claude-dev-suite Pinning Attacks
Pinning = an attacker publishes a tx that blocks your tx from confirming by attaching as a descendant chain that exhausts mempool descendant limits or replacement-fee budgets.
Classic pinning
You publish parent P (commitment, HTLC-timeout, etc.). Attacker publishes child C spending P's output:
- C has very low fee rate but has 24+ unconfirmed ancestors/ descendants of its own (Tx graph).
- Mempool admits C; descendant chain reaches policy limits.
- Now ANY new descendant of P is rejected: descendant limit exceeded.
- You can't CPFP P; you can't RBF (your replacement would conflict with C, but BIP125 rule 5 makes it economically prohibitive).
Result: your honest tx P is "stuck" until mempool eviction (which might never happen if attacker keeps refilling).
Lightning impact
In Lightning context:
- After force-close, you publish HTLC-timeout to claim funds before CLTV expiry.
- Attacker pins HTLC-timeout's anchor / output → can't bump.
- CLTV expires upstream (Alice's HTLC times out before you can claim Bob's).
- You lose the in-flight HTLC.
Mitigation 1: Anchor outputs (BOLT-3 update, ~2020)
Each commitment tx gets two 330-sat anchor outputs (one per party):
- Spendable immediately by the party.
- Spendable by anyone after 16 blocks.
You can spend your anchor with a high-fee child to CPFP the commitment. Defeats simple pinning of commitment.
But: HTLC-timeout txs themselves still vulnerable.
Mitigation 2: Ephemeral anchors + TRUC v3 (BIP431, 2024)
- Ephemeral anchor:
value=0, trivial script, MUST be spent in same package. - TRUC v3: ≤ 1 descendant, ≤ 10 kvB tx size, sibling eviction.
Combined: pinning surface is removed. Attacker can't keep stacking descendants; honest party can sibling-evict bad children cheaply.
Mitigation 3: Package relay (BIP331)
submitpackage lets you submit parent + your CPFP child atomically.
Mempool computes effective package fee rate. No race against attacker.
State as of late 2025
- Anchor commitment + TRUC v3 + ephemeral anchors + package relay combined → robust against known pinning variants.
- Legacy non-anchor channels still exist; gradual migration.
- Channels opened post-BIP431 activation should use TRUC v3 by default per impl.
Implementation status
| Impl | TRUC v3 | Ephemeral anchors | Package CPFP |
|---|---|---|---|
| LDK | yes (post 0.0.130) | yes | yes |
| LND | yes (0.18+) | yes | yes |
| CLN | yes | yes | yes |
| Eclair | yes | yes | yes |
Common confusions
- Pinning vs replacement cycling — both attack mempool
policy but differently:
- Pinning: block your tx from confirming.
- Cycling: drain your bumping budget while attacker's tx sits.
- Pinning vs channel jamming — different layers (on-chain vs off-chain).
- Anchor outputs mitigate commitment pinning but not necessarily HTLC pinning. Full fix: anchor + TRUC v3 + ephemeral anchors.