UnifiedFloppyTool

Use when implementing, extending, or wiring UFT's forensic audit trail and report-export subsystem (`include/uft/uft_audit_trail.h`, `include/uft/uft_forensic_report.h`). Trigger phrases: "audit trail", "audit event", "forensic report", "hash chain", "audit-trail wiring", "report exporter", "PDF/HTML/JSON report", "neuer event type", "risk scoring". Covers the 37 event types, 6 export formats (JSON/HTML/PDF/ Markdown/Text/XML), 14 report sections, hash-chain integrity, and chain-of-custody invariants. DO NOT use for: generic logging (`logging_setup.py` style — that's not forensic), debug printf reformatting, GUI-side notification wiring (→ `uft-qt-widget`).

Use when generating synthetic flux/bitstream/sector test data (SCP, HFE, KryoFlux RAW, ADF, D64). Trigger phrases: "erzeuge testdaten", "synthetic flux file", "test fixture für X", "PoC input data", "minimal valid SCP", "generate fake disk image", "weak bit fixture", "corrupted CRC fixture". Fully functional generators for 5 formats. DO NOT use for: real disk captures (copyright risk), performance fuzzing (use real dumps), UI testing (Qt test framework instead).

Use when implementing a new HAL backend for hardware controllers (Greaseweazle, SCP-Direct, KryoFlux, FC5025, XUM1541, Applesauce, UFI). Trigger phrases: "neuer HAL backend", "HAL backend für X", "controller X implementieren", "implement X HAL", "SCP-Direct HAL", "XUM1541 HAL real", "Applesauce HAL", "UFI hardware". Scaffolds C HAL + Qt provider + registry. DO NOT use for: bugfix in existing backend (→ structured-reviewer), changes to capability flags only (→ quick-fix), firmware code on the device side (separate UFI repo), protocol reverse engineering (different domain).

Use when adding, removing, or auditing entries in UFT's copy-protection title databases (`src/protection/c64/c64_protection_db.c`, `src/protection/uft_protection_classify.c`, `uft_protection_extended.c`). Trigger phrases: "protection db", "known title", "neuer kopierschutz eintrag", "c64 protection", "amiga protection", "duplicate title", "title lookup", "substring shadow bug". Captures the title-lookup contract (substring + bidirectional `strstr` → first-match-wins) and the de-dup invariant that was violated in v4.1.5 (9 stale duplicates shadowed canonical entries). DO NOT use for: protection-detection *algorithm* (→ `uft-protection-scheme`), generic database refactor (this skill is data-curation only), ML classifier training data (→ `uft-ml-training-data`).

Socratic mentoring system for HackTheBox players. This skill should be used when users request help with HackTheBox challenges using patterns like "/htb [box-name]" or "help me with [box-name] on HackTheBox". The skill fetches complete writeups, understands the full solution path, and guides players through Socratic questioning without giving direct answers.

Use when closing gaps from the Master Coding Standards v1.0 (MysticFoxDE, see memory/coding_standards.md §11). Trigger phrases: "H-1 / H-2 / H-9 / F-2 / D-2 fix", "capability manifest", "deprecate alias", "GUI HAL wiring test", "SPEC_STATUS marker", "preflight in converter X", "compliance gap", "rule X verletzt", "standards- konformität". This skill enforces the rule numbers, the 3-part error contract (F-4), and the rollout pattern (additive in v4.x, mandatory in v5.0). DO NOT use for: brand-new HAL backends (→ uft-hal-backend), format plugins (→ uft-format-plugin), bugs unrelated to the standard (→ uft-debug-session / quick-fix).

Use when implementing, reviewing, or debugging CRC/checksum code for UFT formats (MFM/FM/Amiga/Apple/DEC). Trigger phrases: "CRC16 für X", "MFM CRC falsch", "Amiga-Checksum", "Apple checksum", "CRC polynomial X", "verify CRC against reference", "neue CRC variante". Includes Python reference impl for cross-check without requiring reveng installation. DO NOT use for: network CRCs (different domain), file-format checksums that aren't CRC (MD5/SHA → use hash-engine), flux-level bit errors.

