Web3 Hunt Foundation

WEB3 HUNT FOUNDATION

Mindset + Recon + Setup. Read this before touching any new target's code.

See also: [[Web3 Audit]], [[Methodology]], [[Trust Map]], [[Wild Mode]], [[Web3 Grep Arsenal]]

PART 1: THE HUNTER MINDSET

The Core Mental Shift

You are NOT looking for "vulnerabilities" in the abstract. You are looking for specific actions an attacker can take TODAY that result in profit.

Everything flows from one question: "What can I STEAL, FREEZE, or DESTROY — and what do I END UP WITH?"

The Bug Validation Template

Apply to every finding before writing a single line:

I am an attacker. I will:
1. SETUP:   What do I need? (wallet, capital, any whitelisted permissions?)
2. CALL:    Exact transactions, exact order, exact function names
3. RESULT:  What do I end up with that I didn't start with?
4. COST:    Gas + capital + flash loan fee + any other expense
5. DETECT:  Can anyone stop or reverse this?
6. NET ROI: I gained X at cost of Y. Is Y << X?

If you can't fill in steps 2 and 3 with specific function calls → it's not a real bug. Stop. Move on.

10 Attacker Questions (Ask For Every External Function)

1. What if amount = 0? Does anything revert or silently pass?
2. What if I call this function twice in the same block?
3. What if I call this before initialize() is called?
4. What if I front-run this transaction?
5. What if the external call fails? Does state get half-updated?
6. What if the token has fee-on-transfer? Does amount received ≠ amount sent?
7. What if I pass address(0) or a malicious contract as an address param?
8. What if I pass type(uint256).max as a numeric param?
9. Can I combine this with a flash loan? (zero-cost capital changes the math)
10. Does a sibling function lack the same modifier this function has?

Question #10 explains 19% of all Critical findings. If vote() has onlyRole(VOTER), check poke(), reset(), harvest() — the missing modifier on the sibling IS the bug.

6 Triager Counter-Questions (Disprove Your Own Finding)

Before spending time on a PoC, try to KILL the finding:

1. Is there an upstream check I missed that actually prevents this?
2. Is this documented intended behavior (whitepaper, NatSpec, design decision)?
3. Does exploitation require admin/privileged access? (Usually invalid if yes)
4. Is the economic cost to exploit greater than the gain? (Not viable if yes)
5. Was this flagged in a prior audit as "acknowledged" or "risk accepted"?
6. Is the "sensitive" data already publicly visible to anyone in the web UI?

One YES = KILL. Move on.

5-Minute Rule

If you've been on the same function for 5 minutes with no clear attack path → STOP. Add it to a low-priority list. Move to the next function. Top hunters: 95% fast-reject + 5% deep dives on confirmed leads.

Depth Over Breadth

Don't review 10 protocols in one week. Pick ONE. Spend 3-5 days becoming the expert. Protocol-specific knowledge compounds. The Curve expert found 5 bugs. The 10-protocol tourist found 0.

Inconsistency Is Proof

If functionA() has a security check, and functionB() doesn't — that IS the report. You don't need to fully understand why. The inconsistency proves the developer intended the check.

PART 2: TARGET SCORING — GO / NO-GO

Before touching any code: score the target. Score < 6 → skip.

Target Scorecard

< 4: Skip — too small, too audited, wrong fit 4-5: Only if nothing better available 6-8: Good — spend 1-3 days ≥ 9: Excellent — spend up to 1 week

PART 3: RECON METHODOLOGY (30-Minute Protocol)

Step 1 — Read Immunefi Page (5 min)

Note:
- All in-scope contract addresses + GitHub links
- Out-of-scope list (DO NOT report these)
- Primacy of Impact: YES/NO (YES = more forgiving on novel impacts)
- Max bounty amounts by severity
- Time on Immunefi (newer = fewer duplicates)

Step 2 — Clone + Setup (5 min)

git clone <target-repo>
cd <target-repo>
git log --oneline -20       # Recent changes = freshest bugs here
forge build                 # Must compile clean (fix if not)
forge test                  # Note failures — may indicate known issues
forge coverage              # Untested code = priority review target

Step 3 — Read ALL Prior Audit Reports (15 min)

For each finding, note its status:

• Fixed: Skip
• Acknowledged / Risk Accepted: ⚡ START HERE ⚡
  • Developer knows about it but chose not to fix it
  • Variants, escalations, related attack paths = in-scope and uncovered
• Partially Fixed: Verify fix actually closes ALL attack paths

Find audits: GitHub repo, protocol docs, Immunefi page, Google "[protocol] audit report"

Step 4 — Crown Jewels (2 min)

Ask: "Worst thing an attacker could do to users of this protocol?"

