Claude Skills by agentydragon

Migrate a CNPG PostgreSQL cluster between Proxmox and Hetzner regions via streaming replication with sub-second downtime

Analyze CPAP sleep therapy data from the user's ResMed AirSense 11. Read daily summaries (AHI, leak, pressure, compliance) and per-session waveforms from EDF files served via WebDAV. Use when user asks about sleep quality, CPAP data, AHI, therapy compliance, or sleep analysis.

Eval-only stub skill. Contains no guidance — used as the "skill off" baseline arm so the sandbox shape stays uniform across `--skill on/off` rollouts.

Surface pending followups, natural extensions, incomplete migrations, code quality issues, and "what's next" suggestions. Verify session work is on disk. Use when wrapping up a task or session, or when user asks "what's next", "anything else", "what did we miss".

Deep forensic debugging that never stops until root cause is found or visibility limit is proven. Use when user wants to understand exactly why something is broken, not work around it. Activates on "why is this happening", "dig deeper", "don't work around it", "I want to understand", "find the root cause", "this seems suspicious", or when a problem suggests deeper breakage.

Use this skill for parametric 2D/3D technical drawings using FreeCAD Sketcher and TechDraw. Triggers when the user wants constrained parametric floor plans, mechanical sketches, layout diagrams, or any drawing where dimensions drive geometry. Also use when the user mentions FreeCAD, .FCStd files, Sketcher, TechDraw, parametric CAD, or technical drawings. Use this skill even for seemingly simple 2D layouts — the parametric constraint approach prevents coordinate drift and makes edits safe. Alw...

Ground a skill in tested, runnable examples. Use when creating a new skill or upgrading an existing skill to have CI-verified recipes. Triggers when the user asks to create a skill, add examples to a skill, or verify a skill has tested recipes.

Take and view screenshots of Hetzner Cloud servers via WebSocket VNC console. Use to diagnose issues when text commands fail — boot problems, unresponsive servers, kernel panics, stuck boot screens, graphical output inspection.

Optimal information gathering under uncertainty. Use when the main bottleneck is missing information and the right next step is to reduce uncertainty efficiently: research, compare options, design an experiment, elicit preferences, or decide what questions to ask next. Trigger on requests like "help me figure this out", "what should I find out", "what do I need to know to decide", or any situation where targeted questions, searches, data collection, or experiments would materially improve the...

Record a TODO item persistently in the repo (TODO.md, PLAN.md, or GitHub issue). PRIORITY - execute immediately, preempting any in-progress work. Use when user says "/later <thing>".

Audit and improve linter/checker configuration across any repo and language. Discovers languages, proposes useful checks to enable with real examples, creates PRs for approved checks, documents rejected ones. Also fixes misconfigurations, updates outdated versions, removes stale exclusions, and proposes tightenings.

Design an MCP server tool surface for an AI agent. Starts from common workflows, derives tool schemas, audits for completeness and safety. Use when designing a new MCP server or redesigning an existing one.

Iteratively pave a fiddly procedure into a verified runbook. Use when the user wants to "pave" something, or wants to figure out and document a complex, fiddly procedure (deploys, recoveries, env setup, vendor APIs, scraping, bootstrap sequences) that has rough edges and needs hands-on experimentation to get right — not just dry research. The loop: execute the current draft → observe friction → amend the draft (and extract small tools where useful) → repeat until a clean run requires zero imp...

Interact with Proxmox VMs - screenshots, keystrokes, network info (user)

Paraphrase vague instructions back with context, without executing changes. TRIGGER: any time the user says the word "readback".

Systematic binary reverse engineering toolkit. Extract source code, understand functions, document protocols, compare versions. Uses strings, symbols, disassembly, and differential verification.

Discover and analyze Claude Code session logs from ~/.claude/projects, including finding the current session and extracting tool calls, user messages, and conversation history

Make well-calibrated probability estimates using superforecasting methodology. Use when user asks about probability, likelihood, chance, odds, "will X happen", "when will X happen", "how much will X cost", "what could go wrong", failure modes, risk assessment, forecasting, or any question involving uncertainty and estimation.

Automated dependency updates — reads Renovate dashboard, applies safe updates, produces tested PRs from the agent's fork. One bulk PR for trivial bumps, separate PRs for non-trivial migrations. Use on a schedule or manually.

Verify documentation claims against actual code, finding and fixing stale or incorrect docs. Audit docs for token efficiency — cut what strong LLMs already know, keep local specifics and gotchas. Use when asked to review docs, trim docs, or check docs are accurate.

Diagnose the health of a Claude Code session (CLI or web) by running the observable acceptance criteria in the hook daemon SPEC against the live session. Also runs out-of-SPEC diagnostics (web_setup.sh freshness, claude-hooks pin staleness, bbr runner recycling, git hook origin-URL issues). Use when the user asks "did setup go ok", "why isn't bbr working", "check credentials", "selfcheck", "why do my commits fail", or any question about session health.

Show the current task stack and context. Use when user says "bt", "backtrace", "stack", "where are we", or asks about current progress on a multi-step task.

