kc_ai_skills

Use when the user wants to build a consistent-identity LoRA for an original character — defining the character, generating a face/body-consistent multi-angle dataset (via the gpt-image-gen skill for codex image generation), captioning it, doing base-specific homework, training on a chosen base (Pony / Z-Image / others) on a local GPU, and producing a usable LoRA. This skill ORCHESTRATES the end-to-end pipeline and gates every expensive/irreversible step; it delegates actual image generation to gpt-image-gen and never improvises training settings from memory.

Use when the user attended one or more sessions of a conference (with audio recordings + slide photos) and needs help building (1) faithful per-session reconstructions in markdown (slide visuals + speaker transcript with Whisper hallucination annotations), and (2) a downstream report deliverable whose scope and format are decided interactively with the user. Pipeline phase (raw → mlx_whisper Chinese SRT → per-slide multimodal reconstruction → official agenda cross-check via Playwright if available) is deterministic. Report phase is interactive — always quiz user on scope (single-session / single-day / multi-day synthesis), format (existing template / free-form fallback), recipient (formal / informal), and any business workstream mapping before drafting.

Use when the user is solving an authorized CTF / lab reverse-engineering challenge focused on Windows application authentication or license-check bypass. Guides triage → static analysis → dynamic experiment planning → bypass verification, with strong evidence discipline and VM safety boundaries. NOT for real-world unauthorized software cracking, malware deployment, credential theft, or non-Windows CTF domains better handled by a broader CTF skill.

Use when the user wants to AUTHOR an unattended dispatch for either: (1) a goal-driven evaluator-optimizer loop of the GENERATE-AND-SELECT kind (generate candidates → grade against a rubric → iterate by reason-code → keep the best; the human picks the final selection), or (2) a lean build-to-spec run whose build spec is ALREADY FROZEN (approved ADR / locked design / accepted machine-checkable AC) and the only missing piece is the unattended-execution wrapper. A fresh-session agent runs the dispatch hands-off while the human only watches traffic-light push notifications. Interview-style forcing questions lock the spec, then it emits a self-contained dispatch markdown + a channel-agnostic notification protocol. This is the upstream SPEC AUTHOR, NOT a runtime: the dispatch is run by Claude Code's /goal, a headless `-p` session, or any unattended agent — /goal is the engine, this writes what you feed it. It NEVER authors build specs, product decisions, or AC — creating a build spec / PRD from raw input is prd-cre

Use when the user asks to generate an image via GPT/Codex (e.g. 「叫 gpt 生圖」「幫我用 gpt 生圖」「gpt 畫一個 X」). The skill drafts a Chinese + English prompt pair, iterates with the user until they explicitly approve, then dispatches Codex CLI ($imagegen skill, codex built-in image_gen) in the background, monitors progress, converts the result to a jpg in the current working directory, and writes a sidecar prompt log. Does text-to-image AND img2img — drop a reference image (on-disk file) and it runs Codex `-i` to lock a face/character across scenes.

Use when the user is considering applying to, interviewing with, or accepting an offer from a company and wants due diligence on the company and optionally a specific role. Requires a company name, optionally a job title, then researches current public data, employee reviews, salary signals, product/tech quality, and red flags before giving a grounded recommendation. NOT for generic career coaching or resume editing.

Use when the user wants to test, compare, or choose local Ollama models for their machine. Checks Ollama/GPU state, recommends model sizes from available VRAM, preserves existing benchmark records, pulls only approved models, runs repeatable benchmarks, restores stopped services, and writes a markdown comparison report. NOT for hosted API model evaluation or subjective chat-quality judging without local benchmark commands.

Use when the user wants to lint a repo following the Karpathy LLM Wiki pattern (raw/ + wiki/ + SCHEMA.md / index.md / log.md). The skill detects path, scans wiki pages + schema layer, reports frontmatter gaps, broken in-body links, source traceability breaks, stale claims, orphan pages, index synopsis contradictions, and SCHEMA-vs-reality drift. Phase 1 is read-only report; Phase 2 (fix + log.md LINT entry) only runs after the user explicitly approves.

Use when the user wants to convert one Markdown file into a publication-ready A4 PDF, especially when the source may contain Mermaid diagrams, ASCII diagrams, CJK text, tables, or pandoc/weasyprint edge cases. Works by copying the source to a _pdf.md working file, converting diagrams, escaping PDF-breaking syntax, rendering with pandoc + weasyprint, then self-checking pages. NOT for batch conversion, slide decks, or editing the original Markdown in place.

Use when the user wants to turn a Markdown report into a presentation-quality .pptx via interactive design decisions and a reusable hand-coded build script. Drives pre-analysis, global style choices, optional per-slide layout dialogue, python-pptx composition, and optional LibreOffice render self-check. NOT a generic auto-converter, NOT for PDF output, and NOT a fixed brand-template pipeline — brand integration is handled ad hoc through helper primitives.

Use when the user wants to lint a Claude Code memory directory (~/.claude/memory or custom path) for index inconsistency, stale project state, duplicate / conflicting feedback rules, naming convention violations, frontmatter gaps, and oversized files. The skill detects path, scans root-level *.md, reports findings by severity. Read-only — never auto-fixes, never deletes, never merges.

Use when the user wants to prepare a local project for GitHub or public release. Runs a release-readiness sweep over README structure, bilingual docs, commit hygiene, sensitive-data exposure, broken links, markdown rendering, project layout, tests, CI, Docker, and final cleanup. Fixes issues only inside the target repo and treats secrets/history rewriting as explicit high-risk gates. NOT for publishing without user approval or for private operational runbooks that should not be open-sourced.

