Competitor Pr Finder

Competitor PR Finder

Give it your product URL. It finds your competitors, researches every PR channel they used (news, podcasts, communities), surfaces the channels that appear across multiple competitors (your proven targets), finds the journalist or host for each, and drafts a personalized cold pitch for your product at every tier-1 channel.

---

Zero-hallucination policy: Every channel, journalist name, story angle, and pitch detail in the output must trace to a specific Tavily search result or the fetched product page. This applies to:

• Competitor names: must appear in Tavily search results, not AI training knowledge
• Channel names: must have a URL in the search results
• Journalist/host names: must appear verbatim in a Tavily snippet
• Story angles: extracted from article/episode titles in search results only
• Pitch drafts: reference specific evidence from search data + product analysis

---

Common Mistakes

| The agent will want to... | Why that's wrong | |---|---| | Name a journalist from training knowledge | Every journalist name must trace to a search result snippet. Writing "Sarah Perez covers startups at TechCrunch" from memory is hallucination. | | List channels without evidence URLs | Every channel in the output must have at least one URL from the PR search results proving a competitor was featured there. | | Skip the competitor confirmation step | Always show discovered competitors and wait for the user to confirm. Wrong competitors = wasted searches and a useless output. | | Generate generic pitches ("We'd love to be featured") | Every pitch must reference a specific angle from the evidence AND a specific differentiator from the product analysis. | | Mark a channel as Tier 1 with only 1 competitor occurrence | Tier 1 = 3+ competitors. Tier 2 = exactly 2. Tier 3 = 1. Do not promote channels that haven't proven themselves. | | Use em dashes in output | Replace all em dashes (--) with hyphens. |

---

Read Reference Files Before Each Run

cat references/pr-channel-types.md
cat references/pitch-guide.md
cat references/tier-scoring.md

---

Step 1: Setup Check

echo "TAVILY_API_KEY:    ${TAVILY_API_KEY:+set}${TAVILY_API_KEY:-NOT SET -- required}"
echo "FIRECRAWL_API_KEY: ${FIRECRAWL_API_KEY:+set}${FIRECRAWL_API_KEY:-not set, Tavily extract will be used as fallback}"

If TAVILYAPIKEY is missing: Stop immediately. Tell the user: "TAVILYAPIKEY is required to research competitors and find PR coverage. There is no fallback. Get it at app.tavily.com -- free tier: 1000 credits/month (about 43 full runs at ~23 searches/run). Add it to your .env file."

If only FIRECRAWLAPIKEY is missing: Continue. Tavily extract will be used for the URL fetch.

---

Step 2: Parse Input

Collect from the conversation:

• product_url: the URL to fetch (required, unless user pastes a description directly)
• product_name: optional, derived from page if not provided
• geography: optional -- US / Europe / global. Default: US

If the user provides only a pasted description (no URL): Skip Steps 3 and 4. Go directly to Step 4 (product analysis) using the pasted text as product_content. Set page_source to user_description and note in data_quality_flags.

If neither URL nor description: Ask: "What is the URL of your product or startup? Or paste a short description: what it does, who it is for, and what makes it different from competitors."

Derive product slug:

PRODUCT_SLUG=$(python3 -c "
from urllib.parse import urlparse
import sys
url = 'URL_HERE'
if url.startswith('http'):
    host = urlparse(url).netloc.replace('www.', '')
    print(host.split('.')[0])
else:
    import re
    print(re.sub(r'[^a-z0-9]', '-', url[:30].lower()).strip('-'))
")
echo "Product slug: $PRODUCT_SLUG"

---

Step 3: Fetch Product Page

Primary: Firecrawl (if FIRECRAWLAPIKEY is set)

curl -s -X POST https://api.firecrawl.dev/v1/scrape \
  -H "Authorization: Bearer $FIRECRAWL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url": "URL_HERE", "formats": ["markdown"], "onlyMainContent": true}' \
  | python3 -c "
import sys, json
d = json.load(sys.stdin)
content = d.get('data', {}).get('markdown', '') or d.get('markdown', '')
print(f'Fetched via Firecrawl: {len(content)} characters')
open('/tmp/cprf-product-raw.md', 'w').write(content)
"

Fallback: Tavily extract (if FIRECRAWLAPIKEY is not set)

curl -s -X POST https://api.tavily.com/extract \
  -H "Content-Type: application/json" \
  -d "{\"api_key\": \"$TAVILY_API_KEY\", \"urls\": [\"URL_HERE\"]}" \
  | python3 -c "
import sys, json
d = json.load(sys.stdin)
content = d.get('results', [{}])[0].get('raw_content', '')
print(f'Fetched via Tavily extract: {len(content)} characters')
open('/tmp/cprf-product-raw.md', 'w').write(content)
"

