Create and manage Metabase questions, dashboards, and public links. Use when the user wants to build metrics dashboards, create saved questions with SQL queries, or share analytics publicly.

Turn a rough bug report, feature request, support note, or pull request into a short, plain-language issue focused on the problem and desired behavior. Use when a contributor asks to simplify an issue, explain what a PR is for, create the corresponding issue for a PR, remove implementation detail from a report, or invokes /simple-issue-description.

Log genuine, recurring repository friction to .agents/PAPERCUTS.md — confusing setup, a flaky repo command or script, a misleading in-repo error, stale generated files, or a non-obvious gotcha that will cost the next contributor time. Also use to review, deduplicate, and resolve existing entries. Gate hard before logging: only friction the repository itself can fix counts. Never log the agent's own sandbox/permission errors, shell-scripting mistakes, transient flakiness, or third-party tool quirks the repo can't change.

Turn a PRD into a multi-phase implementation plan using tracer-bullet vertical slices, saved as a local Markdown file in ./plans/. Use when user wants to break down a PRD, create an implementation plan, plan phases from a PRD, or mentions "tracer bullets".

Cargo feature flags for crates/xberg — ORT-incompatible targets (WASM, Android x86_64 emulator), type-only and tract inference companion features, WASM/Android-safe variants, PDF backend, mutually-exclusive ORT variants, platform-conditional deps, aggregate feature sets, and build profiles. Load when adding, wiring, or debugging a Cargo feature, or when reasoning about what compiles on WASM/Android/Windows/macOS-intel targets.

Decide which xberg-io repository owns a cross-repository fix or API, and coordinate compatible changes across Xberg, Alef, enterprise, crawler, LLM, and OCR repositories. Load when work spans sibling repos; do not use for a self-contained Xberg edit.

MIME type detection and extractor routing in core/mime.rs — the FORMATS registry that EXT_TO_MIME and SUPPORTED_MIME_TYPES are derived from, the path-based and bytes-based detection functions, priority-based registry selection, wildcard MIME families, and the real procedure for adding a format. Load when adding a format, wiring an extractor to a MIME type, or debugging why a file routes to the wrong (or no) extractor.

Practice-profile interview that recommends and installs a starter pack of community legal skills. This IS the cold start for the whole ecosystem — it asks what kind of lawyer you are and recommends what to install first. Use on fresh install, when the user says "get me started" or "what should I install", or to re-run the integration-availability check after adding or removing an MCP connector.

Guided customization of your commercial contracts practice profile — change one thing without re-running the whole cold-start interview. Adjust risk posture, escalation contacts, playbook positions, NDA triage preferences, house style, review preferences, or matter workspace paths. Use when the user says "change my [thing]", "update my profile", "edit my playbook", "tune my config", or "customize".

Case status summary by audience — client-facing (plain language), internal (for the professor), or court-ready (formal caption format per local rules). Same facts, different framing and depth. Use when a student needs to update the client, brief the professor, or prepare a court status report.

Evaluate a skill against the Legal Skill Design Framework — thirteen design parameters (including trust-surface, freshness, schema validation, and conflict detection), three legal failure modes, and a three-band verdict (Ready / Some Concern / Material Concerns). Use when deciding whether to trust a community skill before installing it, before deploying a first-party skill to your team, or whenever the user asks "should I trust this?" or "is this skill well-designed?". Runs automatically as part of /legal-builder-hub:skill-installer.

Run the cold-start interview — learns your privacy practice and writes CLAUDE.md from your policy, DPA template, and a reference PIA. Use on first run, when CLAUDE.md is missing or has placeholders, or when the user says "set up the privacy plugin", "onboard me", "configure privacy", or wants to re-run the interview or re-check integrations.

Install a community skill from a watched registry. Reads the allowlist first, fetches, shows the RAW SKILL.md (not just a summary), runs structural trust checks, runs skills-qa, and only writes files after explicit user approval. Use when the user says "install [skill]", picks install from browse, or provides a direct skill URL.

Track the IP portfolio — registrations, renewals, maintenance fees, and use declarations. Use when checking what's renewing, adding or updating an asset, recording a maintenance filing, or auditing the register for gaps, lapses, and use-in-commerce questions. Receives handoffs from prosecution and clearance work.

Infringement triage across trademark, copyright, patent, and trade secret — a flag list with the factors cutting each way, not a finding. Use when assessing whether someone is infringing your IP or whether you might be infringing theirs, when a knockoff or copycat surfaces, or when deciding whether a matter is worth pursuing and how.

Draft a cease-and-desist letter (send mode) or triage one you received (receive mode). Use when asserting your rights against an infringer with a demand letter calibrated to your enforcement posture, or when an incoming C&D needs triage into a structured options memo with a recommendation.

Run the cold-start interview to learn your IP practice and write your practice profile. Use on first install when the practice profile is missing or still contains placeholders, when re-onboarding with --redo, or when re-probing integrations with --check-integrations after connecting or disconnecting an MCP. This is the ONLY skill that should run on a fresh install.

Freedom-to-operate triage — a structured first look at potentially blocking patents, not an FTO opinion. Use when a product, process, or feature is being evaluated for blocking patents, when asked whether anything stops a launch, or to build a claim-chart first pass against the most plausible patents before patent counsel review. This skill never concludes a product is clear to launch.

Optimizes transformer attention with Flash Attention for 2-4x speedup and 10-20x memory reduction. Use when training/running transformers with long sequences (>512 tokens), encountering GPU memory issues with attention, or need faster inference. Supports PyTorch native SDPA, flash-attn library, H100 FP8, and sliding window attention.

Use this skill when an InsForge maintainer has finished an OSS repo change and is ready to open, update, or submit the InsForge PR. Runs the release-quality deterministic E2E gate by building a package.json-derived InsForge test image tag, deciding whether sibling agent-e2e fixture coverage must change, dispatching the Deterministic Fixture E2E workflow, waiting for results, and triaging failures before PR submission.