Eval-only stub skill. Contains no guidance — used as the "skill off" baseline arm so the sandbox shape stays uniform across `--skill on/off` rollouts.

Apply full judgment to "what would the user want to do, know, or reconsider next?" — from pending followups and incomplete migrations up to premortem risks, architectural rethinks, and owner's-eye strategy for the work lane. Verify session work is on disk. Use when wrapping up a task or session, or when user asks "what's next", "anything else", "what did we miss".

Deep forensic debugging that never stops until root cause is found or visibility limit is proven. Use when user wants to understand exactly why something is broken, not work around it. Activates on "why is this happening", "dig deeper", "don't work around it", "I want to understand", "find the root cause", "this seems suspicious", or when a problem suggests deeper breakage.

Inspect Forgejo repositories through the REST API and web endpoints, especially Actions CI, task/run metadata, logs, package registry tags, haku-state CI, and Flux/image rollout debugging. Use when diagnosing Forgejo Actions failures, missing logs, branch badge state, package/image publish gaps, or when endpoint shape is unclear.

Parametric 2D/3D technical drawings with FreeCAD Sketcher and TechDraw — constraint-driven floor plans, mechanical sketches, dimension-driven layouts. Use for anything FreeCAD/.FCStd, even seemingly simple 2D layouts, and read before writing any FreeCAD scripting code.

Download and view photos from a public Google Photos share link (photos.app.goo.gl or photos.google.com/share). Use whenever a Google Photos link should be viewed, described, or analyzed — web_fetch alone returns album metadata, never the pixels.

Ground a skill in tested, runnable examples. Use when creating a new skill or upgrading an existing skill to have CI-verified recipes. Triggers when the user asks to create a skill, add examples to a skill, or verify a skill has tested recipes.

Find an IKEA product and produce a correctly-scaled STL mesh ready for FreeCAD (search → GLB download → Draco-decompress → STL; also converts a bare .glb). Use when the user wants an IKEA item as a 3D/CAD model or a .glb turned into an STL.

Optimal information gathering under uncertainty: pick the research step, experiment, comparison, or question that reduces uncertainty fastest. Use when missing information is the bottleneck — "help me figure this out", "what do I need to know to decide".

Audit documentation, notes, runbooks, wiki-style knowledge bases, prompts, comments, UI/help copy, and other natural-language information systems for SSOT drift, duplication, stale claims, unclear ownership, and maintenance gaps. Produce prioritized, evidence-backed improvement suggestions with stable shorthand IDs like A, B, C so the user can choose what to execute.

Record a TODO item persistently in the repo (TODO.md, PLAN.md, or GitHub issue). PRIORITY - execute immediately, preempting any in-progress work. Use when user says "/later <thing>".

Audit and improve linter/checker configuration across any repo and language. Discovers languages, proposes useful checks to enable with real examples, creates PRs for approved checks, documents rejected ones. Also fixes misconfigurations, updates outdated versions, removes stale exclusions, and proposes tightenings.

Design an MCP server tool surface for an AI agent. Starts from common workflows, derives tool schemas, audits for completeness and safety. Use when designing a new MCP server or redesigning an existing one.

Iteratively pave a fiddly procedure (deploys, recoveries, env setup, vendor APIs) into a verified runbook: execute the draft, observe friction, amend, repeat until a clean run needs zero improvisation. Trigger on "pave this" or "let's figure out and document X".

Paraphrase vague instructions back with context, without executing changes. TRIGGER: any time the user says the word "readback".

Systematic binary reverse engineering toolkit. Extract source code, understand functions, document protocols, compare versions. Uses strings, symbols, disassembly, and differential verification.

Audit and prune large sets of remote GitHub branches using branch-specific semantic evidence, linked successor or merged-work proof, calibrated estimates of the owner's deletion decision, a high-throughput HTML decision cockpit, and SHA-guarded deletion. Use when asked which branches are droppable, stale work should be reviewed, branches without PRs need ranking, a branch-cleanup report or cockpit is wanted, or an explicitly approved branch batch should be deleted.

Discover and analyze Claude Code session logs from ~/.claude/projects, including finding the current session and extracting tool calls, user messages, and conversation history

Make well-calibrated probability estimates using superforecasting methodology. Use when user asks about probability, likelihood, chance, odds, "will X happen", "when will X happen", "how much will X cost", "what could go wrong", failure modes, risk assessment, forecasting, or any question involving uncertainty and estimation.

Automated dependency updates — reads Renovate dashboard, applies safe updates, produces tested PRs. One bulk PR for trivial bumps, separate PRs for non-trivial migrations. Use on a schedule or manually.

Verify documentation claims against actual code, finding and fixing stale or incorrect docs. Audit docs for token efficiency — cut what strong LLMs already know, keep local specifics and gotchas. Use when asked to review docs, trim docs, or check docs are accurate.

Navigate to the Blue Shield of California member portal message center and download all letter PDFs to `~/Downloads/bsc-letters/`. The user is already logged in. Start at: `https://www.blueshieldca.com/memberwebapp/connect/message-center/Inbox/` ---