Every captured commit receives deterministic security triage and a separate communication-quality score. Security candidates and broader second-pass signals receive full-patch Ollama analysis.
Message quality measures whether a commit identifies its scope, purpose, rationale, testing, and supporting references. It does not change the security-severity score.
This change fixes a wallet synchronization check in LND. Previously, the wallet only compared block heights to decide if it was fully caught up with the Bitcoin chain. During a chain reorganization (a temporary fork where one block is repl…
State synchronization check previously used height only, which is insufficient across reorgsFix adds block hash comparison to prevent false-positive sync statusCould cause LND to proceed while wallet state reflects an invalidated or not-yet-updated chain tip
This commit fixes a missing permission check on a new Lightning routing RPC called XFindBaseLocalChanAlias. Without the entry, any valid macaroon could call that RPC, bypassing the normal access-control list. The patch adds the required pe…
Missing macaroon authorization entry for an RPC endpointPotential privilege escalation: any valid macaroon could invoke an unprotected RPCRegression test added to enforce macPermissions coverage
This large LND patch changes when an ordinary invoice is considered 'paid'. Previously, releasing the preimage (the secret that lets the payer claim the funds) immediately marked the invoice as settled. Now the invoice enters a new interme…
New intermediate invoice/HTLC state prevents premature settlement reportingOn-chain HTLC outcomes persisted without opt-in to support restart recoveryForwarding package marker persisted before preimage release to avoid replay rejection on restart
This commit only adds documentation to the release notes. It describes a previously merged code change (PR 11555) about how LND handles invoice finality, but the commit itself contains no code, bug fixes, or behavior changes. There is noth…
This change fixes a bug in LND's invoice handling for multi-part payments (MPP and AMP). When a payment is split into several pieces (HTLCs), each piece is accepted individually. Previously, an old accepted piece could become too close to …
Fixes missing revalidation of accepted HTLC CLTV expiries at MPP/AMP set completionPrevents settlement of shards whose on-chain claim window has become unsafe while heldAdds new fail resolution ResultHtlcSetExpiryTooSoon and cancels the entire affected set
This commit changes how LND's wallet locks (leases) unspent coins through its RPC interface. It removes a recently added 'release after N confirmations' option and replaces it with an explicit 'no expiry' flag. Old clients that still send …
Breaking API change that reserves removed protobuf fields to prevent accidental reintroductionExplicit rejection of legacy `release_after_spend_confs` parameters from old clients instead of silent fallbackValidation prevents conflicting no-expiry + duration lease settings before wallet state is touched
This commit refactors how the Lightning Network Daemon (LND) stores and queries channel graph data so that the same channel or node can exist under both old (v1) and new (v2) gossip protocols without the database returning stale, missing, …
Cross-version data consistency: prevents a v2 channel shell from shadowing a usable v1 channel, which could affect routing decisions and RPC output.Cache consistency: ensures the in-memory pathfinding cache and the SQL-backed RPC reads agree on which channel version and policies are active.Bug fix: zombie edge info now carries the correct channel ID instead of 0, which could have caused incorrect channel identification during resurrection.
This is a large bug-fix patch for LND (a Bitcoin Lightning Network node) that tightens correctness rules across many subsystems. It prevents several ways a node could mis-handle channel data, accept bad network gossip, commit corrupted dat…
Database transaction commit now aborts on read errors (etcd) - prevents silent data corruption / inconsistent stateGossip zombie marking removed for invalid funding script - prevents denial-of-service / channel graph poisoningWire message length bounding added to watchtower protocol - prevents memory exhaustion / malformed message parsing
This commit changes how LND handles multi-part Lightning payments when one part fails after the preimage was already released. Previously, the invoice could be marked as fully paid even if one shard was later reclaimed by the payer on-chai…
Changes invoice settlement semantics for released-but-failed HTLCsAdds new persisted HTLC state `HtlcStateFailed`Prevents invoice from reaching `ContractSettled` when any released shard fails
This change tightens the rules for opening Lightning payment channels by refusing a setting called the CSV delay when it is set to zero. The CSV delay is the waiting period a user must endure before reclaiming their own funds from a channe…
Rejects a zero CSV delay that would eliminate the breach-remedy windowCentralizes the check in VerifyConstraints so all channel types and directions are coveredAdds explicit error type and unit test for the new constraint
This commit fixes two bugs in how Lightning Network channel data is saved to disk. First, it stops extra 'aux' signatures meant only for advanced taproot channels from being saved on ordinary channels, where they don't belong. Second, it p…
Persisted state could become unreadable due to writer/reader size mismatchAuxiliary signatures for taproot overlay channels were persisted on plain channel HTLCsFix adds explicit serialization bound to prevent unreadable channel records
This commit adds the first end-to-end BOLT 12 'Offers' flow to LND: a new dev-only RPC and CLI command to create offers, plus a dispatcher that listens for incoming onion-message invoice requests and replies with signed BOLT 12 invoices. I…