Use when a UFT build fails on any platform/compiler/Qt-version combination, when the CI matrix goes red, or when adding code that touches platform- specific APIs. Trigger phrases: "build fehlschlag windows", "msvc c11 fehler", "ci red", "qmake fehler", "vla msvc", "macos arch fehler", "mingw build kaputt", "windows linker error", "qt 6.10 c++20 fehler", "ci matrix red", "linux build only", "platform-specific build issue", "stale object files", "header conflict windows". Maps every error to the FIX-class that resolves it. DO NOT use for: runtime bugs that build fine (→ uft-debug-session), packaging/installer issues (separate concern), Qt UI errors (→ uft-qt-widget), compiler optimization tuning (→ uft-benchmark), STM32-firmware-portability of UFT code (→ uft-stm32-portability — different concern, prepares UFT for embedded; THIS skill is about desktop/CI matrix).

Use when reproducing a bug, isolating a minimal failing input, and writing the regression test that will prevent it returning. Trigger phrases: "bug repro", "minimal reproducer", "regression test für bug", "crash bei X isolieren", "ASan finding", "UBSan issue", "welcher commit hat X kaputt gemacht", "bisect fails". Standardizes the repro → isolate → fix → test cycle. DO NOT use for: feature requests (different workflow), performance regressions (→ uft-benchmark), build errors (→ uft-format-plugin verification section).

Use when adding a new module to UFT's DeepRead pipeline or OTDR-booster (8+12 modules exist). Trigger phrases: "neues DeepRead modul", "OTDR modul für X", "adaptive decode pass für Y", "recovery module", "weighted voting pass". Pipeline pattern: each module takes flux+annotations, produces refined flux+annotations. DO NOT use for: flat decoder changes (→ src/flux/), protection detection (→ uft-protection-scheme), format plugins (→ uft-format-plugin), top-level PLL (→ algorithms/).

Use when adding a new disk-image format plugin to UFT. Trigger phrases: "neuer format plugin", "füge plugin für X hinzu", "X format unterstützen", "add X plugin", "plugin für Y schreiben", "support Z disk image". Scaffolds the 6 touchpoints (plugin.c + enum + registry + .pro + test + CMakeLists). DO NOT use for: modifying existing plugins (→ structured- reviewer), conversion between formats (→ uft-format-converter), filesystem-layer work (→ uft-filesystem), flux decoder changes (→ flux/).

Use when crafting prompts for an LLM to work effectively on UFT — for Claude Code sessions, Claude Chat reviews, GPT-4 PR reviews, or any LLM you brief on a UFT task. Trigger phrases: "prompt für ML decoder review", "wie briefe ich Claude für hotpath optimization", "system prompt UFT", "agent prompt", "prompt engineering für UFT review", "reviewer prompt format". Templates that channel AI_COLLABORATION.md rules into ready-to-paste prompts. DO NOT use for: writing code itself (different skills), prompt engineering for unrelated projects (this is UFT-specific), conversational chat without a task (just talk), hyperparameter tuning prompts for ML training (separate ops domain).

Use when modifying or reviewing code under src/recovery/ or any function that fabricates, repairs, or reconstructs sector/bitstream/flux data. Trigger phrases: "recovery code", "sector recovery", "bitstream recovery", "multi-read voting", "CRC correction", "fehlende bytes rekonstruieren", "recovery pass", "confidence_map", "weak bits handhaben", "majority voting für sektoren". Enforces the data-integrity invariants that distinguish a forensic preservation tool from a "best-effort" recovery tool. DO NOT use for: DeepRead pipeline modules (→ uft-deepread-module), top-level format plugins (→ uft-format-plugin), CRC polynomial implementation (→ uft-crc-engine), filesystem-level salvage (→ uft-filesystem).

Use when preparing a UFT release (version bump, tag, changelog, CI verification, artefact publishing). Trigger phrases: "release v4.2", "version bump", "tag new release", "prepare release", "changelog für X", "bump version to", "release UFT". Executes the 8-step release checklist consistently. DO NOT use for: hotfix branches (separate workflow), internal dev builds, deploying pre-releases to package managers, producing DEB/RPM packages (separate skill later).

Use when reviewing or writing UFT code that must also compile for UFI firmware (STM32H723ZGT6, Cortex-M7). Trigger phrases: "läuft das auf STM32", "firmware-kompatibel", "dual-target", "firmware-safe", "FPU limit", "Intrinsics portabel", "SRAM budget", "runs on UFI firmware". Enforces Hardware-Dualität from AI_COLLABORATION.md §5. Includes executable portability check script. DO NOT use for: pure GUI/Qt code (desktop-only), hardware-provider code (desktop-only), code explicitly marked `#ifdef UFT_DESKTOP_ONLY`.