Work backward from impact to code:

• "Steal deposits" → find: withdrawal functions, access control on transfer
• "Mint infinite tokens" → find: mint functions, who calls them, what checks
• "Freeze all funds" → find: emergency functions, time locks, role assignments
• "Steal all rewards" → find: reward distribution, distributor role, harvest functions

Step 5 — Architecture + Fund Flow (3 min)

Draw the money flow (even mentally):

User USDC
   ↓ deposit()
[Protocol Vault] ──→ External Protocol (Aave/Compound/Uniswap)
   ↓ yield accumulates
[Reward Distributor] ──→ Users via claim/harvest

Find WHERE VALUE ACCUMULATES. That contract = highest priority.

Key state variables to map:

• Total deposited / total assets / total shares
• Per-user balance tracking (how is it updated?)
• Reward accumulator (index, per-share, per-second?)
• Role assignments (owner, admin, governance, distributor)
• Time locks (timestamps, epochs, cooldowns)

Step 6 — Static Analysis (5 min)

# Slither — 93 detectors, fast
slither . --exclude-low --filter-paths "test|lib|node_modules"
slither . --detect reentrancy-eth,unprotected-upgrade,arbitrary-send-eth

# Aderyn — Rust-based, Foundry-native
aderyn . --output report.md

# Read output → note HIGH/CRITICAL only
# Tools catch ~30-40% of bugs. Human review finds the rest.

PART 4: RECON CHECKLIST

Run through this before any deep review:

PROGRAM:
[ ] Max bounty noted per severity
[ ] ALL in-scope contracts listed (name + address)
[ ] Out-of-scope list read — nothing to falsely report
[ ] Primacy of Impact: YES/NO noted
[ ] Program launch date noted (new = good)

PRIOR AUDITS:
[ ] All audit PDFs downloaded and scanned
[ ] Each finding: status noted (Fixed/Ack/Risk Accepted)
[ ] Acknowledged items in notes as starting points

CODEBASE:
[ ] git clone + forge build passes
[ ] git log checked — recent commits noted
[ ] forge coverage run — untested functions noted
[ ] Slither + Aderyn run — high/critical noted

ARCHITECTURE:
[ ] Fund flow drawn
[ ] Crown jewels identified (where value lives)
[ ] External dependencies mapped (Chainlink, Uniswap, Aave, etc.)
[ ] ALL privileged roles found (onlyOwner, onlyRole, etc.)
[ ] Proxy/upgradeable pattern identified (if any)

ATTACK SURFACE:
[ ] All external/public non-view functions listed
[ ] Mint/burn functions located
[ ] Withdraw/emergencyWithdraw functions located
[ ] Upgrade/migration functions located
[ ] Oracle dependencies found
[ ] Signature/permit usage found
[ ] Cross-contract interactions mapped

ATTACK SURFACE BY PROTOCOL TYPE

DEX / AMM:
- Oracle manipulation (getReserves, slot0 = flash-loan manipulable)
- Rounding in pool math (1-wei attacks × flash swap)
- Missing slippage protection (sandwich vector)
- Fee-on-transfer token handling

LENDING / BORROWING:
- Collateral valuation (oracle → overborrow)
- Liquidation logic (bad debt creation, self-liquidation)
- Interest accrual rounding (favors borrower or protocol?)
- Flash loan → inflate collateral → borrow → repay

VAULT / YIELD:
- First depositor share inflation (ERC4626)
- Donation attack via direct balanceOf transfer
- Strategy rug (malicious strategy contract)
- Reward accounting timing (enter/exit attacks)

BRIDGE / CROSS-CHAIN:
- Message replay (missing nonce/nullifier)
- Signature replay (no chainId)
- Validator set manipulation
- Destination execution reentrancy

STAKING / RESTAKING:
- Reward distribution timing attacks
- Slashing logic errors
- Role never granted → permanent lock
- Withdrawal queue multi-field desync

WHEN TO RE-READ THIS CHAIN

→ NEXT: 02-bug-classes.md