Claude Skills by grahama1970

Deterministic public-domain text corpus chunker with seeded selectors for reproducible sampling across engineering/government corpora.

Collaborative argumentative walkthrough for complex implementations. REQUIRES /interview (user context) and /ask consult (persona review) BEFORE writing. Combines claim verification, Mermaid diagrams, structured tables, and adversarial human review into a prosecution brief.

Detect, classify, and mark Controlled Unclassified Information (CUI) per 32 CFR Part 2002. Scans documents and ArangoDB collections for CUI indicators, applies proper markings, and tracks CUI flow through the system.

Build a dedicated, recall-ready client knowledge base in /memory from a client brief, website/API docs, OpenAPI specs, and GitHub repos, then wire it into /live-evidence as the glance-card knowledge graph for interviews, meetings, and sales calls. Use when the user says "curate client", "build a client knowledge base", "prep me for an interview/meeting with <company>", or asks to turn docs plus repos into /memory recall knowledge.

Control and inspect Cursor Cloud Agents through the official Cursor Agents API. Use this when users ask for Cursor Agents, webcursor, cursor.com/agents control, Cursor Cloud Agent runs, Cursor agent usage, or Cursor agent API checks.

Unified Embry OS development dashboard. Aggregates daemon health, LLM metrics, shadow agreement, cascade state, skill health, active tasks, Chutes quota, and project-state into a single Rich TUI, JSON, or plain-text view.

Report on data completeness for the SPARTA QRA pipeline (Control -> URL -> Knowledge -> QRA). Queries DuckDB to show coverage percentages at each stage.

Automated URL fetch failure handling with strategy exhaustion, memory learning, and human-in-the-loop recovery. Use when fetches fail and you need intelligent retry, pattern learning, and human collaboration.

Automated PDF failure analysis and fixture generation. Takes failed PDF URLs, identifies breaking patterns, and generates minimal fixtures via fixture-tricky for regression testing. Supports batch mode and combined stress test generation.

Force the project-agent to use a real debugger instead of guessing: set breakpoints where the problem might be, stop execution at those breakpoints, inspect live variable state, and analyze the observed runtime state before patching. Use when a project agent is stuck, sees confusing or repeated failures, suspects state mutation, routing, async, serialization, cache, closure, test fixture, UI/backend mismatch, or any bug where logs/static reading would lead to speculation. Use before further p...

Discover new books using OpenLibrary API with Federated Taxonomy integration. Recommends books based on subjects, authors, and HMT bridge attributes. Can be invoked via /dogpile books.

Research and enrich professional contacts and marketing prospects. Uses /dogpile to find current company, role, recent news, and company intelligence. Ingests CSV contact lists and outputs enriched profiles to /memory.

Search and retrieve cinematic 3D LUTs (.cube) from web sources or cache.

Discover new movies using TMDB API with Federated Taxonomy integration. Recommends movies based on genres, directors, and HMT bridge attributes. Can be invoked via /dogpile movies.

Discover new music using MusicBrainz and ListenBrainz APIs. Recommends music based on Horus taste profile and HMT bridge attributes. Can also be invoked via /dogpile music.

Discover actors and reference talent via TMDB API with HMT taxonomy. Find actors by name, filmography, or bridge attributes. Used by create-cast for character reference discovery.

Compatibility shim for legacy distill callers. Forwards to /doc2qra while preserving the historical CLI interface.

Convert any document (PDF, URL, text) into Question-Reasoning-Answer pairs with a document summary. Stores to memory for later recall.

Deep research aggregator that searches Brave (Web), GitHub (Code/Issues), ArXiv (Papers), YouTube (Videos), optional Context7 library docs, and optional feed/archive/book sources. Provides a consolidated Markdown report with an ambiguity check, grounded synthesis, and Agentic Handoff.

Model integrity monitor. Detects LLM quality degradation via active probes (NIAH, SKILL.md extraction) and passive signals (hook denials, transcript corrections, duplicate reads). CUSUM drift detection on quality time series. D3 dashboard for visualization.

Verifies relationships between a source text (e.g. Episode, Task) and existing Knowledge Graph lessons. Runs KNN/Hybrid search to find candidates, then uses LLM (scillm) to verify "verifies", "contradicts", or "related" stances with strict rationales.

Dual-backend embedding service for semantic search. Text-only (MiniLM, 384d) and multimodal (Qwen3-VL-Embedding-2B via vLLM, 2048d). Persistent Docker containers with automatic routing by modality.

Verifies relationships between a source text (e.g. Episode, Task) and existing Knowledge Graph lessons. Runs KNN/Hybrid search to find candidates, then uses LLM (scillm) to verify "verifies", "contradicts", or "related" stances with strict rationales.

Read and explain Embry OS configuration from embry.yaml

Skill wrapper for the Embry Dashboard (Tauri app). Launch, navigate tabs, and screenshot the dashboard for design review.

Agent-readable Embry voice control-plane contract for speaking, listening, cancelling, replaying, and inspecting live voice/chat turns. Use when a project agent needs to control Embry Chatterbox voice, RealtimeSTT listener state, shared Chat UX synchronization, memory/Tau routing, orb state, or session replay through explicit endpoints instead of ad hoc browser scripts.

This skill encodes Emil Kowalski's philosophy on UI polish, component design, animation decisions, and the invisible details that make software feel great.

