agent-skills

Claude Code skills I wrote and use daily. Thirteen of them, grouped by what they actually do rather than by how they're implemented.

Each skill is a directory under skills/ containing a SKILL.md with YAML frontmatter, plus any scripts or reference docs it needs.

The skills

Document and media pipelines

skill	what it does
scan-organize	Batch pipeline for scanned paper: separates, deskews, OCRs, classifies, tags and renames, then writes metadata back to the filesystem. Never deletes — everything lands in a staged review directory first.
document-file-handler	Pulls text, tables and metadata out of PDF / DOCX / XLSX / PPTX using local CLI tools, so binaries never get loaded into the context window whole.
lineart-svg	Photo to single-colour SVG icon through a checkpointed review loop. Raster generation and vectorisation are separated by a human approval gate.
bookfinity-build	Generates Gridfinity book-box covers for Bambu Studio, including the .3mf emboss-text surgery that silently prints stale text if you edit it by hand.

Machine and session telemetry

skill	what it does
hardware	Writes a structured snapshot of the machine's physical hardware to a single markdown file, for another AI session to read as context. Idempotent.
software	Same idea for the dev toolchain — languages, package managers, IDEs, local model runtimes, installed skills.
token-dashboard	Cost, token, cache-hit and tool-usage analytics over local Claude Code session logs. Reads ~/.claude/projects only; nothing leaves the machine.
agent-monitor	Real-time dashboard of sessions, tool calls and subagent orchestration, fed by Claude Code hooks.
daily-tech-summary	Aggregates a day's engineering activity into a committed markdown log. Reads local browser history and Gmail — see Privacy below.

Working method

skill	what it does
orchestrator	Runs non-trivial work as a delegated multi-agent pipeline — assess, route, plan, delegate, review, verify — instead of editing files directly.
context-engineering	Token-budget discipline for large or unfamiliar repos: build a compact codebase map rather than reading everything.
workflow-visualizer	Turns processes, code paths, state machines and dependency graphs into Mermaid diagram source.
reins	Drives your real, logged-in browser from the shell. Wraps a third-party CLI — see Credits.

Install

Copy the ones you want into your skills directory:

git clone https://github.com/RAEngineered/agent-skills
cp -r agent-skills/skills/token-dashboard ~/.claude/skills/

Claude Code picks them up on the next session start. They're independent — there is no shared runtime, so take one or take all thirteen.

Paths

These were written on Windows against my own layout. Machine-specific paths are replaced with placeholders you'll need to set:

placeholder	meaning
%USERPROFILE%	your home directory
%DOCS%	wherever you keep working documents
%PICTURES%	your image directory

Privacy

Two skills touch data worth knowing about before you run them:

• daily-tech-summary reads local browser history databases (Chrome, Edge), Gmail via an MCP connector, and local AI session logs, to reconstruct what you did that day. It writes the result to a git repo. Read references/data-sources.md and decide for yourself before pointing it anywhere.
• reins drives your actual signed-in browser, which means it operates with your live sessions and cookies. That is the entire point of it, and also the entire risk.

Everything else is local-only and read-only against your own files.

Credits

Some of these are wrappers around other people's tools. The skill is mine; the tool underneath is not:

• reins installs and drives @karnstack/reins.
• document-file-handler defers to the Anthropic-provided pdf / docx / xlsx / pptx skills when the job is authoring rather than extraction.

If you're looking for the large community skill packs, those live upstream and are deliberately not vendored here — see everything-claude-code and caveman.

Licence

MIT — see LICENSE.