메뉴
AI-powered security scanner — OWASP Top 10, CWE Top 25, KMIP authorization, FIPS gating, memory safety, side-channel, supply chain, and 20 vulnerability families. Use when asked to review code security, audit KMIP access control, or scan for vulnerabilities.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | security-review |
| description | AI-powered security scanner — OWASP Top 10, CWE Top 25, KMIP authorization, FIPS gating, memory safety, side-channel, supply chain, and 20 vulnerability families. Use when asked to review code security, audit KMIP access control, or scan for vulnerabilities. |
An AI-powered security scanner that reasons about code the way a security researcher would — tracing data flows, understanding component interactions, and catching vulnerabilities that pattern-matching tools miss. Covers all major software vulnerability families relevant to a Rust + TypeScript KMS with FFI to OpenSSL and PKCS#11 HSMs.
is_allowed() calls, unauthorized key material access)unsafe blocks), side-channel resistance, supply chain integrityFollow these steps in order:
Read .github/skills/shared/anti-hallucination.md before any analysis. All rules in
that file are mandatory for this skill. Do not proceed to Step 1 until you have read it.
If a path was provided (e.g. /security-review crate/server/src/core/), scan only that scope.
If no path given, scan the entire project starting from the root.
Identify the primary language(s): Rust (main), TypeScript (UI), Bash (scripts).
Cargo.toml and Cargo.lock for known vulnerable crates (CVEs in cargo audit watchlist).
openssl, ring, rustls, hyper, actix-web, serde.ui/pnpm-lock.yaml for vulnerable npm packages..mise/scripts/ for insecure subprocess calls.Scan ALL files (config, CI, Dockerfiles, shell scripts, Rust source) for:
.env files accidentally committeddebug!, trace!, info! with sensitive variable names)Flag any // TODO: debug — remove before shipping markers left in the codebase.
crate/server_database/src/ — raw query construction with user input.mise/scripts/ — $variable without quotingcrate/server/src/routes/crate/server/src/core/operations/ that return or modify key material without calling the access-control check (typically db.is_allowed(...) or equivalent in crate/access/)
KmipObject (Get, Export, Decrypt, Sign, MAC, Derive) must verify the caller owns or has been granted access to the key UIDLocate operations must filter results to objects the caller can access — not return global object listscrate/server/src/middlewares/ — alg:none, weak secrets, missing expiry validationinsecure feature flag code that disables auth must be strictly gated with #[cfg(feature = "insecure")]crate/access/ module — check that grant/revoke operations require owner or admin privilegeReview each file in crate/server/src/core/operations/:
Locate scope results to the authenticated user's owned + granted objects?Create, Register, Import operations linking the new object to the authenticated user as owner?EnsureAuth middleware being enforced?#[cfg(feature = "non-fips")] usage — verify it is at function/module level, never inline inside function bodies#[cfg(feature = "non-fips")] gatescrate/server/src/start_kms_server.rs scope registrations for non-FIPS routes are wrapped in #[cfg(feature = "non-fips")] { ... }insecure feature does not bypass FIPS algorithm enforcementDisplay implszeroize crate usage on key material)crate/server/src/routes/aws_xks/, etc.) — verify URL validationcrate/crypto/src/rand::thread_rng() used for key generation instead of rand::rngs::OsRngapply_openssl_dir_env_if_needed() is called before any Provider::try_load()dangerouslySetInnerHTML, unescaped user content in React componentsui/src/Critical for Rust unsafe blocks and FFI to OpenSSL / PKCS#11 / HSM libraries:
crate/hsm/*/src/, crate/crypto/src/)crate/kmip/src/)unsafe blocks — require // SAFETY: comment; flag every unsafe without one#[serde(deny_unknown_fields)] on security-critical structsui/src/) — Object.assign or spread with user-controlled keysis_allowed() check and state mutationOnceLock / Mutex / RwLock — deadlock potential or poisoned lock propagation in error pathsCreate + Destroy on the same UID — race to key existenceregex::Regex::new(...)) and TypeScript for catastrophic backtrackingReKey across large object sets)== on &[u8] secrets instead of constant_time_eqRUSTSEC-2023-0071 in SECURITY.md)alg:none / algorithm substitution — alg header not validated against server-side expected algorithm listkid injection — path traversal or SQL injection via the kid header fieldjku / x5u injection — SSRF via header-specified key URL (server fetches attacker-controlled URL)exp) validation — check insecure feature gate does not disable it in production buildscode_verifier is actually validated server-sideSameSite cookie attribute, absent CSRF tokens on state-changing endpointsX-Frame-Options or frame-ancestors CSP directivereturn_url, redirect, next parameters in auth flowsContent-Security-Policy, Strict-Transport-Security, X-Content-Type-Options, Referrer-Policy, Permissions-PolicyAccess-Control-Allow-Origin combined with Access-Control-Allow-Credentials: trueCargo.toml audit with cargo deny check advisoriescosmian_*) potentially claimable on crates.io*, >=) that could admit malicious patch releasestest_data submodule points to the expected Cosmian repo and branchCargo.lock and ui/pnpm-lock.yaml integrity (unexpected changes without corresponding Cargo.toml / package.json changes)Destroy), privilege grants/revokes, key export — all must be loggeddebug! / trace! / info! callsDestroy operations are irrecoverable — must produce a durable audit log entryDestroy be called on an Active key without first calling Revoke/Deactivate?Activation Date, Deactivation Date, Sensitive, or Extractable attributes?Non-Exportable attribute must be blocked in Export / Get operationsLocate information disclosure: Locate returning objects owned by other principals (even metadata without key material is information leakage)Register / Import overwrite an existing object UID owned by a different user?Trace user-controlled input from entry points (HTTP params, KMIP request fields, TTLV-deserialized bytes) to sinks (DB queries, exec calls, key material output, file writes).
