peter537/peter537-agent-plugin

Perform read-only, redacted reviews of repositories, Git changes or ranges, paths, and subsystems for personal data, private records, credentials, developer-machine metadata, re-identifiable samples, anonymization failures, and disposable one-time migrations or conversions. Use only when the user explicitly requests a personal-data, privacy, anonymization, repository-leak, sensitive-data exposure, or disposable-migration review. Do not invoke merely because code is being committed, released, published, or open-sourced, and do not use for ordinary code audits, dependency reviews, performing dataset anonymization, legal-compliance opinions, or implementation work.

Audit every software dependency and supply-chain input in a repository, including direct and transitive packages, nested projects and workspaces, build plugins, CI actions, container images, vendored code, submodules, and runtime-loaded extensions. Use for dependency or package reviews; before adding, installing, upgrading, replacing, or removing dependencies; when manifests or lockfiles change; or when checking exact resolved versions, vulnerabilities, provenance, integrity, licenses, malicious packages, dependency confusion, maintainer risk, or supply-chain attacks. Do not use for broad source-code audits unrelated to dependencies.

Review existing or proposed tests and agent evaluation cases for useful failure detection, redundancy, implementation coupling, and maintenance cost; consolidate or remove tests when authorized. Use for dedicated test-value audits and test-suite cleanup. Excludes ordinary test writing, concrete bug or flake diagnosis, production-code retirement, building missing verification, and independent acceptance verification.

Build missing verification paths and strengthen passing but ineffective tests for established behavior. Use for dedicated requests to author a harness, fixture, meaningful assertion, readiness check, persistence observation, or safe reset procedure, including already-specified missing harness assertions. Excludes routine test-table extensions, test-value audits and removal, concrete bug or flake repair, independent acceptance verification, documentation-only context work, and MAUI-specific browser hosting.

Orchestrate reproducible, multi-source software and product research through ChatGPT Deep Research and verify the resulting evidence. Use for complex or ambiguous questions requiring synthesis across external sources, official-documentation comparison, vendor or standards evaluation, UI reference analysis, conflicting claims, or traceable capture sets with URLs, dates, and citations. Do not use for a single fact, one-document lookup, ordinary web search, or repository-only investigation.

Audit and improve the semantic health of comments and docstrings embedded in source, tests, configuration, build files, and scripts. Use for clear requests to review or clean up misleading, redundant, stale, misplaced, missing, commented-out, TODO/FIXME, API-documentation, doctest, or tool-directive comments. Do not use for documentation sets, wording-only editing, broad code audits, generated or vendored content, fixture text where comment syntax is data, or executable-code simplification.

Perform evidence-backed, read-only audits of software repositories or scoped changes for correctness, security, maintainability, readability, state and data ownership, testability, performance, architecture, file organization, dependency risk, proportionate sensitive-data risk, accidental complexity, overengineering, and unnecessary implementation or change surface. Use for comprehensive codebase audits, security reviews, or deep review of a diff, branch, pull request, file, or subsystem. Do not use for dedicated bug diagnosis, dependency-only reviews, personal-data or repository-exposure audits, pre-addition supply-chain gates, dedicated source-comment health or codebase-pruning passes, dedicated C# or Python simplification, routine implementation, simple style feedback, or a narrowly specified fix that does not request an audit.

Research and clarify complex software changes before implementation. Use for architecture proposals, migrations and refactors, parity or retirement inventories, sequencing and dependency analysis, scope and tradeoff decisions, ADR candidates, acceptance gates, and other software work that needs repository evidence plus targeted user questions. Do not use for straightforward changes with an already decision-complete specification.

Diagnose and, when authorized, repair concrete, non-obvious local-repository or CI defects. Use for specific bug investigation, reproduction, root-cause analysis, flaky tests or regressions, and build, integration, environment, data-dependent, or measured performance failures. Broken, throwing, failing, or slow behavior alone is insufficient. Excludes quick error explanations, possible-cause lists without investigation, issue summaries, proactive code or performance audits, ordinary feature work, known trivial fixes, and active production or security-incident response.

Audit software-repository documentation against implementation evidence and report findings without editing; update, refresh, or rebuild it when requested. Use for comprehensive documentation reviews and maintenance of README files, project guides, navigation, documentation tooling, and evidence-backed Mermaid diagrams. Do not use for ordinary prose drafting, proofreading, copyediting, tone or voice work, repository-only analysis, application-code changes, or dependency- and package-only installation or change requests.

