Program Selection

Your time is the most expensive thing you have. Pick wrong, lose a month.

When to invoke

Trigger phrases:

• "should I hunt this program"
• "is this H1 worth it"
• "pick me a Bugcrowd target"
• "evaluate this program"
• "what target next"

Do not invoke when: the user has already committed to a target and is asking about recon or hunting.

The 7 Selection Criteria

Score each from 1-5. Hunt programs scoring ≥ 25/35.

1. Payout History (weight: ×2)

• Look at the last 50 disclosed reports on the program page
• Median bounty for the severity you're confident in
• Programs paying only minimums are red flag — they downgrade aggressively
• 5/5: median Critical > $5k, median High > $2k
• 3/5: median Critical $1-5k, High $500-2k
• 1/5: minimums only, or "swag/rep only"

2. Response Time

• HackerOne shows avg time to first response and avg time to triage
• Bugcrowd shows similar in program stats
• 5/5: first response < 24h, triage < 3 days
• 3/5: first response < 72h, triage < 7 days
• 1/5: response > 1 week — your dupe risk skyrockets

3. Scope Size

• Count in-scope assets (domains, mobile apps, APIs, smart contracts)
• Bigger scope = more attack surface, but more competition
• 5/5: wide scope *.target.com with subsidiary domains
• 3/5: 5-20 specific assets
• 1/5: one URL, no subdomains

4. Scope Quality (out-of-scope analysis)

Read the OOS list carefully. Red flags:

• "Self-XSS won't be paid" (fine, standard)
• "No subdomain takeover" → kills entire vuln class
• "No reports about CSP / cookie flags / missing headers" (good — saves time)
• "Findings must demonstrate impact" → high bar, plan PoC accordingly
• 5/5: OOS is sensible; high-impact classes welcome
• 1/5: too many vuln classes excluded

5. Competition Level

• Public programs: high competition, dupes likely
• Private invitation: lower competition, higher payout typical
• Signal of saturation: if disclosed reports show heavy clustering on the same paths, the program is "hunted dry" for the easy stuff
• 5/5: private invite, < 6 months old
• 3/5: public, < 12 months old
• 1/5: public, > 3 years old (low-hanging gone)

6. Tech Stack Match

• Are you strong in the tech they use?
• E.g., if you're a GraphQL/JS specialist and the target is a Java/Spring monolith, that's a mismatch
• 5/5: stack matches your top 2 skills
• 3/5: partial overlap
• 1/5: tech you've never touched

7. Program Hygiene

• Has the program closed reports as N/A or Informative > 30%?
• Have they downgraded severities controversially? (check disclosed activity)
• Does the manager respond to triagers?
• 5/5: clean record, fair severity assessment
• 1/5: known for harsh downgrades or ghosting

Step-by-Step Workflow

1. Pull program metadata

For HackerOne:

# View program page (manually copy URL into Claude / browser)
# Key fields to extract:
#   - Avg time to first response
#   - Avg time to triage
#   - Avg bounty (last 90 days)
#   - Disclosed report count
#   - In-scope assets

# If using HackerOne API (requires auth):
# https://api.hackerone.com/v1/hackers/programs

For Bugcrowd:

# Navigate to program → Activity tab → "Recent Validations"
# Key fields:
#   - VRT (Vulnerability Rating Taxonomy) used
#   - Average payout per VRT bucket
#   - Last activity

2. Score against the 7 criteria

Open templates/program-score-card.md and fill in.

3. Read 10 disclosed reports

• Pick the most recent 10
• Note: vuln classes that paid, severity disputes, triager tone
• This is the single best signal of "what works here"

4. Tech stack fingerprint

Quick passive recon (no scanning yet):

# DNS lookup
dig +short target.com

# Cert transparency for breadth
curl -s "https://crt.sh/?q=%25.target.com&output=json" | jq -r '.[].name_value' | sort -u | head -30

# Wappalyzer-style fingerprint via httpx
echo "target.com" | httpx -tech-detect -title -status-code -json

5. Decision

Total Score (out of 35)	Decision
30-35	Hunt immediately — high ROI expected
25-29	Hunt with focus — pick 1-2 strong vuln classes
20-24	Time-box 1 week — pull out if nothing found
< 20	Skip — better targets exist

Output template

# Program Selection Report: <program_name>

**Platform:** HackerOne / Bugcrowd / Intigriti
**Date evaluated:** YYYY-MM-DD
**Scope:** *.example.com + mobile (iOS, Android)

## Scoring (35 max)

| Criterion | Score | Note |
|---|---|---|
| Payout history (×2) | 4/5 (8/10) | Median High = $1,500 |
| Response time | 5/5 | First response avg 6h |
| Scope size | 5/5 | ~80 subdomains found |
| Scope quality | 3/5 | Subdomain takeover excluded |
| Competition | 2/5 | Public for 4 years |
| Tech stack match | 4/5 | React + GraphQL = my strong stack |
| Program hygiene | 4/5 | Fair triage, no controversies |
| **TOTAL** | **26/35** | **HUNT WITH FOCUS** |

## Recommended attack vectors
1. GraphQL endpoint at api.example.com — introspection, batching, deep-query attacks
2. New mobile app v3.x — re-scan after recent release
3. Auth flow — OIDC implementation, possibly custom

## Pass on:
- Subdomain takeover (excluded)
- Anything CSRF — they auto-mark N/A

## Time budget: 5 days

Cross-references

• [[scope-analysis]] — once you decide to hunt, fully parse the scope
• [[threat-modeling-mindmap]] — turn scope into hunt plan
• [[continuous-monitoring]] — set up watcher on private programs

Common pitfalls

1. Chasing big logos. Big company ≠ big payout. Many enterprise H1 programs pay min.
2. Ignoring competition. A 4-year public program on Yahoo's scope is picked clean.
3. Going after CVSS-only programs without considering business impact. Bugcrowd's VRT often pays more than CVSS would suggest.
4. Overweighting your familiarity bias. "I know X tech" ≠ "X tech still has bugs here".
5. Not reading the OOS list before scanning. You burn hours on a class they exclude.

Always-rejected check

Before committing, verify the program does NOT auto-reject:

• Missing security headers (X-Frame-Options, CSP without exploit)
• Theoretical CSRF on state-changing forms without PoC
• Self-XSS
• Email enumeration without bypass
• Outdated software without exploit
• TLS/SSL config issues without working PoC
• Subdomain takeover (if excluded)
• Open redirects (if excluded, OR if no impact chain)
• Logout CSRF
• Brute force without rate-limit evidence

→ See docs/always-rejected-list.md for the full list.