Linked Spec Consistency
Use when one change crosses multiple Markdown requirements, decisions, verification notes, and release-state documents, and the maintainer needs an explicit relation ledger plus executable evidence that links, identifiers, coverage, retired language, and claims about current behavior still agree.
Long Running Agent Sessions
Use when a multi-step agent task must continue without repeated prompting or resume safely after interruption while controlling duplicate effects, concurrent runners, cancellation, approvals, and retry limits.
Safe Download Verification
Use when assessing a download URL or file before execution. Separate source provenance, transport, integrity, publisher identity, signature evidence, and malware observations; stop on mismatches and never disable platform protections.
Evidence Grounded Research
Use when researching a disputed, fast-changing, or poorly standardized topic that requires primary sources, current official documentation, evidence grading, terminology boundaries, and explicit fact/judgment/unknown labels.
Windows Cli Portability
Use when a command or setup script behaves differently across Bash, Git Bash, Windows PowerShell, PowerShell, or a native Windows CLI. Preserve argument boundaries, stdin, paths, encoding, exit codes, and side-effect controls through observable contracts.
Github Repo Investigation
Use when a user names a public GitHub project or repository and wants a read-only, evidence-labeled assessment of its identity, positioning, architecture, audience, activity, license, non-goals, and due-diligence risks without cloning it by default.
Fail Closed Boundary Testing
Use when testing parsers, adapters, APIs, queues, plugin boundaries, streaming readers, or multi-source merges where malformed, incomplete, contradictory, unavailable, or oversized input must not become success or trigger side effects.
Ai Competition Due Diligence
Use when deciding whether an AI competition, hackathon, or innovation challenge is real, open, relevant, and worth entering. Build a time-bounded evidence ledger for identity, eligibility, dates, rules, judging, intellectual property, data rights, costs, and submission path.
Openai Compatible Provider Diagnostics
Use when investigating why an OpenAI-style endpoint behaves differently across non-stream, stream, tool, multimodal, or agent requests. Provides a layered manual workflow and a bounded classifier for caller-supplied HTTP/SSE metadata; it does not execute provider or capability probes.