New network-facing dispatcher answers invoice requests from any peer; rate and concurrency limits are appliedBlinded-path PathID check prevents reply linking to self-created paths, but TODO notes the check should move once offers can have offer_pathsCreateOffer RPC is restricted to dev builds and requires offchain:write macaroon permission
This change fixes a bug in LND's invoice handling for AMP (Atomic Multi-Path) payments. Previously, if someone tried to pay a fresh shard for an AMP set that was already fully paid, the invoice updater could return an internal error that w…
Denial-of-service vector: an internal updater error could force-fail a shared channel link, disrupting unrelated sibling HTLCsAMP invoice state machine edge case: reusable invoice stays open for other sets but must reject new shards for already-settled setsClean failure path introduced: ResultInvoiceNotOpen instead of error-driven link failure
This is a large feature merge for LND that adds a new optional RPC subsystem called switchrpc. It lets an external program hand a pre-built onion packet to LND for direct dispatch over a Lightning channel, instead of LND doing its own path…
New optional RPC surface (switchrpc) that accepts externally crafted onion packets and directly dispatches HTLCsIdempotency/deduplication added via InitAttempt durable write-ahead log and ErrPaymentIDAlreadyExistsStartup orphan cleanup fails half-open or pre-commit attempts to avoid hangs
This commit fixes a bug in LND's transaction sweeping logic. Previously, if the Bitcoin backend rejected a sweep transaction for a temporary or unknown reason (for example, the mempool was full, a network error occurred, or the rejection r…
Funds-availability / liveness fix: temporary or unknown mempool rejections could leave sweep inputs unspent.Error-handling reclassification: previously fatal errors now retryable.Retry metadata preservation: next fee rate and fee cap kept across attempts.
This change is a hardening fix for LND's remote-signing feature. Previously, if a node was configured to act as an outbound remote signer but was compiled without the required 'signrpc' and 'walletrpc' sub-servers, it would silently fall b…
silent-degradation-to-no-op preventedconfiguration-misconfiguration now surfaces as fatal errorbuild-tag-dependency explicitly validated at runtime
This commit only fixes blank-line spacing in a single test file to satisfy lint checks. It adds one blank line after a function signature and removes one trailing blank line. There are no code behavior changes, no bug fixes, and no securit…
This change updates two GitHub Actions used in the project's automated lint workflow so they are locked to specific commit hashes instead of floating version tags. A floating tag could be silently repointed to a different, possibly malicio…
CI/CD supply-chain hardeningthird-party action pinning by commit hashprevention of tag-movement attacks
This commit fixes the GitHub Actions CI setup for running a custom code linter. It makes sure Docker can read and write build caches correctly and that the built linter image is available for later steps. There is no user-facing security f…
This patch fixes a database bug in the Lightning Network Daemon (LND) where very long lists of unsigned, acknowledged channel updates could be silently truncated when saved to disk. Because the storage format used a 16-bit counter, lists l…
Integer truncation / wrap-around in serialization format (uint16 count)Silent data loss on restart due to undersized length fieldDefensive deserialization with count-modulo validation
Expose the no-timeout lifetime explicitly for unused output leases. Document owner cancellation and first-confirmed-spend cleanup, and add boolean acknowledgement fields so clients can verify the selected mode.
78/100 · AdequateMessage clarity
✓ Descriptive subject✓ Names a concrete action or component✓ Provides detailed explanatory context✓ Mentions testing or verification
Why it was queued
signing or wallet pathsecond-pass: security-sensitive path
AI analysis · Informational 19/100
This commit only changes the API definition file (a .proto file) for LND's wallet RPC. It adds new boolean flags and documentation describing how long-term output leases work. There is no actual code logic change here—just the interface and comments—so it does not by itself fix or introduce a security vulnerability. It is a design/documentation/API surface change that may be related to a larger feature around lease lifetimes.
AI review queuedMerge pull request #11383 from ziggie1984/build-go-1.27.3by ziggieXXX · bfe5039f · Oct 20, 2026 · 34 filesMessage 73 · AdequateLow 28Details
Commit message · ziggieXXX
Merge pull request #11383 from ziggie1984/build-go-1.27.3
build: bump Go language to 1.26.10 and toolchain to 1.27.3
73/100 · AdequateMessage clarity
✓ Specific, descriptive subject✓ Names a concrete action or component✓ Provides an explanatory body✓ Links an issue, advisory, or supporting reference
This commit is primarily a routine update of the Go compiler/toolchain version used to build LND (from Go 1.27.1 to 1.27.3, and the language version from 1.26.8 to 1.26.10) across many build files, Docker images, and sub-modules. It also includes a few small test-code cleanups that make error handling stricter. The change itself is not a security patch for LND logic, but it moves the project to a newer Go point release that may contain upstream fixes for compiler/runtime issues. There is no explicit vendor statement that this update fixes a security vulnerability.
AI review queuedbuild: bump Go language and toolchain versionsby ziggie · d67c9edf · Oct 20, 2026 · 30 filesMessage 80 · StrongInformational 15Details
Commit message · ziggie
build: bump Go language and toolchain versions
Keep lnd compatible with the previous stable Go release by updating the language version in every tracked go.mod file to Go 1.26.10.