Use when implementing functions from a UFT_SKELETON_PLANNED header (uft_ml_decoder.h, uft_ml_training_gen.h, or any future skeleton). Trigger phrases: "implementiere uft_ml_decoder", "ML decoder fertig bauen", "training_gen Funktion implementieren", "skeleton ausfüllen", "promote skeleton to real implementation", "M1 MF-011 implement". Enforces the "no phantom features" rule: implement the function or remove the prototype, never both. DO NOT use for: implementing uft_ml_protection.c (already complete → uft-ml-protection-classifier), designing new ML APIs from scratch (write design doc first), porting external ML libs (separate concern), generating training data only (→ uft-ml-training-data).

Use when modifying or extending uft_ml_protection.c — the cosine- similarity copy-protection classifier. Trigger phrases: "neue protection im ML classifier", "V-MAX/RapidLok/CopyLock reference vector", "ml_protection feature engineering", "protection classifier threshold", "classifier features hinzufügen". This is the ONE complete ML module in UFT — pure C, no external libs. DO NOT use for: heuristic protection detection (→ uft-protection-scheme — different layer), CNN/neural decoders (→ uft-ml-implement-skeleton), training data generation (→ uft-ml-training-data), uft_ml_decoder.h (skeleton, not classifier).

Use when generating labeled training data for UFT's ML modules — flux captures with ground truth, augmented synthetic patterns, or protection-classifier reference samples. Trigger phrases: "training data für ML decoder", "labeled flux samples generieren", "ground truth pairs", "synthetic flux augmentation", "training set für classifier", "ml training generator implementieren", "ml_training_gen Funktion". Bridges flux fixtures and ML inputs. DO NOT use for: just test fixtures (→ uft-flux-fixtures), generating new disk-image formats (→ uft-flux-fixtures), implementing the ML decoder itself (→ uft-ml-implement-skeleton), labeling existing flux (different — that's manual annotation).

Use when measuring performance of decoder/PLL/CRC hotpath code, verifying speedup claims with before/after numbers, or detecting regressions. Trigger phrases: "benchmark für X", "messe wie schnell", "before/after speedup", "regression test performance", "is this really 2× faster", "micro-benchmark". Full working harness included. DO NOT use for: memory profiling (use valgrind/massif directly), Qt UI latency (use QTest), integration tests (use ctest).

Use when adding filesystem-layer support (AmigaDOS, FAT12/16, CBM DOS, Apple DOS 3.3, ProDOS, CP/M). Trigger phrases: "filesystem für X", "AmigaDOS hinzufügen", "FAT support", "file listing für X", "extract files from Y image", "directory traversal". Separate layer from format plugins — plugins read tracks, filesystems read files. DO NOT use for: adding new disk-image container (→ uft-format-plugin), conversions between containers (→ uft-format-converter), flux-level decoding.

Use when adding a NEW conversion path between two format plugins (e.g. D64→G64, SCP→ADF, IMG→HFE). Trigger phrases: "konvertierung X nach Y", "convert X to Y", "neuer conversion path", "X zu Y converter", "format bridge Y→X". 44 paths exist — pattern is stable. DO NOT use for: adding a new source/target format (→ uft-format-plugin), filesystem traversal (→ uft-filesystem), flux-to-flux timing preservation (→ uft-deepread-module).

Use when adding a copy-protection DETECTOR to UFT's read-only forensic pipeline. Trigger phrases: "neuer protection detector", "X copy protection erkennen", "V-MAX detector", "Copylock detection", "Rob Northen erkennen", "fuzzy bits detector", "long track detection". 35+ schemes registered; pattern is stable. DO NOT use for: removing/cleaning protections (UFT is read-only by design), format plugins (→ uft-format-plugin), filesystem semantics (→ uft-filesystem), protection-scheme reverse-engineering (write a spec doc first, then this skill).

Use when adding a NEW Qt6 widget, tab, or dialog to UFT's GUI. Trigger phrases: "neues Qt-widget", "neuer tab", "add Qt panel", "widget für X analyse", "GUI-dialog für Y", "analysis panel für Z", "OTDR-style panel". Enforces MASTER_PLAN Regel 2 (no GUI without working backend). DO NOT use for: modifying existing widget (→ structured-reviewer), pure styling changes (→ quick-fix), Qt Designer UI-file-only tweaks, plotting with QCustomPlot/QtCharts (those have their own patterns).