Checkpoint:

python3 -c "
content = open('/tmp/cprf-product-raw.md').read()
if len(content) ', a['industry_taxonomy']['l2'], '>', a['industry_taxonomy']['l3'])
print('Differentiators:')
for d in a['differentiators']:
    print(f'  - {d}')
"

---

Step 4b: Phase 1 -- Competitor Discovery

ls scripts/research.py 2>/dev/null && echo "script found" || echo "ERROR: scripts/research.py not found -- cannot continue"

python3 scripts/research.py \
  --phase discover \
  --product-analysis /tmp/cprf-product-analysis.json \
  --tavily-key "$TAVILY_API_KEY" \
  --output /tmp/cprf-competitors-raw.json

Print results for AI review:

python3 -c "
import json
data = json.load(open('/tmp/cprf-competitors-raw.json'))
print(f'Searches run: {len(data[\"competitor_searches\"])}')
for s in data['competitor_searches']:
    print(f'\nQuery: {s[\"query\"]}')
    print(f'Answer: {s.get(\"answer\",\"\")[:400]}')
    for r in s.get('results', [])[:5]:
        print(f'  - {r[\"title\"]} | {r[\"url\"]}')
        print(f'    {r.get(\"content\",\"\")[:200]}')
"

AI instructions: Read the search results above. Pick exactly 5 competitor companies that:

1. Are named in the search result titles, answers, or snippets
2. Are in the same L3 niche as the product being analyzed
3. Are actual competing products (not agencies, consultancies, or list articles)
4. Are distinct from each other (not the same company under different names)

For each competitor write: name, url (from the search result where they appeared), description (one sentence from snippet), source_url (the search result URL where they were found).

---

Step 5: Competitor Confirmation

Show the discovered competitors to the user:

python3  0:
    failures.append(f'INFO: {nf_count} field(s) marked "not found in search data" -- verify before outreach')

# Check 6: tier 1 channels have evidence URLs
for ch in result.get('tier_1_deep_dives', []):
    if not ch.get('evidence_urls'):
        failures.append(f'Warning: {ch["channel_name"]} has no evidence_urls')

if 'data_quality_flags' not in result:
    result['data_quality_flags'] = []
result['data_quality_flags'].extend(failures)

json.dump(result, open('/tmp/cprf-final.json', 'w'), indent=2)
print(f'QA complete. {len(failures)} issues addressed.')
for f in failures:
    print(f'  - {f}')
if not failures:
    print('All QA checks passed.')
PYEOF

Present the output:

## PR Intel: [product_name]
Date: [today] | Competitors researched: [N] | Tier 1 channels: [N] | Tier 2 channels: [N]

---

### Your Product
[one_line_description]
Differentiators: [list]
Competitors researched: [names]

---

### Tier 1 Channels (Proven Beats -- Found in 3+ Competitors)

*These channels have already covered multiple companies in your space.*

| Channel | Type | Found in | Journalist/Host | Approach |
|---|---|---|---|---|
[one row per tier 1 channel]

---

### Deep Dives + Cold Pitches

#### 1. [Channel Name] (Tier 1 -- [Type], found in [N] competitors)

Covers: [channel_overview]
Covered competitors: [found_in_competitors with evidence URLs]
Story angle they used: [why_they_covered_competitors]
Journalist/Host: [journalist_name] | Beat: [journalist_beat]
How to reach: [approach_method]

**Cold pitch:**
Subject: [subject]

[body -- 3-4 sentences]

---

[repeat for each tier 1 channel]

---

### Tier 2 Channels (Warm -- Found in 2 Competitors)

| Channel | Type | Found in | URL |
|---|---|---|---|
[one row per tier 2 channel]

---

### Tier 3 Channels (Discovery -- Found in 1 Competitor)

[comma-separated list of channel names with URLs]

---

### 3 Bonus Hooks (Angles Your Competitors Didn't Use)

1. [hook_text]
2. [hook_text]
3. [hook_text]

---
Data notes: [data_quality_flags, or "None"]
Saved to: docs/pr-intel/[PRODUCT_SLUG]-[DATE].md

Save to file and clean up:

DATE=$(date +%Y-%m-%d)
OUTPUT_FILE="docs/pr-intel/${PRODUCT_SLUG}-${DATE}.md"
mkdir -p docs/pr-intel
echo "Saved to: $OUTPUT_FILE"

rm -f /tmp/cprf-product-raw.md /tmp/cprf-product-analysis.json \
      /tmp/cprf-competitors-raw.json /tmp/cprf-competitors-confirmed.json \
      /tmp/cprf-pr-raw.json /tmp/cprf-pr-patterns.json \
      /tmp/cprf-journalist-results.json /tmp/cprf-final.json
echo "Temp files cleaned up."

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Varnan-Tech
• Source: Varnan-Tech/opendirectory
• License: MIT
• Homepage: https://www.opendirectory.dev

Install and usage instructions live in the source repository linked above.