Move build and release tooling to Go 1.27.3, including CI, Docker images, the Makefile, protobuf generation, the linter configuration, and mobile build documentation. Update the installation documentation and official Linux archive checksums for Go 1.26.10 as well.
80/100 · StrongMessage clarity
✓ Descriptive subject✓ Names a concrete action or component✓ Uses a recognizable type or scope✓ Provides detailed explanatory context
Why it was queued
second-pass: unusually broad change
AI analysis · Informational 15/100
This commit is a routine maintenance update that changes which version of the Go programming language is used to build LND. It does not change any LND source code, fix a bug, or address a security vulnerability. The only changes are version numbers in build configuration files, Docker images, CI settings, and documentation.
AI review queuedMerge pull request #11373 from starius/commit-fee-rate-floorby ziggieXXX · 216caf5a · Oct 20, 2026 · 8 filesMessage 73 · AdequateModerate 62Details
Commit message · ziggieXXX
Merge pull request #11373 from starius/commit-fee-rate-floor
multi: enforce the commitment fee rate floor
73/100 · AdequateMessage clarity
✓ Specific, descriptive subject✓ Names a concrete action or component✓ Provides an explanatory body✓ Links an issue, advisory, or supporting reference
Why it was queued
signing or wallet pathmerge-commit duplicate discountsecond-pass: security-sensitive path
AI analysis · Moderate 62/100
This patch fixes a bug where LND could propose or accept Lightning channel commitment transaction fees that were too low to be valid. Specifically, with a setting of max-commit-fee-rate-anchors=1, LND would propose 250 sat/kw, below the 253 sat/kw floor its own channel state machine requires. This could prevent channels from carrying payments or cause peers to reject or fail to sign updates. The fix enforces the 253 sat/kw minimum when opening channels, updating fees, and at startup configuration.
AI review queuedMerge pull request #10919 from GeorgeTsagk/local-reputation-subsystemby Suheb · 3abd5967 · Oct 16, 2026 · 32 filesMessage 73 · AdequateInformational 21Details
Commit message · Suheb
Merge pull request #10919 from GeorgeTsagk/local-reputation-subsystem
[1/?] Local reputation: subsystem core, read only
73/100 · AdequateMessage clarity
✓ Specific, descriptive subject✓ Names a concrete action or component✓ Provides an explanatory body✓ Links an issue, advisory, or supporting reference
This commit adds a new experimental 'local reputation' feature to the Lightning Network Daemon (LND). It watches HTLC forwarding events and computes per-channel reputation scores, but only logs the results. It does not change routing decisions, alter wire messages, or write anything to disk. The feature is enabled by default but can be disabled with a config flag. The main security-relevant design is the panic boundary that prevents bugs in this new subsystem from crashing the HTLC forwarding path.
AI review queuedMerge pull request #11335 from ziggie1984/channeldb-remove-dead-v6-deserializerby ziggieXXX · a804c68c · Oct 16, 2026 · 1 fileMessage 73 · AdequateInformational 15Details
Commit message · ziggieXXX
Merge pull request #11335 from ziggie1984/channeldb-remove-dead-v6-deserializer
channeldb: remove dead v6 close summary deserializer
73/100 · AdequateMessage clarity
✓ Specific, descriptive subject✓ Names a concrete action or component✓ Provides an explanatory body✓ Links an issue, advisory, or supporting reference
This commit simply deletes an unused, deprecated helper function that was only kept around for old database migrations. Removing dead code does not change how the program behaves or introduce a security issue.
AI review queuedMerge pull request #11334 from gijswijs/fee-ceiling-clampby Yong · 65c3263c · Oct 15, 2026 · 6 filesMessage 73 · AdequateModerate 60Details
Commit message · Yong
Merge pull request #11334 from gijswijs/fee-ceiling-clamp
lnwallet/chainfee: clamp web API fee estimates to a ceiling
73/100 · AdequateMessage clarity
✓ Specific, descriptive subject✓ Names a concrete action or component✓ Provides an explanatory body✓ Links an issue, advisory, or supporting reference
Why it was queued
signing or wallet pathmerge-commit duplicate discountsecond-pass: security-sensitive path
AI analysis · Moderate 60/100
This commit adds an upper limit (ceiling) to fee estimates that LND pulls from an external web API. Previously, only a lower bound existed, so a malicious or malfunctioning fee API could return extremely high fee rates. Those inflated rates could then flow into wallet operations and cause users to overpay transaction fees. The patch clamps both per-block fee estimates and the minimum relay fee rate to a maximum of 10,000 sat/vbyte. It also ties the sweeper's maximum allowed fee rate to the same ceiling so the two limits stay consistent.