Use when the user wants to evaluate a GitHub repository before installing, running, forking, or depending on it. Takes a GitHub repo URL, cleans tracking params, shallow-clones to /tmp, inspects dependency/supply-chain risk, static vulnerability patterns, issue-reported security problems, maintainer health, and produces a risk summary. NOT for reviewing the user's own PR diff or for running untrusted code.

Use when the user wants to rewrite Markdown prose into a conversational, humorous, self-deprecating engineering tone while preserving technical accuracy and document structure. Rewrites prose sections only, keeps code blocks, diagrams, tables, English summary blocks, and factual claims intact. NOT for changing requirements, adding new technical content, or editing non-Markdown artifacts.

Use when the user wants privacy-respecting web, image, news, or video search through a configured local SearXNG instance instead of external search APIs. Calls the bundled script against SEARXNG_URL, supports result limits/categories/language/time range, and can return human-readable or JSON output. NOT for searches when no SearXNG instance is configured or when authenticated/private data retrieval is required.

Use when the user wants to register, inspect, manually run, or remove scheduled Claude skills with Telegram push notifications. Presents a menu, discovers schedulable skills with headless-prompt frontmatter, converts natural-language schedules into cron entries with conflict confirmation, writes managed config, and verifies notification delivery. NOT for one-off task execution without scheduling or for running skills that lack a headless prompt.

Use when the user wants a spec-driven development workflow for implementing a feature in the current codebase, from fuzzy idea or existing active spec through requirements, technical plan, tasks, implementation, verification, and closure report. Auto-detects project/spec state, writes persistent files under specs/, asks one grounded question at a time when requirements are ambiguous, and stops at stage gates. NOT for stakeholder PRDs (prd-create), ADO ticket breakdown (prd-breakdown), single architecture-decision records (adr), or already-frozen tasks that should simply be implemented.

Use when the user is unsure which kc_ai_skills workflow skill to use, or describes work involving PRD, SD/software design, FR, AC/acceptance criteria, ADR, tickets, implementation, debugging, release checks, or unattended agent execution. Triage the request with at most one clarifying question when needed, explain the route in plain language, then hand off to the right specialist skill. This is an entry router only: it does not write PRDs, specs, ADRs, tickets, or dispatches itself.

Use when user wants to draft a PRD (Product Requirements Document) from raw input (meeting transcripts, hand-waved descriptions, scattered decisions). Workflow: load org's PRD Guideline + writing discipline → lock execution mode (human-run vs unattended-agent-run) → ingest raw input → quiz user numbered-list iterate to fill §1-§15 → draft v0.1 → handle stakeholder merge (review feedback, surface conflicts) → lock v1.0 + sanitize per ADO publication contract → publish to ADO Wiki. For an agent-run PRD, §13 carries the unattended-execution discipline (machine-checkable AC + traffic-light + 3-exits + stop-and-ask), aligned with goal-engineer's loop-run-protocol. NOT for packaging an ALREADY-FROZEN build spec (approved ADR / locked design / machine-checkable AC) into an unattended dispatch — that is goal-engineer's lean build dispatch. Pure prompt-driven — Claude is the runtime, no Python helper. Trigger phrases: 寫 PRD / PRD 撰寫 / prd-create / 初版 PRD / 起 PRD.

Use when a durable technical decision was just made in conversation and should be recorded, or when user asks to write/record an ADR (開 ADR / 記個決策 / 這要不要 ADR). Runs a three-gate check FIRST and actively talks the user out of writing one when the decision doesn't qualify — then writes a lightweight (title + 1-3 sentences) ADR following the repo's own ADR conventions if any exist. Repo conventions always override this skill's defaults. NOT for requirement specs (spec / prd-create) or for rewriting history (superseded ADRs get a new ADR, never an edit).

Use when the user reports misbehaving software to investigate — a bug, crash, wrong output, flaky behavior, or '為什麼壞掉/不會動/查一下' — and the root cause is not yet established. Enforces building a red-capable reproduction command BEFORE any hypothesis is allowed, then 3-5 ranked falsifiable hypotheses before testing any single one. NOT for conceptual questions, code reading requests, or feature work. When a failing command already exists, Step 1 is pre-satisfied — still apply, jumping straight to the hypothesis discipline.

Use when the user says 先討論 / wants to align on a fuzzy idea before building, or before any non-trivial implementation whose requirements are still ambiguous — interviews the user until shared understanding is confirmed, one question at a time, each with a suggested answer; facts answerable from the filesystem/code are looked up, never asked. Optionally maintains the repo's CONTEXT.md glossary when vocabulary settles during discussion. Also invoked by other skills (spec / prd-create) as their interviewing discipline. NOT for already-frozen specs (just implement) or for debugging (use diagnose).

Use when user wants to break a PRD into Azure DevOps work items via vertical-slice plan. Workflow: read PRD → quiz user to lock durable decisions → draft slices with HITL/AFK + blocked_by → call az CLI directly via Bash to create items + Predecessor relations. Idempotent re-runs via fingerprint markers embedded in description. Pure prompt-driven — Claude is the runtime, no Python helper, no install ceremony. Trigger phrases: prd-breakdown / 拆 PRD / 切 vertical slice / 推 slice 到 ADO.