly with people enriched.

TOTAL=$(wc -l < {OUTPUT_DIR}/people.jsonl)
ICP_FITS=$(node -e '
const fs = require("fs");
const fits = new Set(fs.readFileSync("{OUTPUT_DIR}/icp_fits.txt", "utf-8").split("\n").filter(Boolean));
const slug2name = {};
for (const slug of fits) {
  const md = fs.readFileSync(`{OUTPUT_DIR}/companies/${slug}.md`, "utf-8");
  const m = md.match(/^company_name:\s*(.+)$/m);
  if (m) slug2name[slug] = m[1].trim();

Then ask via AskUserQuestion — clean two-option choice with the quantified cost on each:

AskUserQuestion(questions: [
  {
    question: "Enrich which speakers?",
    header: "Enrichment scope",
    multiSelect: false,
    options: [
      { label: "ICP fits only", description: "${ICP_FITS} speakers, ~$((ICP_FITS * LANES)) calls (recommended)" },
      { label: "All speakers", description: "${TOTAL} speakers, ~$((TOTAL * LANES)) calls" }
    ]
  }
])

Save the chosen scope as ENRICH_SCOPE=icp_fits or ENRICH_SCOPE=all. If the user picks “All speakers” and TOTAL × LANES > 600, print a warning and ask once more — that’s a 10+ minute run with hundreds of tool calls.

Step 8b — Filter and batch

# Build _people_to_enrich.jsonl based on ENRICH_SCOPE
if [ "$ENRICH_SCOPE" = "all" ]; then
  cp {OUTPUT_DIR}/people.jsonl {OUTPUT_DIR}/_people_to_enrich.jsonl
else
  node -e '
const fs = require("fs");
const fits = new Set(fs.readFileSync("{OUTPUT_DIR}/icp_fits.txt", "utf-8").split("\n").filter(Boolean));
const slug2name = {};

Then in a single message, dispatch one Agent call per batch (up to 6 per message) with the prompt from references/workflow.md → “Person Enrichment”. Each subagent’s prompt should include:

• {SKILL_DIR}, {OUTPUT_DIR}, {DEPTH} (deep | deeper)
• {USER_COMPANY}, {USER_PRODUCT}, {ICP_DESCRIPTION}
• {EVENT_NAME} (from recon.json .title)
• {LANES} → 2 for deep mode, 4 for deeper mode (substituted into # browse call N/{LANES})
• {PEOPLE_BATCH} → contents of _batch_people_aa (each line a JSON record from people.jsonl)

Agent dispatch (skeleton, repeat per batch in one message):

Agent(
  description: "Person enrichment batch aa",
  prompt: <Person Enrichment prompt from workflow.md with all placeholders substituted; PEOPLE_BATCH = cat _batch_people_aa>,
  subagent_type: "general-purpose"
)
Agent(
  description: "Person enrichment batch ab",
  prompt: <same template, PEOPLE_BATCH = cat _batch_people_ab>,
  subagent_type: "general-purpose"
)
... up to 6 per message

After all subagents return, verify the people files exist:

ls {OUTPUT_DIR}/people/*.md | wc -l
# Should equal wc -l _people_to_enrich.jsonl

Step 9: Compile Report

Generate the company-grouped HTML index, alternate views, and CSV in one command:

node {SKILL_DIR}/scripts/compile_report.mjs {OUTPUT_DIR} --open

This generates:

• {OUTPUT_DIR}/index.html — people grouped by company, ranked by company ICP score (opens in browser)
• {OUTPUT_DIR}/people.html — filterable speaker list (alternate view)
• {OUTPUT_DIR}/companies.html — ICP-ranked company table with attendees
• {OUTPUT_DIR}/results.csv — cold-outbound-ready spreadsheet

Then present a summary in chat:

## Event Prospecting Complete — {Event Name}

- **Total speakers extracted**: {count}
- **Unique companies**: {count}
- **ICP fits (score ≥ {threshold})**: {count}
- **Speakers enriched**: {count}
- **Score distribution** (companies):
  - Strong fit (8-10): {count}
  - Partial fit (5-7): {count}
  - Weak fit (1-4): {count}
- **Report opened in browser**: {OUTPUT_DIR}/index.html

Show the top 5 people cards as a markdown table sorted by company ICP score, then offer to:

• Adjust --icp-threshold and re-run Steps 6-9
• Export the CSV to a CRM