AI review queuedMerge pull request #11330 from Roasbeef/remote-next-rev-optionby ziggieXXX · 6052d0bc · Oct 15, 2026 · 21 filesMessage 73 · AdequateModerate 61Details
Commit message · ziggieXXX
Merge pull request #11330 from Roasbeef/remote-next-rev-option
lnwallet+chanstate: reject unexpected revocations, make RemoteNextRevocation an fn.Option
73/100 · AdequateMessage clarity
✓ Specific, descriptive subject✓ Names a concrete action or component✓ Provides an explanatory body✓ Links an issue, advisory, or supporting reference
This commit hardens how LND handles revocation messages during Lightning channel state updates. Previously, parts of the code treated a missing 'next revocation point' as a nil pointer, which could lead to state updates being applied in the wrong order or to invalid revocations partially updating the channel state. The change makes the optional nature of that point explicit, validates the revocation before touching stored secrets, and rejects unexpected revocations cleanly. It is a defensive bug-fix rather than a confirmed exploited vulnerability.
AI review queuedMerge pull request #11246 from Roasbeef/fix-musig2-session-lifecycleby Olaoluwa Osuntokun · 7542f418 · Oct 9, 2026 · 13 filesMessage 73 · AdequateModerate 51Details
Commit message · Olaoluwa Osuntokun
Merge pull request #11246 from Roasbeef/fix-musig2-session-lifecycle
lnwallet: clean up transient MuSig2 sessions
73/100 · AdequateMessage clarity
✓ Specific, descriptive subject✓ Names a concrete action or component✓ Provides an explanatory body✓ Links an issue, advisory, or supporting reference
Why it was queued
signing or wallet pathmerge-commit duplicate discountsecond-pass: security-sensitive path
AI analysis · Moderate 51/100
This commit fixes how LND manages temporary MuSig2 signing sessions for Taproot Lightning channels. Previously, these sessions could be left alive longer than necessary or leak if a signing round failed or was aborted. The patch releases the backing signer session as soon as the local node no longer needs it, and adds regression tests to confirm sessions are cleaned up. It also fixes a related error-handling bug where an invalid MuSig2 partial commitment signature was rejected without including the commitment height and transaction details that the remote peer needs for debugging.
AI review queuedMerge pull request #11191 from bitromortac/2604-bolt12-1hby ziggieXXX · f3a8f4e8 · Oct 1, 2026 · 21 filesMessage 58 · ThinLow 37Details
Commit message · ziggieXXX
Merge pull request #11191 from bitromortac/2604-bolt12-1h
bolt12: finalize the codec package
58/100 · ThinMessage clarity
✓ Specific, descriptive subject✓ Names a concrete action or component✓ Links an issue, advisory, or supporting reference! No meaningful explanatory body
Why it was queued
signing or wallet pathparser or protocol pathmerge-commit duplicate discountsecond-pass: unusually broad changesecond-pass: security-sensitive path
AI analysis · Low 37/100
This commit finalizes LND's new BOLT 12 codec package. It is mostly a refactor: public Encode/Decode helpers are renamed to internal encodeBech32/decodeBech32, validation helpers are unexported, and new helpers such as EncodeSigned and OfferHash are added. The most user-visible change is that the bech32 decoder no longer rejects very long strings on its own; it now expects the caller (onion-message envelope, RPC, or CLI) to enforce size limits. The commit also adds a display-only DecodeInvoiceStringUnvalidated entry point and tightens signing so that SignInvoice/SignInvoiceRequest run writer validation before producing a signature. There is no explicit bug fix or security patch in the diff; the changes are defensive API hardening.
Merge pull request #11277 from GeorgeTsagk/maxfeeratio-ceiling
[1.5/3] aux revocation: allow maxFeeRatio above 1.0
73/100 · AdequateMessage clarity
✓ Specific, descriptive subject✓ Names a concrete action or component✓ Provides an explanatory body✓ Links an issue, advisory, or supporting reference
Why it was queued
signing or wallet pathmerge-commit duplicate discountsecond-pass: security-sensitive path
AI analysis · Low 36/100
This change loosens a safety cap on Bitcoin transaction fees for a specific LND wallet RPC. Previously, callers could not allow fees larger than the transaction's total output value. Now they can opt in to a ratio up to 5x the output value, with a hard ceiling and an audit log warning. The intended use is legitimate: sweeping or re-creating tiny 'asset carrier' outputs that hold only about 1000 satoshis of Bitcoin but carry valuable off-chain assets. Without the higher cap, those operations could be blocked when on-chain fees spike. The change is defensive rather than a vulnerability fix, but it does increase the damage a compromised caller or malicious user with RPC access could do by overpaying fees.
AI review queuedbuild: skip CI for docs-only changes and run a fast subset on draftsby ziggie · 098f8445 · Sep 24, 2026 · 1 fileMessage 100 · StrongInformational 15Details
Commit message · ziggie
build: skip CI for docs-only changes and run a fast subset on drafts
Every push to a PR currently runs the full CI matrix regardless of what changed or whether the PR is still being iterated on. Introduce two measures to cut down on wasted runners while keeping all jobs in parallel, so a single run still reports as many failures as possible:
1. A ci-gate job runs first and detects documentation-only PRs, i.e. PRs that only touch files under docs/ or markdown files. Renamed files are checked under both their old and new name so that moving a source file into docs/ doesn't count as a documentation change. The cheap commit and release notes checks always run. The way the other jobs are skipped depends on whether they have a matrix:
- Non-matrix jobs (static checks, lint) and non-required jobs are skipped at the job level. A skipped job still reports its status, so the required checks in branch protection are satisfied. - Matrix jobs with required legs (unit tests, cross compilation, pinned dependencies, basic itests) are kept instantiated and skip their steps instead, in the same way the no-itest label already works. A job-level condition is evaluated before the matrix is expanded, so a skipped matrix job would never report the per-leg check names that branch protection requires and the PR would stay blocked.