Check for vulnerabilities that only appear when examining multiple files together:
For each finding:
Output the report in this format:
## Security Review: [scope]
### Summary
| Severity | Count |
|----------|-------|
| 🔴 CRITICAL | N |
| 🟠 HIGH | N |
| 🟡 MEDIUM | N |
| 🔵 LOW | N |
| ⚪ INFO | N |
### Findings
#### [SEVERITY] [Category]: [Title]
- **File**: `path/to/file.rs:line`
- **Vulnerable code**: `<snippet>`
- **Risk**: What can an attacker do with this?
- **Confidence**: High / Medium / Low
- **Fix**: See patch below
### Dependency Audit
[CVEs or clean]
### Secrets Scan
[Findings or clean]
### KMIP Authorization Audit
[Missing access checks or clean]
### FIPS Gating Audit
[Inconsistent gating or clean]
For every 🔴 CRITICAL and 🟠 HIGH finding, generate a concrete patch:
State explicitly: "Review each patch before applying. Nothing has been changed yet."
For every 🔴 CRITICAL and 🟠 HIGH finding where the fix is already present in the codebase
(i.e., the vulnerability existed before and is now resolved), add a new entry to
SECURITY.md under the current year using this format:
#### COSMIAN-<YYYY>-<NNN> — <Short title matching the finding title>
| Field | Value |
| ---------- | ----- |
| Severity | <Critical / High / Moderate / Low> |
| Published | <DD Month YYYY> |
| Affected | from <first affected version> before <fix version> |
| Fixed in | <version or "next release"> |
| Found by | <source, e.g. "Copilot security-review"> |
| References | <PR or issue link if available, else —> |
**Summary:** <Two-to-four sentences: what the vulnerability was, how it could be triggered.>
**Impact:** <What an attacker could do.>
**Mitigation:** <What operators must do: upgrade, config change, workaround.>
Numbering: read the highest existing COSMIAN-<YYYY>-NNN in SECURITY.md for the current
year and increment by 1. If no entry exists for the current year, start at 001.
Also add the entry to the table of contents and the summary table at the bottom of SECURITY.md.
If the finding is not yet fixed (i.e., it is a new finding requiring a patch), do not
add it to SECURITY.md yet — it will be added after the fix is merged and verified.
/threat-modelIf any finding involves a trust boundary change — a new authentication method, a new unauthenticated endpoint, a new outbound call, a new inter-service trust, or a new HSM/cloud integration — append this note to the report:
⚠️ Architectural review recommended: this finding affects a trust boundary. Run
/threat-model(incremental mode) to update the STRIDE-A threat inventory before merging./security-reviewcovers tactical code vulnerabilities;/threat-modelcovers system-level threats that may not be visible at the code level.
| Severity | Meaning | Example |
|---|---|---|
| 🔴 CRITICAL | Immediate exploitation risk, data breach likely | Key material returned without auth check, RCE |
| 🟠 HIGH | Serious vulnerability, exploit path exists | BOLA/IDOR on KMIP objects, hardcoded secrets |
| 🟡 MEDIUM | Exploitable with conditions or chaining | Missing rate limit, SSRF with partial control |
| 🔵 LOW | Best practice violation, low direct risk | Verbose errors, missing security header |
| ⚪ INFO | Observation worth noting, not a vulnerability | insecure feature left enabled in a config file |
Write the next version entry at the top of CHANGELOG.md by summarizing all changes since the last tagged release. Use when preparing release notes.
Audit GitHub Actions workflows for efficiency and recommend fixes to reduce CI minutes and costs. Use when asked to improve CI performance.
Fix CI failures in a loop until all GitHub workflow runs on the current branch are green. Use when CI is failing and needs automated repair.
Comprehensive cryptographic audit: FIPS 140-3, BSI TR-02102, ANSSI, NIST SP 800-series compliance, algorithm allow-list, key sizes, feature-flag gating, OpenSSL provider init, key lifecycle, multi-standard matrix, and academic cryptanalysis cross-check. Use when touching crate/crypto/, algorithm selection, or key management code.
Comprehensive security audit orchestrator: invokes /security-review, /cryptography-review, /threat-model, and /standards-review in sequence. Produces a unified go/no-go report. Use for full security audit before release or after significant changes.
Release readiness gate: runs all AI audit skills in sequence and produces a go/no-go report. Use before triggering the release workflow.