Test design, vulnerability assessment, and quality assurance (13K skills)
Audit code, frontend behavior, and agent traces for ignored errors, fallbacks that hide failures, and unsupported success reports. Findings identify locations, severity, user impact, and evidence, with repair handoffs based on reproduced failures.
Design and summarize fair comparisons of coding agents on reproducible tasks. Assess quality, correctness, cost, time, tool use, and verification separately, then recommend an agent for a specific scenario using observed run evidence.
Investigate an AI agent that stalls, repeats tool calls, or loses its goal. Compare possible causes against recorded evidence and prepare a small, reversible recovery step, with unresolved questions clearly identified.
Configure a shared guard that blocks destructive shell commands before local AI agents execute them. Adjust blocking rules, connect agent integrations, and test why commands are blocked or allowed.
Check assumptions before changes to APIs, pricing, quotas, or defaults. This skill combines research, measurements from real services, documented results, and product-owner approval to assess customer impact beyond passing tests, with follow-up verification after deployment.
Review dependency handling, package downloads, credentials, and deployed file cleanup in APM. The skill defines checks for safe paths, centralized credential access, and refusing content when its integrity or signature cannot be verified.
Review OpenAPM specification changes through four specialist perspectives covering schemas, registries, dependency resolution, and web architecture. The skill combines findings and mechanical checks into one advisory comment, separating immediate fixes from work deferred to later versions.
Test local Tika changes with Jazzer, a tool that mutates inputs to uncover failures. Build parser fuzzers, check existing files, reproduce crashes, add starting samples, and retest fixes for memory exhaustion, hangs, and unexpected exceptions.
Inspect files with Apache Tika to identify their actual format, embedded content, revision traces, and risk indicators. Save extracted evidence and file hashes while distinguishing recorded claims from observed structure, without asserting authorship or validating signatures.
Compare Tika character encoding detection using one build and two configurations. Find files whose detected encoding changed, cross-check text quality signals, and inspect detector traces to investigate garbled text and identify which detector contributed to the result.
Compare text and metadata extracted by two Tika builds from the same file collection. Generate reports to investigate content differences, encoding changes, new exceptions, and missing embedded documents, then repeat the evaluation after fixes.
Checks whether work is truly complete before reporting success, committing changes, or creating a pull request. It runs the relevant checks, reads their full output, and ties status reports to current evidence.
Investigate bugs, failing tests, and unexpected behavior before proposing changes. Follow four stages to gather evidence, compare working examples, test a specific cause, and verify a focused fix with a test that reproduces the problem.
Scan a website for broken links, including links that return 404 or 500 errors. Crawl internal pages, identify broken outbound links, and report the pages containing them so you can locate and fix the references.
Diagnose bugs, intermittent failures, and performance regressions through structured logging and reproduction. It ranks possible causes with confidence scores, tests competing explanations, and reports whether the evidence proves the cause or leaves the diagnosis incomplete.
Guide app automation on iOS, tvOS, macOS, and Android using help matched to the installed agent-device CLI. Tasks include screen inspection, taps, text entry, scrolling, and collecting logs, network information, or performance evidence.
Review a Marin access-change pull request by decrypting user identities and explaining their roles and resources. After explicit human confirmation, merge the request and apply only the verified permission changes through Pulumi.
Compare Zephyr changes using the same prepared dataset and matching execution settings. Analyze Finelog metrics for each processing stage to distinguish CPU efficiency, elapsed time, and peak memory changes, then report improvements, regressions, or tradeoffs.
Evaluate whether a proposed LinkedIn integration fix addresses its linked issue by reviewing the report, code changes, and tests. Flag implementation concerns and optionally compare live results with a genuine recorded baseline after approval.
Investigate LinkedIn integration bugs by comparing issue reports with source code. Separate established facts from unverified claims, identify missing evidence, and optionally check reproduction in a live environment after explicit approval.
Review a git diff or selected files with parallel agents to find behavior regressions, security risks, reliability problems, and compatibility or test gaps. Receive filtered, evidence-backed findings and prioritized recommendations without changing files.
Check whether infrastructure code actually runs when a user triggers it. Trace calls from hooks or commands through scripts and modules, verify registration and permissions, and test the complete path to uncover unreachable code.
Develop features and fix bugs through test-driven development: write a failing test, add minimal code to pass, then clean up. The workflow assigns agents to planning, test writing, implementation, and validation of the full test suite.
Run a staged testing workflow that checks type errors first, executes unit and integration tests in parallel, and runs end-to-end tests only after they pass. Receive pass and fail counts with details about failures.