Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
61 changes: 61 additions & 0 deletions .github/ai-review/config.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,61 @@
# Review config for the multi-model AI review (yetanotherco/actions pr_review_multi.yml), read from the
# PR base branch. RISC-V zkVM: executor, STARK prover/verifier, recursion.
# Layout checked with `git ls-tree` of main (2026-11-06): bin/ crypto/{crypto,ecsm,math,math-cuda,stark}
# executor/ prover/{src/constraints,src/tables,src/recursion.rs} syscalls/ spec/ tooling/ bench_vs/.
threat_model = "threat_model.md"

# Areas are Cargo packages (dirs with a Cargo.toml), else the first `unit_depth` path components.
unit_depth = 2

# Soundness-critical code. Files not listed are reviewed as sensitive too (default_sensitive is on and
# low_risk is empty), so every part gets every reviewer; this list orders the review of a large PR.
sensitive = [
"prover/**", # AIR tables, constraints, LogUp/bus, trace generation, recursion.rs
"crypto/stark/**", # STARK prover and verifier, FRI, transcript
"crypto/math/src/**", # field and polynomial arithmetic
"crypto/crypto/src/**", # hashes, Merkle commitments, Fiat-Shamir
"crypto/ecsm/src/**", # elliptic-curve precompile
"executor/src/vm/**", # instruction semantics the constraints must match
"syscalls/src/**", # guest/host ABI, private-input and commit paths
"**/constraints/**",
"**/*verif*",
"**/*recursion*",
"**/recursion/**",
"**/continuation*/**",
"formal/**", # Lean / Clean models (not present on main yet)
"Cargo.toml", # workspace dependencies (supply chain)
"rust-toolchain*",
]

# Not reviewed (in addition to lockfiles, binaries and the built-in generated/vendored patterns).
ignore = [
"bench_vs/**", # comparison benchmarks against other zkVMs
"**/*.elf",
"**/*.log",
"**/*.log/**",
"executor/programs/**/*.json",
"crypto/math-cuda/kernels/**/*.ptx",
]

# Extra test globs (on top of tests/, test/, *_test.rs, ...).
test_globs = [
"executor/programs/asm/test_*",
"tooling/ethrex-tests/**",
"tooling/ethrex-fixtures/**",
"crypto/math/fuzz/**",
]

# Reviewers for parts with no sensitive file (only used if low_risk or default_sensitive = false makes
# some files non-sensitive).
nonsensitive_lanes = ["claude", "gpt", "kimi"]

# Parts matching these are reviewed first and, in a split PR, get twice the per-reviewer budget: keep
# the list to the code where a bug breaks soundness.
priority = [
"crypto/stark/src/**",
"prover/src/constraints/**",
"prover/src/tables/**",
"prover/src/recursion*",
"executor/src/vm/**",
"syscalls/src/**",
]
46 changes: 0 additions & 46 deletions .github/ai-review/matrix.json

This file was deleted.

37 changes: 0 additions & 37 deletions .github/ai-review/prompts/general.md

This file was deleted.

10 changes: 0 additions & 10 deletions .github/ai-review/prompts/lanes/verify.md

This file was deleted.

19 changes: 19 additions & 0 deletions .github/ai-review/threat_model.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,19 @@
# yetanotherco/lambda_vm

RISC-V zkVM: execution trace generation, a STARK prover and verifier, recursion and
continuations, and formal models (Lean / Clean) under `formal/`.

Priorities, soundness above all:
- Underconstrained columns or constraints: a value the prover can choose freely that the
constraints should pin down.
- Missing range checks (limbs, bytes, carries, multiplicities).
- LogUp / bus errors: wrong multiplicities, unbalanced sends and receives, a bus term that a
malicious prover can cancel.
- Integrity of preprocessed and committed tables (a table the verifier trusts but does not fix).
- Padding rows that have side effects (bus interactions, memory writes) or skip constraints.
- The verifier accepting an invalid proof; missing checks on proof shape or public inputs.
- Continuation / recursion linking errors (state handed from one segment or layer to the next
without being bound).
- Panics or unbounded work in the verifier on a malformed proof, including allocations or
recursion sized by proof-supplied counts, lengths or nesting.
- Formal models that diverge from the implementation, or prove a weaker statement than claimed.
Loading
Loading