claude-code

Personal knowledge base manager. Processes inbox files (PDF, MD, DOCX, RTF, TXT, HTML, PNG, URL), extracts metadata, renames with citation conventions, generates summaries, maintains a searchable index, answers questions grounded in indexed sources, performs full-text keyword search, moves pipeline output into topic folders, and builds slide decks from indexed or external content via a slides skill.

Multi-step workflow to create presentation slides from content (academic papers, articles, notes). Reads the source (splitting PDFs, extracting text from Word/RTF, or reading markdown/text/LaTeX directly), summarizes the content, generates Beamer slides with figures and tables, then converts to PPTX using the style guide. Reuses any pre-existing text extract or summary in the output directory to avoid redundant work. Default deck opens Title, then Methodology, then Summary (a preview of conclusions) and includes Limitations and Conclusions slides near the end, each when appropriate. Optional `structure=` parameter (mba [default], teaching, faculty, professional, consulting, working) and `register=` parameter (business [default], technical) forward to the beamer skill to select the deck skeleton and language level; `audience=` is accepted as a deprecated alias for `structure=`.

Beamer slide generation with citation strategy decision, code-first matplotlib figures, outline checkpoint with content inventory, audience-aware rhetoric, Devil's Advocate slides, code blocks, and transition slides. Quality audit runs as a single agent at the sonnet tier. Beyond generate mode, it supports edit, audit, and convert-to-PPTX modes for working on a previously generated deck, including an optional PDF-vs-PPTX conversion audit. Edit mode auto-activates when the working directory contains a `*_build/slides.tex`. Adapted from Scott Cunningham's beautiful_deck approach.

Knowledge base sync and health check for the knowledge-base skill. Scans all topic folders for unindexed documents, removes index entries for missing files, refreshes topic descriptions, cleans up stale split directories, and reports health issues (missing summaries, naming inconsistencies, orphaned files, duplicate content).

Audit recent Claude Code session transcripts for recurring friction (repeated corrections, routing misfires, repeated reprompts, enforcement gaps, post-upgrade regressions) and recommend conservative changes to your personal skills, CLAUDE.md, protocols, or settings. Report-only: it proposes, it never edits. Prefers no change. Triggers on "skill audit", "audit my skills", "audit recent sessions".

Academic paper summarizer for papers, research articles, and scholarly publications. Structured summaries with citations, thesis, conclusions, methodology, implications, and critical analysis.

Summarizer for news articles, blog posts, videos, and podcasts. Structured summaries with citations, thesis, themes, and critical analysis. Not for academic papers; use summary-academic.

Run any question, idea, or decision through a council of 5 AI advisors who independently analyze it, peer-review each other anonymously, and synthesize a final verdict. Adapted from Andrej Karpathy's LLM Council methodology.

Generate standalone TikZ diagrams (pipelines, hierarchies, cycles, hub-and-spoke, thematic) compiled to PDF, with a mandatory independent audit agent that catches overlap and routing defects before delivery.

Download, split, and deeply read academic PDFs. Use when asked to read, review, or summarize an academic paper. Splits PDFs into 4-page chunks, reads them in small batches, and produces structured reading notes, avoiding context window crashes and shallow comprehension.

Deep, interactive variant of ai-council with three user-in-the-loop checkpoints (clarify, surface assumptions, iterate). Same five advisors and anonymous peer review. Use when the answer is expensive if wrong, or when you are still genuinely unsure what you are asking.

Analyze forwarded content for factual accuracy, then draft a reply in the user's voice. Use when the user provides forwarded content and asks for analysis, fact-checking, or a reply.