soundcheck

Detects supply-chain risks in dependency manifests, lockfiles, install commands, and CI pipelines — missing lockfiles, absent vulnerability scanners, unverified installs, install hooks that pipe remote content to a shell, and AI-hallucinated package names. Use when writing package install commands, dependency manifests, CI/CD configs, build scripts, or code that loads external packages.

Deep review that audits API contracts for mismatches between what callers assume and what implementations enforce. Designed for nightly CI, pre-release scans, or manual deep audits — runs in tens of minutes, not seconds. Surfaces bugs that single-pass OWASP review misses — caller/callee invariant gaps, trust-anchor confusion, predicate misnaming.

Runs a full OWASP/CWE security audit via isolated subagents. Use when the user types /security-review or explicitly requests a full security audit of the current code.

Detects weak password storage, flawed JWT validation, sessions surviving logout, and missing MFA in authentication code. Use when writing login flows, session management, password storage, JWT handling, MFA implementation, credential validation, or any code that authenticates users. Also invoke when implementing "remember me" functionality or API key management.

Detects missing ownership checks, broken role enforcement, and IDOR vulnerabilities that let users access other users' resources. Use when writing code that checks authorization, enforces resource ownership, handles IDOR (object-level authorization), processes server-side requests to external URLs, or implements access control middleware.

Detects multi-threaded code where a lock is held across a blocking operation, lock acquisition order risks deadlock, lock-free atomics use the wrong memory ordering, or double-checked locking is missing the necessary barrier. Use when writing or modifying code that acquires multiple locks, uses atomic operations with explicit memory orders (std::memory_order_*, atomic.LoadAcquire, Ordering::Relaxed), implements lazy initialization with double-checked locking, or calls a blocking operation (I/O, sleep, await, blocking channel send) inside a locked region.

Detects library-internal cryptographic correctness bugs that pattern matchers and crypto-failures skills miss — AEAD nonce reuse, ECDSA k-value reuse, length-extension on bare hashes, padding-oracle exception distinguishability, and branching on secret material. Use when writing code that calls a crypto primitive directly (AEAD encrypt/decrypt, ECDSA/DSA sign, raw hash, RSA decrypt). Distinct from cryptographic-failures, which covers application-layer choices like MD5 for passwords; this skill is about how primitives are wired up.

Detects weak or broken cryptography that lets attackers recover plaintext passwords, forge tokens, or decrypt sensitive data. Use when writing code that encrypts or decrypts data, hashes passwords or tokens, generates random values for security purposes, manages cryptographic keys, or configures TLS/SSL settings. Also invoke when storing sensitive data at rest.

Detects forms and state-changing endpoints missing CSRF protection. Use when writing HTML forms that submit POST/PUT/DELETE requests, configuring session cookies, or setting up CSRF middleware for web applications. Also invoke when disabling or bypassing CSRF protections in framework configuration.

Detects error handlers that leak stack traces or fail open on exception. Use when writing error handlers, exception catching blocks, try/catch/finally constructs, or API error responses. Also invoke when an application could fail open on exception, or when stack traces might reach end users.

Detects autonomous agents that take irreversible or high-impact actions without human approval. Use when building autonomous LLM agents, implementing multi-step agent pipelines, writing code where LLM output triggers real-world actions (file writes, API calls, emails, database changes, code execution), or designing agentic workflows with tool use.

Detects file upload handlers that accept executable content or write to web- accessible paths without validation. Use when writing file upload handlers, processing multipart form data, saving uploaded files to disk or cloud storage. Also invoke when accepting user-supplied filenames or storing uploads in a web-accessible directory.

Detects GraphQL schemas without depth limits, cost analysis, or introspection controls. Use when writing GraphQL schemas, resolvers, or server configuration. Also invoke when setting up Apollo Server, graphql-yoga, Strawberry, gqlgen, or any GraphQL framework without explicit depth limiting, cost analysis, or introspection controls.

Detects API keys, passwords, tokens, and credentials embedded directly in source code. Use when writing code that contains API keys, passwords, tokens, connection strings, or private keys as string literals. Also invoke when embedding credentials in configuration files, environment setup scripts, or test fixtures that could be committed to version control.

Detects HTTP response header construction from user input vulnerable to CRLF injection. Use when writing code that sets HTTP response headers using values from user input, request parameters, or external data. Also invoke when constructing email headers, Content-Disposition filenames, or Location headers from caller-controlled strings.

Detects missing security controls — rate limits, MFA, re-auth — that should have been designed in from the start. Use when designing authentication flows, implementing rate limiting, building business logic for financial or access-sensitive operations, or writing code that enforces security requirements at the application layer.

Detects sensitive data written to unprotected local files, preference stores, or SQLite. Use when writing code that stores sensitive data (credentials, tokens, PII) to local files, platform preference stores (NSUserDefaults, SharedPreferences, UserDefaults), SQLite databases, or localStorage without encryption at rest.

Detects unsafe rendering or execution of LLM output that enables XSS, command injection, or second-order injection. Use when writing code that renders LLM output to a UI, executes LLM-generated code or shell commands, or passes LLM responses to downstream APIs or databases. Also invoke when displaying markdown or HTML from an LLM response.

Detects LLM tool definitions without parameter validation, allowing prompt injection to escalate into file/shell/network access. Use when writing LLM tool definitions, function schemas for tool use, plugin or extension handlers, or any code that defines what actions an LLM can take via tools. Also invoke when implementing tool parameter validation.

Detects unsafe deserialization, unverified software updates, and tampered CI artifacts. Use when writing deserialization code, processing pickled or marshalled data, implementing software update mechanisms, consuming CI/CD artifact downloads, or handling data from untrusted sources that gets reconstructed into objects.

