name	crypto-review
description	Crypto-misuse review — non-constant-time comparison of secrets (timing side-channels), missing/compiler-elidable zeroization of secrets, and non-cryptographic RNG used to mint secrets. The crypto-reviewer agent confirms each candidate handles a secret and promotes real issues into .kuzushi/findings.json (source "crypto-review"). Distinct from /sast (injection) and /sharp-edges (API design).
context	fork
agent	crypto-reviewer
user-invocable	false

Crypto review

Find the crypto-implementation footguns that look fine but leak or weaken secrets.

Run node "${CLAUDE_PLUGIN_ROOT}/scripts/cmd/crypto-review-prepare.mjs" --target "<repo root>" (optionally --input '{"maxCandidates":30}'). If it reports no-candidates, say so and stop. Read the prep's prepPath.
For each candidate, open the site and first confirm the value is secret-derived, then assess by category: timing-side-channel (variable-time compare of a secret), missing-zeroization (secret not securely wiped / elidable memset), or weak-crypto-rng (non-CSPRNG minting a token/key/nonce). Decide finding / candidate / rejected.
Write the { candidates: [...] } bundle to the prep's draftPath, then run the assembleCommand — it validates verdicts/categories and promotes them into .kuzushi/findings.json (source:"crypto-review").
Report findings by category (file:line, the leak/weakness, and the constant-time / secure-zero / CSPRNG fix to use).

On code with no cryptographic / secret handling — nothing to review.
For injection or API-design misuse — that's /sast, /taint-analysis, and /sharp-edges.

"Just a string compare." → If a side is a secret (MAC/token/hash), a variable-time compare leaks it; confirm secret-derivation before clearing.
"memset wipes it." → A plain memset on a dying buffer is often optimized away; only a secure-zero primitive counts.
"The RNG is random enough." → Math.random/rand() are predictable; for secrets that's a finding.

name	crypto-review
description	Crypto-misuse review — non-constant-time comparison of secrets (timing side-channels), missing/compiler-elidable zeroization of secrets, and non-cryptographic RNG used to mint secrets. The crypto-reviewer agent confirms each candidate handles a secret and promotes real issues into .kuzushi/findings.json (source "crypto-review"). Distinct from /sast (injection) and /sharp-edges (API design).
context	fork
agent	crypto-reviewer
user-invocable	false

Find the crypto-implementation footguns that look fine but leak or weaken secrets.

Run node "${CLAUDE_PLUGIN_ROOT}/scripts/cmd/crypto-review-prepare.mjs" --target "<repo root>" (optionally --input '{"maxCandidates":30}'). If it reports no-candidates, say so and stop. Read the prep's prepPath.
For each candidate, open the site and first confirm the value is secret-derived, then assess by category: timing-side-channel (variable-time compare of a secret), missing-zeroization (secret not securely wiped / elidable memset), or weak-crypto-rng (non-CSPRNG minting a token/key/nonce). Decide finding / candidate / rejected.
Write the { candidates: [...] } bundle to the prep's draftPath, then run the assembleCommand — it validates verdicts/categories and promotes them into .kuzushi/findings.json (source:"crypto-review").
Report findings by category (file:line, the leak/weakness, and the constant-time / secure-zero / CSPRNG fix to use).

On code with no cryptographic / secret handling — nothing to review.
For injection or API-design misuse — that's /sast, /taint-analysis, and /sharp-edges.

"Just a string compare." → If a side is a secret (MAC/token/hash), a variable-time compare leaks it; confirm secret-derivation before clearing.
"memset wipes it." → A plain memset on a dying buffer is often optimized away; only a secure-zero primitive counts.
"The RNG is random enough." → Math.random/rand() are predictable; for secrets that's a finding.

crypto-review