All job-level conditions use !cancelled() so that a failure of the gate job itself results in the full CI running rather than everything being skipped.
2. Draft PRs only run a fast subset: static checks, lint, the pinned dependency check, the non-race unit test legs and the basic itests (18-22 min per leg). The large itests (up to 86 min per leg), Windows and macOS itests, cross compilation, the race unit legs (47-56 min) and the backwards compatibility test are skipped until the PR is marked ready for review. ready_for_review and converted_to_draft are added to the pull_request trigger types so that changing the draft state triggers a run with the right subset; the concurrency group cancels the previous one.
Push and merge_group runs carry no PR and always run everything.
100/100 · StrongMessage clarity
✓ Specific, descriptive subject✓ Names a concrete action or component✓ Uses a recognizable type or scope✓ Provides detailed explanatory context✓ Explains rationale or failure mode✓ Mentions testing or verification
This commit only changes how the project's automated GitHub test pipeline is organized. It adds a 'gate' job that decides whether a pull request only touches documentation, and if so, skips the heavy code tests. It also makes draft pull requests run a smaller, faster set of tests. There is no change to the actual Lightning Network software, wallet logic, cryptography, or network handling, so it does not create a security vulnerability in the product users run.
Merge pull request #11212 from ziggie1984/disable-legacy-channels
multi: stop opening legacy channels
58/100 · ThinMessage clarity
✓ Specific, descriptive subject✓ Names a concrete action or component✓ Links an issue, advisory, or supporting reference! No meaningful explanatory body
Why it was queued
signing or wallet pathparser or protocol pathmerge-commit duplicate discountsecond-pass: security-sensitive path
AI analysis · Moderate 62/100
This change stops LND from opening new Lightning channels using the old 'legacy' commitment format. The legacy format makes it harder to recover funds if something goes wrong, because the money owed to you is tied to a secret key that changes with each channel state. From now on, LND will either use a safer default or refuse the channel open. Existing legacy channels keep working normally; only brand-new legacy channels are blocked.
AI review queuedmulti: make RemoteNextRevocation an fn.Optionby Olaoluwa Osuntokun · 4bdb3320 · Sep 23, 2026 · 18 filesMessage 78 · AdequateLow 32Details
Commit message · Olaoluwa Osuntokun
multi: make RemoteNextRevocation an fn.Option
In this commit, we change OpenChannel.RemoteNextRevocation and ChannelCloseSummary.RemoteNextRevocation from a bare *btcec.PublicKey to an fn.Option[btcec.PublicKey], and have LightningChannel.RemoteNextRevocation return one as well. The remote party's next revocation point is legitimately absent until they send channel_ready (and, for taproot channels, until we have their nonces), so the type should say so rather than leaving it to convention.
With an Option, code can no longer move the point into a field that assumes it's always set without first deciding what to do when it's absent. Every place that needs the point now has to unwrap it: the revocation rotation, restoring a pending remote commitment, the anchor resolution and chain watcher paths for a pending remote commitment, and a retransmitted channel_ready in the peer. Each returns the new ErrNoNextRevocation if it's missing. The presence checks in the funding manager, link and peer become IsSome/IsNone.
We hold the key by value rather than by pointer, so a present point is always a real key: there's no Some(nil) to fall into. In the same spirit, InsertNextRevocation and InitNextRevocation now take the key by value, which moves the only pointer dereference to the funding manager, where the point comes straight off a decoded channel_ready message.
The on-disk format is unchanged. The revocation state still omits the point when it's absent, and the close summary still writes a presence flag ahead of it. The frozen migration packages keep their own copies of these types, so they're untouched.
We add a round trip test for a channel with no next point, and for close summaries with and without one. The peer test for a retransmitted channel_ready previously passed a channel with no next point and inserted an empty value. It now asserts that case is rejected, and that a real point makes it into the peer's copy of the channel.
78/100 · AdequateMessage clarity
✓ Descriptive subject✓ Names a concrete action or component✓ Provides detailed explanatory context✓ Mentions testing or verification
Why it was queued
signing or wallet pathsecond-pass: security-sensitive path
AI analysis · Low 32/100
This commit is a defensive code-quality change in LND that makes the 'remote next revocation point' an explicit optional value instead of a bare pointer that could be nil. It forces every piece of code that needs the point to check whether it is actually present before using it, and returns a clear error (ErrNoNextRevocation) when it is missing. The on-disk database format is unchanged, and new tests are added for channels and close summaries that lack the point. The change reduces the risk of accidental nil-pointer use or logic bugs around channels that have not yet received channel_ready, but it is not a fix for a known active exploit.