Prepare, assess, implement, and verify safe browser-development paths for .NET MAUI Blazor Hybrid UI by reusing or creating a companion Blazor Web host, or by creating an isolated temporary browser harness around host-neutral Razor Class Library UI. Use when a MAUI Blazor Hybrid project needs browser inspection, browser-driven development, responsive or DOM verification, shared MAUI-and-Web UI, host-specific service adapters, or diagnosis of a browser host that prerenders but does not become interactive. Do not use for ordinary Blazor Web applications, native-only MAUI/XAML work, visual-polish-only requests, or browser testing unrelated to MAUI Blazor Hybrid.

Prove and safely retire dead or obsolete repository surface, including unused code, exports, routes, and assets or obsolete flags, experiments, compatibility, and migration paths. Use for clear requests to find or remove dead code, retire obsolete behavior, or prune a codebase. Excludes broad code audits, local C# or Python simplification, dependency removal, bug diagnosis, architecture redesign, feature work, and cleanup justified only by style.

Simplify explicitly scoped C# or Python code without changing its established behavior. Use for clear requests to remove unnecessary indirection or generated boilerplate, harden weak type handling, de-slop an implementation, perform a focused simplification review, or apply already-configured simplification enforcement. Do not use for broad repository audits, repository-wide dead or obsolete surface pruning, bug diagnosis, architecture migrations, feature development, formatting-only cleanup, dependency installation, UI or prose work, or other languages.

Assess supplied failures, near misses, or proposed rules and recommend one evidence-backed disposition without implementing it. Use only when explicitly invoked as $retrospective. Do not use for automatic session learning, transcript archives, active incident response, concrete bug repair, broad code audits, or ordinary implementation planning.

Review AGENTS.md, skill instructions, discovery descriptions, and related agent guidance for authority conflicts, stale commands, unclear applicability, unnecessary loading, duplication, and approval friction. Apply instruction fixes when explicitly requested. Use for dedicated reviews of how guidance steers agents; not ordinary specified instruction edits, wording-only polishing, documentation-set reconciliation, retrospective disposition selection, broad code audits, or implementation planning.

Capture, update, or resume a task checkpoint when explicitly requested, including natural-language requests to save a handoff or continue from a checkpoint. Preserve scope, decisions, evidence limits, repository state, and remaining work across interruptions. Do not save progress automatically or use for ordinary summaries, new architecture planning, research, retrospective learning, documentation audits, or durable verification-context authoring.

Create, audit, refine, polish, or redesign task-oriented product interfaces, including generic-looking UI, using repository evidence and rendered verification. Also assess, create, or maintain persistent project design context from repository evidence and explicit human decisions on direct request, without requiring UI changes or rendered verification. Covers hierarchy, interactions, responsive behavior, accessibility, and design-system consistency, including Streamlit and Blazor Web Apps. Excludes wording-only copy, ordinary non-UI framework or backend work, image-only tasks, marketing, editorial, and commerce surfaces, and documentation maintenance other than the requested design context.

Create, assess, or maintain a repository's authoritative verification guidance from existing local evidence. Use only when the user explicitly invokes $verification-context to document verification commands, entrypoints, environments, fixtures, evidence limits, persistence boundaries, reset procedures, and cleanup. Do not use to execute acceptance verification, perform comprehensive documentation audits, plan unresolved behavior, add dependencies or harnesses, diagnose bugs, run one named check, or prepare a release.

Verify whether an already-completed local-repository or CI change satisfies its acceptance criteria through reproducible, layered evidence and complete cleanup. Use when the user clearly asks to prove, validate, acceptance-test, or end-to-end verify completed behavior. Do not use for implementing or repairing a change, running one named check, broad auditing, root-cause diagnosis, planning, UI-quality review, dependency work, active incidents, deployment, release, or publication.

Draft, edit, proofread, audit, or apply human feedback to repository prose for clarity, reader fit, and voice while preserving meaning, source language, and format. Includes documentation and wording in UI, CLI, comments, and localization strings. Exclude authoritative verification-context authoring beyond wording-only edits; without explicit $verification-context invocation, that is ordinary repository work. Also exclude semantic source-comment audits, code-only work, translation-only requests, summarization, extraction, and factual verification. Poetry and fiction require explicit invocation.