microsoft/PromptKit v0.3.0

v0.3.0 — C/C++ Analysis Ecosystem, Adversarial Bug Hunting, and 41 New Components

Repository: microsoft/PromptKit

Tag: v0.3.0

Published: 2026-03-26T16:53:23Z

Prerelease: no

Release notes:

PromptKit v0.3.0

The library nearly doubled. This release adds 41 new components — bringing PromptKit from 55 to 96 components — with a major focus on C/C++ code analysis, cross-compiler portability, and rigorous adversarial review workflows.

Highlights

🔍 C/C++ Analysis Ecosystem — A complete review toolkit built from research-validated patterns and public standards:

• 7 research-cited C++ best practice patterns (CPP-1 through CPP-7), grounded in published work from MSRC, Lu et al., Bloch, and others
• 10 MSDN-public Win32 API convention patterns (WG-001 through WG-010)
• 10 performance-critical C API patterns (PC-001 through PC-010) for system libraries, game engines, and drivers
• 8 WinRT design patterns (WRT-001 through WRT-008) for Windows Runtime APIs
• 21 MSVC↔Clang/GCC portability patterns across 7 categories
• A dedicated review-cpp-code template that composes all of the above

🐛 Adversarial Bug Hunting — A forensic review mode where every finding must survive challenge:

• adversarial-falsification protocol: the LLM must disprove each finding before reporting it
• exhaustive-path-tracing protocol: every file gets a coverage ledger proving full analysis
• kernel-defect-categories taxonomy: 14 defect classes (K1–K14) for OS kernel and driver code
• exhaustive-bug-hunt template tying it all together

🔧 Automated Remediation — Systematic, build-verified batch fixing:

• fix-compiler-warnings template: SARIF-driven warning remediation with build-verify loops
• compiler-diagnostics-cpp protocol: specific resolution strategies for C4456–C4459 shadowing, implicit conversions, deprecated features, and pragma suppression removal
• minimal-edit-discipline guardrail: prevents collateral damage from automated fixes (encoding safety, type preservation, formatting preservation)
• find-and-fix-bugs template: autonomous scan→fix→build→iterate workflow

📋 Finding Classification & Structured Findings — A reusable framework for any diagnostic-to-report pipeline:

• finding-classification reasoning protocol: three-way classification (Exact Match / Variant / New Pattern) with confidence analysis and catalog update proposals
• structured-findings format: consolidated findings with severity/confidence classification, root cause analysis, and source fidelity markers
• cpp-review-patterns taxonomy: 19 labels across 6 categories for classifying C++ review findings

---

New Components

Protocols (+17)

| Protocol | Category | Description | |----------|----------|-------------| | cpp-best-practices | analysis | 7 research-validated C++ review patterns | | win32-api-conventions | analysis | 10 Win32 API naming/typing/design patterns | | performance-critical-c-api | analysis | 10 patterns for perf-critical flat C APIs | | winrt-design-patterns | analysis | 8 WinRT activation/deferral/enumeration patterns | | compiler-diagnostics-cpp | analysis | Systematic C++ warning resolution strategies | | msvc-clang-portability | analysis | 21 cross-compiler portability patterns | | kernel-correctness | analysis | OS kernel lock/refcount/cleanup analysis | | adversarial-falsification | guardrail | Requires disproving findings before reporting | | minimal-edit-discipline | guardrail | Constrains code mods to be minimal and verifiable | | finding-classification | reasoning | Three-way finding classification with confidence | | exhaustive-path-tracing | reasoning | Per-file coverage proofs with path tracing | | rfc-extraction | reasoning | Extract requirements from RFCs/internet-drafts | | invariant-extraction | reasoning | Extract constraints/state machines from specs | | test-compliance-audit | reasoning | Audit test code against validation plans | | integration-audit | reasoning | Audit cross-component integration points | | workflow-arbitration | reasoning | Multi-agent workflow progress evaluation | | requirements-reconciliation | reasoning | Reconcile multi-source requirements |

Templates (+16)

| Template | Category | Description | |----------|----------|-------------| | review-cpp-code | code-analysis | C/C++ review composing best practices + memory safety | | exhaustive-bug-hunt | code-analysis | Adversarial line-by-line review with coverage proof | | find-and-fix-bugs | investigation | Autonomous scan→fix→build→report workflow | | fix-compiler-warnings | investigation | Batch warning remediation from SARIF/compiler output | | discover-tests-for-changes | testing | Find relevant tests for local code changes | | scaffold-test-project | testing | Test project scaffolding (gtest/pytest/jest/TAEF/etc.) | | generate-commit-message | devops | Structured commit messages from staged changes | | extract-rfc-requirements | standards | Extract structured requirements from RFCs | | extract-invariants | standards | Extract constraints from specs or code | | reconcile-requirements | standards | Reconcile requirements from multiple sources | | audit-test-compliance | document-authoring | Audit tests against validation plan | | audit-integration-compliance | document-authoring | Audit cross-component integration | | author-implementation-prompt | code-generation | Prompt for spec-compliant implementation | | author-test-prompt | code-generation | Prompt for spec-compliant test generation | | author-workflow-prompts | code-generation | Multi-agent workflow prompt assets | | author-north-star | document-authoring | Strategic north-star vision documents |

Formats (+3)

| Format | Description | |--------|-------------| | structured-findings | Consolidated diagnostic findings with severity/confidence | | exhaustive-review-report | Per-file coverage ledgers with falsification proofs | | north-star-document | Strategic vision documents |

Taxonomies (+2)

| Taxonomy | Description | |----------|-------------| | cpp-review-patterns | 19 labels across 6 C++ review categories | | kernel-defect-categories | 14 OS kernel/driver defect classes (K1–K14) |

Personas (+3)

| Persona | Description | |---------|-------------| | implementation-engineer | Builds correct code from specifications | | test-engineer | Writes thorough, spec-driven tests | | workflow-arbiter | Evaluates...