In this commit, we refine revocation processing and extend its transition coverage. The state-transition and key checks now run before their corresponding revocation bookkeeping updates.
The tests cover the rejected cases for tweakless and taproot channels, verify that their revocation bookkeeping stays unchanged, and confirm that the next valid transition succeeds.
78/100 · AdequateMessage clarity
✓ Descriptive subject✓ Names a concrete action or component✓ Provides detailed explanatory context✓ Mentions testing or verification
Why it was queued
signing or wallet pathsecond-pass: security-sensitive path
AI analysis · Moderate 59/100
This change tightens the order of checks when LND processes a revocation message from the other party in a Lightning channel. Previously, the code could add a secret to its revocation store before confirming the revocation was actually expected and matched the right commitment point. Now it checks those conditions first, so a malformed or unexpected revocation cannot partially update local state. The patch is defensive hardening rather than a clear fix for an active exploit.
Merge pull request #11223 from gijswijs/legacy-dust-retribution-fix
lnwallet+contractcourt: skip trimmed HTLCs in legacy breach retribution
73/100 · AdequateMessage clarity
✓ Specific, descriptive subject✓ Names a concrete action or component✓ Provides an explanatory body✓ Links an issue, advisory, or supporting reference
Why it was queued
signing or wallet pathmerge-commit duplicate discountsecond-pass: security-sensitive path
AI analysis · Moderate 56/100
This update fixes a bug in how LND handles old-style punishment transactions when a channel partner tries to cheat. Previously, tiny (dust) HTLCs were left as blank placeholder entries in the punishment data, which could cause the node to crash or behave incorrectly when trying to claim funds back. The fix removes those blank entries and adds safety checks further down the line.
Merge pull request #11190 from Roasbeef/zpay32-reject-duplicate-payment-hashes
zpay32: reject duplicate payment hash fields
73/100 · AdequateMessage clarity
✓ Specific, descriptive subject✓ Names a concrete action or component✓ Provides an explanatory body✓ Links an issue, advisory, or supporting reference
Why it was queued
parser or protocol pathmerge-commit duplicate discountsecond-pass: security-sensitive path
AI analysis · Moderate 62/100
This change tightens how LND reads Lightning invoices (BOLT 11). Previously, if an invoice contained more than one payment hash field, LND would silently keep the first one and ignore the rest. Now it rejects the invoice entirely. The release notes frame this as an interoperability improvement rather than a security fix, but the behavior removes a potential source of confusion or invoice manipulation where a second, different payment hash could be present but ignored.
In this commit, we reject BOLT 11 invoices that contain more than one payment hash (`p`) field. `zpay32.Decode` previously kept the first supported payment hash and ignored later fields, so the interpreted payment hash depended on field order.
The decoder now tracks whether a `p` field has appeared separately from whether its contents parsed successfully, and any later `p` field returns `ErrDuplicatePaymentHash`. This covers identical hashes, distinct hashes, and a malformed first field followed by a valid one. The change is deliberately stricter than the current BOLT 11 text, which tells a reader to use the first `p` field; it is motivated by lightning/bolts#1357, and will be backported to the v0.20.5 and v0.21.4 releases.
We also replace the stale comment above parse32Bytes, which quoted the old BOLT 11 rule that a reader must skip over 32-byte fields with an unsupported length. The current reader requirements say the opposite: a reader must fail the payment if a fixed-length field (p, h, s, n) does not have the correct length. The new comment states that reporting the field as absent is a known divergence, and explains why the end result for the payment hash is the same either way.
In the tests, we cover identical and distinct duplicate hashes, a malformed first field followed by a valid one, and the signed vectors. A valid variant of the flipped vector keeps the wrong-length h and n fields, so the coverage that such fields are skipped survives. The duplicate vector in the decode table pins ErrDuplicatePaymentHash through a new wantErr field rather than asserting any decode error.
98/100 · StrongMessage clarity
✓ Descriptive subject✓ Names a concrete action or component✓ Provides detailed explanatory context✓ Explains rationale or failure mode✓ Mentions testing or verification✓ Links an issue, advisory, or supporting reference
Why it was queued
parser or protocol pathsecond-pass: security-sensitive path
AI analysis · Moderate 62/100
This change tightens how LND reads Lightning invoices. Previously, if an invoice contained more than one payment hash field, LND would silently keep the first valid one and ignore the rest. Now it rejects such invoices outright. This prevents an attacker from potentially tricking a wallet into interpreting a different payment hash than intended, which could lead to payments being sent to the wrong destination or failing in unexpected ways. The fix is being backported to two release branches.
In this commit, we move the two tests that still opened legacy channels over to static remote key. Both produce the same commitment shape and the same fees, so no assertion moves.
The watchtower case opted into legacy deliberately. The remote signer one did not: neither of its nodes runs with committweak, so both support static remote key, and the channel only came out legacy because an empty channel_type was accepted without any check. That is the bug this PR fixes, sitting in our own suite.