Episodic Memory Archiver. Stores full conversation transcripts with embeddings and analysis into ArangoDB. Tracks UNRESOLVED sessions for reflection with structured failure episodes (trigger/diagnosis/action/outcome), K~4 similar failure retrieval, user behavioral profiling, and federated taxonomy classification.

Behavioral evaluation of skills using fixture-based test cases. Runs eval.json fixtures per skill to verify correct output, exit codes, latency budgets, and compose chain pipelines. Produces transparent Markdown and JSON reports with diffs for failures. Complements sanity.sh (smoke tests) and skills-ci (static analysis) with behavioral correctness.

Self-improving evidence case convergence loop. Runs question banks against /create-evidence-case, diagnoses mismatches, and tracks convergence across cycles.

Live evidence case pipeline viewer with NVIS sentence markup and agent chat well. Shows gates firing, entity grounding, QRA evidence, and verdicts in real-time. Chat well connects to /subagent-service for human-agent course correction.

Generate and validate end-of-work feature explainers for technical interviews and developer transparency: strict JSONL records, teleprompter notes, source/debugger bindings, editable Excalidraw architecture boards, optional SVG previews, and question routing.

NIST OSCAL JSON export of compliance evidence from ArangoDB graph

Extract framework control references from documents and create graph edges to sparta_controls. Supports NIST, CWE, ATT&CK, SPARTA, D3FEND, ISO. 3-tier extraction: regex → RapidFuzz → classifier (future).

Extract control IDs, domain phrases, taxonomy tags, and relationship data from question text. Composes /memory (ArangoDB recall to load sparta_controls vocabulary). Returns structured EntityExtractionResult that defines the shape of evidence cases, conversations, and QRA reviews before any LLM runs. Zero LLM cost — Flashtext (Aho-Corasick) + RapidFuzz (Levenshtein). NO REGEX.

Extract structured JSON from HTML using Schematron3B, deterministic table extraction, and Vision-based media OCR.

Self-improving PDF extraction engine. Rust-native via pdf_oxide. Replaces PyMuPDF (AGPL) with MIT-licensed extraction: text, blocks, sections, tables, figures, profiling, engineering detection. Shadow-LEGO cascade for automatic quality improvement via predict-extract-verify loop.

Composable PDF table extraction skill. Hybrid Rust+compiled-Python architecture replacing Camelot. Rust hotspots (image processing, geometry) via PyO3, compiled Python parsers (lattice, stream, network, hybrid), pdf_oxide for text extraction, polars for output. Shadow-LEGO self-correcting with strategy routing via /assistant classify. Agentic: builds and self-corrects its own extraction pipeline.

Persona-driven datalake quality assessment with Margaret Chen (Pratt & Whitney) and Jennifer Cheung (NIWC Pacific). Dynamic annealing thresholds, batch review, and /ask consult integration.

Extract one local document through the canonical Extractor application command. Use when the user says "extract this", "extract pdf", "convert to markdown", "parse this file", "process document", or provides a document.

Stage cleaned chapter text and extract quote-grounded minimal fact/QRA candidates from source text or primary/context chunks, with streaming model calls, retries, validation, and durable artifacts.

Fetch web pages, PDFs, and documents with automatic fallbacks and content extraction. Use when user says "fetch this URL", "download this page", "crawl this website", "extract content from", "get the PDF", or provides URLs needing retrieval.

Iterative visualization development lab for /create-figure. Compose novel D3 visualizations from primitives (marks, scales, encodings), test rendering, evaluate against user intent, and promote successful compositions to /create-figure as first-class commands. Self-improving: tracks render success, visual fidelity, and user satisfaction.

Search a codebase or UI for places that don't animate but should, and reject everything that shouldn't. Read-only; it proposes motion with exact values, it does not implement it. Use when the user asks "what could be animated here?" or wants to "make this feel more alive". For fixing existing animations, use improve-animations or review-animations instead.

Generate adversarial PDF content that breaks extractors. Creates false-positive tables, malformed tables, cursed text, and layout traps. Extensible registry for adding new edge cases discovered from real-world PDFs.

Convert natural language requests into verifiable formal specs using interview loops.

Ensure movie subtitles exist and are verified on disk. Use when the user says get subtitles, download subtitles, verify English SRT, ask Bazarr for subtitles, ensure movie subtitles, or find subtitle-backed movie candidates.

Commit and push ONLY the named files onto origin/main, from any dirty checkout, branch, or worktree. Use when asked to commit and push, land changes, push relevant files to main, gcp, or scoped commit. Thin front door over ops-worktrees land; never stages repo-wide, never stashes, never switches branches, never deletes worktrees (the ops-worktrees reaper owns reclamation).

Deep GitHub repository and code search. For implementation research, uses the sibling brave-search skill to discover candidates, ranks them by requested criteria, stars, and relevance, clones the leaders to /tmp, verifies a detected entry point, then locates and evaluates the requested code.

Read-only goal-drift auditor. Holds each project's immutable goal as a typed record derived only from real human prompts, inspects what the project actually produced (tickets first, then artifacts, commits, and absences), and labels every action SERVES_GOAL, SUPPORTS_INDIRECTLY, DECLARED_DRIFT, UNTICKETED_WORK, SCOPE_DRIFT or MISSING_EXPECTED with a reason. Runs nightly on cron and reports; it never edits, commits, fixes, or reprioritises. Use when asked whether work is on track, whether an a...