Test design, vulnerability assessment, and quality assurance (13K skills)
Investigate recurring bugs or failures by repeatedly asking why and checking code, tests, logs, and git history. Produce an evidence-based cause chain, distinguish guesses, and propose fixes for both the root cause and the symptom.
Identify high impact code that tests may miss, then check a property using formal verification: systematic checks within stated assumptions and limits. Generate a project specific skill so future agents can rerun the check and keep it aligned with code changes.
Update formal verification, mathematical checks of specified behavior, after code changes. The skill reviews relevant commits, aligns models and verification files with the current implementation, reruns checks, and distinguishes outdated verification from actual code bugs.
Run two review agents in parallel to examine code changes for bugs, security issues, and maintainability problems. Combine their evidence and file references into prioritized findings, removing duplicates and highlighting unresolved uncertainty.
Audit branch or PR changes for bugs, security vulnerabilities, and broken behavior. The review traces effects across modules, checks local development workflows, and looks for features exposed despite flags or internal access restrictions.
Fix bugs with a focused regression test, a check that catches the same problem returning. Confirm it fails for the right reason before the fix and passes afterward, using another practical check when test setup is too costly.
Check performance measurements and AI evaluation results before reporting or acting on them. Identify what limits the result, rule out misleading measurements, and keep the number of runs, variation, and supporting evidence alongside each reported figure.
Verify completed work by running the actual feature, reading real values, and inspecting changes before reporting success. Checks both the outcome and the evidence behind it, favoring repeatable scripts that reviewers can run themselves.
Design and review tests around results users or calling systems can observe. Check that tests detect broken behavior while still passing after internal restructuring, with expected results drawn from requirements rather than the tested code.
Create a project-specific skill that runs an app, follows real user actions, and captures evidence of behavior. It documents startup, interaction, feature checks, and cleanup, then verifies the generated instructions through an actual run.
Audit a project's verification skill and feature map against source code and actual use. Check every documented feature, correct outdated descriptions or verification scripts, and collect proven corrections in one pull request while reporting product bugs separately.
Checks whether performance measurements support a speedup, slowdown, or technology choice. It examines bottlenecks, production settings, errors, repeatability, and completed work, then guides reporting with run counts, ranges, and a clear verdict.
Validate software installation and real user workflows in remote or clean environments. Choose an appropriate backend, isolate untrusted code, gather diagnostic evidence, and clean up task resources across supported operating systems.
Check whether required API keys or other credentials are defined without exposing their values in chat or terminal output. When a credential is missing, provide a terminal command that hides typing and saves it to an environment file.
Run pytest tests for claude-tap after core changes to check the workflow from end to end. Review results for proxy processing, trace storage, cleanup, and viewers, with guidance on where to investigate failures.
Record Playwright browser test actions as .webm videos for pull request reviews, bug fix comparisons, and regression evidence. Combine readable pauses, assertions, and terminal logs to show interface behavior alongside automated checks.
Test JavaScript embedded in HTML with two complementary layers: Python unit tests for algorithms and Playwright browser tests for page behavior. Check calculations, button states, keyboard input, and navigation using prepared test data.
Test real Codex CLI sessions through claude-tap, including resumed conversations, multiple tool calls, and image input. Check captured API activity and viewer behavior, then collect screenshots as evidence for a pull request.
Check screenshots and viewer HTML used as evidence in pull requests. Inspect image dimensions, file size, and blank areas, then use Playwright to catch rendering failures, JavaScript errors, and exposed raw data.
Run end-to-end tests that connect local claude-tap source to the real Claude CLI. Use pytest or interactive tmux sessions to verify conversation traces, tool use, HTML viewer generation, streaming capture, and API key redaction.
Check workflows defined in CLAUDE.md for missing skills, incompatible JSON data formats, and inconsistent file names. Produce Markdown and JSON reports identifying broken handoffs, with an optional dry run using generated sample data.
Review agent skills with repeatable automated checks and a separate LLM assessment of content and code. Combine both scores into JSON and Markdown reports, with concrete improvement items when the final score falls below 90.
Verify gateway fixes and skill loading in a local Nexu and OpenClaw setup. Use internal integration tests, basic CLI or HTTP checks, and connected channels to cover different execution paths and session caching behavior.
Troubleshoot Feishu plugin problems using a FAQ and a diagnostic command. Check account settings, API connectivity, application permissions, and user authorization to identify persistent access issues and find steps to resolve them.