The legacy node arguments and the unreferenced CfgLegacy go away with them, since committweak no longer does anything.
90/100 · StrongMessage clarity
✓ Descriptive subject✓ Names a concrete action or component✓ Provides detailed explanatory context✓ Explains rationale or failure mode✓ Mentions testing or verification
Why it was queued
signing or wallet pathsecond-pass: security-sensitive path
AI analysis · Low 33/100
This commit updates LND's integration tests to stop using an obsolete 'legacy' channel type and adds a test confirming that LND now rejects that legacy type. The commit message says the real fix is elsewhere in the same pull request: previously, an empty channel_type was accepted without checking whether the nodes actually supported the legacy format, which could cause a channel to open in an unintended, deprecated commitment shape. The diff itself only changes tests and removes unused legacy-node configuration flags; it does not contain the actual production-code fix.
Merge pull request #11200 from ziggie1984/build-go-1.27.1
build: bump Go language to 1.26.8 and toolchain to 1.27.1
73/100 · AdequateMessage clarity
✓ Specific, descriptive subject✓ Names a concrete action or component✓ Provides an explanatory body✓ Links an issue, advisory, or supporting reference
Why it was queued
signing or wallet pathparser or protocol pathmerge-commit duplicate discountsecond-pass: unusually broad changesecond-pass: security-sensitive path
AI analysis · Low 29/100
This commit updates the Go programming language version used to build LND from 1.25.13/1.26.6 to 1.26.8 (language version) and 1.27.1 (toolchain). It also regenerates many automatically generated RPC files and adjusts linter settings. The main security angle is that newer Go releases typically fix bugs and security vulnerabilities in the Go runtime and standard library, so staying current is good defensive hygiene. However, the commit itself does not describe any specific vulnerability being fixed, and the code changes are mostly version bumps and generated file updates rather than targeted security patches.
The linter configuration uses default: all, so upgrading golangci-lint also enables checks introduced after v2.4. Adopt canonicalheader, clickhouselint, godoclint, iotamixing, and unqueryvet because they add focused HTTP, documentation, const-block, and SQL safeguards.
Fix the canonical WebSocket protocol header spelling required by the expanded canonicalheader implementation. Update twelve existing doc comments to begin with their declared symbol names as required by godoclint.
Keep modernize disabled for now because its findings require broader code changes that should be reviewed separately against the Go 1.26 support baseline.
92/100 · StrongMessage clarity
✓ Descriptive subject✓ Names a concrete action or component✓ Uses a recognizable type or scope✓ Provides detailed explanatory context✓ Explains rationale or failure mode
Why it was queued
signing or wallet pathparser or protocol pathsecond-pass: security-sensitive path
AI analysis · Informational 15/100
This commit is a routine build-maintenance change. It updates the project's Go linter configuration to a newer version and fixes the resulting style warnings: a WebSocket header spelling and several documentation comments that didn't start with the correct symbol names. There is no functional code change that affects security.
AI review queuedlnrpc: regenerate protos with Go 1.27by ziggie · dd523108 · Sep 18, 2026 · 26 filesMessage 68 · AdequateInformational 15Details
Commit message · ziggie
lnrpc: regenerate protos with Go 1.27
The protobuf generator uses Go's formatting packages and is now compiled inside the Go 1.27.1 Docker image. This changes generated doc-comment formatting and causes rpc-check to report stale checked-in bindings.
Regenerate the RPC bindings with the newly pinned image. The resulting changes affect comments only; protobuf schemas and generated API code stay unchanged.
68/100 · AdequateMessage clarity
✓ Descriptive subject✓ Names a concrete action or component✓ Provides detailed explanatory context
Why it was queued
signing or wallet pathsecond-pass: unusually broad changesecond-pass: security-sensitive path
AI analysis · Informational 15/100
This commit is a housekeeping change that regenerates automatically generated RPC binding files using a newer Go compiler version. The only differences are in code comments (doc-comment formatting), not in actual program logic or API behavior. There is no security impact.
AI review queuedbuild: bump Go language and toolchain versionsby ziggie · b1f9ca66 · Sep 17, 2026 · 30 filesMessage 80 · StrongInformational 15Details
Commit message · ziggie
build: bump Go language and toolchain versions
Keep lnd compatible with the previous stable Go release by updating the language version in every tracked go.mod file to Go 1.26.8.
Move build and release tooling to Go 1.27.1, including CI, Docker images, the Makefile, protobuf generation, the linter configuration, and mobile build documentation. Update the installation documentation and official Linux archive checksums for Go 1.26.8 as well.
80/100 · StrongMessage clarity
✓ Descriptive subject✓ Names a concrete action or component✓ Uses a recognizable type or scope✓ Provides detailed explanatory context
Why it was queued
second-pass: unusually broad change
AI analysis · Informational 15/100
This commit is a routine maintenance update that changes which version of the Go programming language is used to build the lnd project. It does not change any application logic, fix a bug, or address a security vulnerability. The project is simply moving its build tooling to a newer Go release and keeping the minimum required Go version current.