Detects IPC receivers that accept input without verifying caller identity. Use when writing URL scheme handlers, Android intent receivers or exported activities, named pipe or socket listeners, XPC service handlers, or any IPC endpoint that processes caller-supplied input without validating the caller's identity or origin.

Detects compromised or backdoored models loaded from unverified sources, floating tags, or unreviewed registries. Use when writing code that downloads pre-trained models, loads models from registries or file paths, integrates third-party LLM providers, or manages model version selection. Also invoke when setting up automated model updates.

Detects missing security event logs, logged secrets, and log injection through user input. Use when writing logging code, audit trails, error handlers that log, authentication event recording, or any code that writes security- relevant events. Also invoke when logging user inputs, API responses, or system actions that touch sensitive data.

Detects ORM create/update calls that spread request bodies without an explicit field allowlist. Use when writing code that creates or updates database records from request body, form data, or JSON input. Also invoke when spreading, merging, or destructuring request payloads directly into ORM model create/update calls without an explicit field allowlist.

Detects MCP tool handlers vulnerable to malicious inputs, hardcoded secrets, or unrestricted file/shell access. Use when writing MCP server definitions, tool schemas, or tool handler code. Also invoke when registering tools with Claude or building Claude Code extensions that expose file system, shell, or network access.

Detects function-local misuse of memory and resource APIs in C, C++, and Rust unsafe — allocations whose return value is not checked, frees on error paths that race the success path, locks initialized incorrectly, file descriptors leaked across exec. Use when writing or modifying C or C++ code that calls malloc/calloc/realloc/free, mmap/munmap, pthread_mutex_*, fopen/open, or any kernel/library memory or resource primitive. Use when writing Rust code inside an unsafe block that calls a raw allocation or pointer API.

Detects LLM endpoints missing token caps, rate limits, or prompt-length bounds, enabling cost and resource exhaustion. Use when writing LLM API call handlers, setting up inference endpoints, implementing chatbot backends, or configuring token limits for LLM services. Also invoke when accepting user- provided prompts without length constraints.

Detects inference endpoints without authentication or throttling, allowing model weight reconstruction. Use when writing inference API endpoints, deploying LLM-serving infrastructure, implementing model access controls, or configuring rate limiting and authentication for model endpoints.

Detects agent-to-agent calls without authentication, authorization, or permission scoping. Use when writing code that calls other agents, spawns subagents, builds multi-agent pipelines, or passes messages between LLM agents. Also invoke when an orchestrator delegates tasks to worker agents or when agents share tools and permissions.

Detects MongoDB and other NoSQL queries that mix user input into operators or filters without validation. Use when writing MongoDB queries, Elasticsearch queries, or other NoSQL database operations that include user-supplied input. Also invoke when building query filters from request parameters or constructing aggregation pipelines with dynamic values.

Detects untrusted numeric input that flows through a conversion, cast, or parser into a length, size, index, or authorization comparison without bounds-checking the post-conversion value. Use when writing or modifying code that calls atoi/strtol/parseInt/strconv.Atoi on user-supplied or network-supplied input, casts between signed and unsigned integer types, narrows an integer width before a bounds check, or uses an untrusted integer as an allocation size, array index, file offset, length argument to memcpy, or comparison gate for authorization or permission decisions.

Detects weak JWT validation, loose redirect_uri matching, and missing state parameters in OAuth/OIDC flows. Use when writing OAuth2 or OpenID Connect flows, JWT validation logic, token endpoint handling, or redirect URI processing. Also invoke when implementing any code that parses or verifies JWTs.

Detects redirects to user-controlled URLs that enable phishing and OAuth callback abuse. Use when writing code that redirects users to a URL from request parameters, form input, or any caller-controlled source. Also invoke when building login flows with "return to" URLs or OAuth callback redirects.

Detects systems that treat LLM output as authoritative fact in consequential decisions without human review. Use when writing code that displays LLM output as authoritative fact, uses LLM decisions to gate consequential outcomes without human review, or builds automated pipelines where LLM judgment drives downstream actions.

Detects file operations with user-controlled paths vulnerable to ../ traversal, absolute paths, or symlink escapes. Use when writing code that opens, reads, writes, or deletes files using paths constructed from user input. Also invoke when serving static files, handling file downloads by name, or resolving paths from URL parameters or request bodies.

Detects unsafe privilege handling in SUID/SGID binaries, environment-variable trust in privileged code, insecure umask, temp-file races, and symlink-follow bugs in /tmp. Use when writing or modifying SUID/SGID-installed code, code that drops or elevates privileges via setuid/setgid/seteuid/seteuid_r, anything that reads PATH/LD_*/IFS environment variables, code that creates files in /tmp or other world-writable directories, or anything that opens a file whose path may contain a symlink controlled by a less-privileged user.

Detects LLM prompts built from user input or retrieved documents that let attackers hijack the model's instructions. Use when writing code that constructs LLM prompts from user input, builds system prompts, implements RAG pipelines, or processes external documents fed to a model. Also invoke when external data can influence LLM context.

Detects JavaScript deep-merges or property assignment from user input that can pollute Object.prototype. Use when writing JavaScript or TypeScript code that deep-merges, clones, or extends objects using user-controlled input. Also invoke when using lodash merge/set, Object.assign with dynamic keys, or recursive property copy on untrusted data.

Detects check-then-act sequences on shared state vulnerable to TOCTOU manipulation. Use when writing check-then-act sequences on shared state, file operations that check existence before read/write, balance updates without locking, or any code where two operations on the same resource are not atomic.

Detects RAG pipelines that ingest external documents into LLM context without sanitization or trust gating. Use when building RAG pipelines, ingesting external documents into vector stores, or retrieving content from external sources to include in LLM context. Also invoke when writing code that fetches URLs or parses documents for LLM prompts.