AI review queuedlnwallet: reject initial commitments that leave a channel unusableby Olaoluwa Osuntokun · 99362af6 · Sep 17, 2026 · 4 filesMessage 95 · StrongLow 47Details
Commit message · Olaoluwa Osuntokun
lnwallet: reject initial commitments that leave a channel unusable
In this commit, we replace the single balance check on the initial commitment with the two checks that decide whether such a channel could ever be used: the BOLT-02 requirement the receiver of open_channel MUST apply, and a comparison of each output against the reserve its own side must maintain.
The first is the spec's, and it is what the previous revision of this commit did alone. The spec measures both to_local and to_remote against the single channel_reserve_satoshis the initiator named in open_channel, which is the reserve the initiator requires of us. It stays because it is a MUST, even though it can reject a channel the initiator could still spend on: that happens when the reserve we name in accept_channel is the lower of the two, and the spec requires the rejection anyway. That case is pinned by a test so a later change cannot drop it by accident.
The second is additive. It compares to_local against the reserve the initiator named for us and to_remote against the reserve we named for them in accept_channel, which is exactly the condition under which neither side can add an HTLC. The two reserves are independent, so this is not the same check as the first, and neither implies the other: it catches the channel the spec comparison accepts where neither side can spend, for example an initiator naming a minimal reserve of us while its own balance lands at or below the reserve we require of it.
That second check is a local policy, in places stricter than the letter of the spec, and is deliberately additive rather than a reinterpretation of the first. It costs us no usable channel: every rejection it contributes beyond the spec's is a channel where neither side can spend, which the property test asserts directly.
Both checks run in handleSingleContribution rather than in NewChannelReservation, as that's the first point where both reserves are known: the one the initiator imposes on us arrives in open_channel, while the one we impose on them is only settled once the channel acceptor (which may override our default) has had its say.
The error carries a sentinel the funding manager can match on, so the decision does not depend on the wording of the message we send the peer.
95/100 · StrongMessage clarity
✓ Specific, descriptive subject✓ Names a concrete action or component✓ Provides detailed explanatory context✓ Explains rationale or failure mode✓ Mentions testing or verification
Why it was queued
signing or wallet pathsecond-pass: security-sensitive path
AI analysis · Low 47/100
This commit tightens the rules LND uses when accepting a new Lightning channel. Before, LND only applied one BOLT-02 check on opening balances. Now it also rejects channels where, because of each side's reserve setting, neither party could ever send a payment. The change prevents 'dead on arrival' channels that look valid to the spec but cannot actually be used. It is a defensive policy fix, not a response to a disclosed vulnerability or active exploit.
AI review queuedfunding: bound the commit fee rate an initiator can impose on usby Olaoluwa Osuntokun · 0f62144d · Sep 17, 2026 · 3 filesMessage 73 · AdequateModerate 64Details
Commit message · Olaoluwa Osuntokun
funding: bound the commit fee rate an initiator can impose on us
In this commit, we add the other half of the BOLT-02 requirement here: the receiver MUST fail the channel if it considers feerate_per_kw unreasonably large. The spec leaves the threshold up to us, so we take the same approach as CLN and Eclair and cap at 10x our own estimate for the same conf target (3) we'd use for a channel we open ourselves.
A pure multiple of our estimate is a bit brittle on its own: on a quiet chain, or when the estimator falls back to the relay fee, 10x of a very small number is still a very small number, and we'd start turning away channels opened at a rate that's perfectly sane in absolute terms. So we floor the bound at 100 sat/vbyte, which no realistic mempool needs a commitment to beat. If the estimator errors out entirely we fall back to that floor rather than failing the flow.
That conf target is now a single constant shared with the rate we pay on a channel we open ourselves, so the two cannot drift apart: the point of the bound is to track the rate we'd consider necessary for a timely unilateral close, and a second hardcoded 3 would let those diverge silently.
The bound is consulted only after the cheap admission checks have passed, so a peer that is over its pending-channel limit cannot drive fee estimation work on the funding goroutine.
Since all channels are single funder today, an initiator that proposes an absurd rate mostly burns its own capacity. But it doesn't only do that: the open still costs us a pending channel slot, and left unchecked it lands us with a confirmed channel that can never be used and has to be closed on chain. Rejecting up front is cheaper for both of us.
The error carries a sentinel the funding manager can match on, so the decision does not depend on the wording of the message we send the peer.
73/100 · AdequateMessage clarity
✓ Specific, descriptive subject✓ Names a concrete action or component✓ Provides detailed explanatory context
Why it was queued
signing or wallet pathsecond-pass: security-sensitive path
AI analysis · Moderate 64/100
This change prevents a remote user from opening a Lightning channel with an outrageously high on-chain fee. Before the patch, an initiator could set a commitment fee so large that most of the channel's money would be eaten by fees, leaving a channel that is effectively unusable and must be closed on the blockchain. The patch caps the fee a remote peer can impose at ten times LND's own fee estimate, with a hard floor of 100 sat/vbyte, and rejects the channel if the proposed fee exceeds that cap.