ai-skills

Agent skills and agent definitions for Claude Code.

Skills

Skills fall into two groups plus one standalone advisor. The thinkies group under skills/thinkies/ collects reasoning and thinking techniques; it also installs as a bundle through the thinkies plugin (see Extensions). The tech group under skills/tech/ collects technical and engineering skills. The one flat skill, de-residency-advisor (skills/de-residency-advisor/), coaches non-EU expats preparing for German government appointments on visa, residency, and citizenship, researching each answer live and citing every claim.

thinkies

Skill	Reach for it when	Description
argue-the-opposite	You hold a position and nobody has pushed back yet	Stress-test a position by building the strongest counter-case
ask-questions	You're about to ask someone something, and it has to land	Ask the user the next good question, or make the better non-question move
ask-respond	A simple-looking question carries assumptions nobody checked	Structured Q&A that decomposes questions before answering
ask-what-breaks	You've reached a conclusion and want to know what would overturn it	Find defeaters that would break a conclusion
assess-current-knowledge	You can't say which of your beliefs you've actually checked	Map what's known vs assumed vs unknown
audit-chain-of-thought	An argument reaches a big conclusion and you can't see which step carries it	Tag reasoning steps by inference type
branch-possibilities	Every option on the table is a variation of the first idea	Generate fundamentally divergent directions from one starting point
calibrate-confidence	A claim is stated flatly and you can't tell how much of it the evidence carries	Match certainty to evidence strength
check-notes	You worked something out earlier and want the finding back, not a rerun	Find saved notes and run records and play them back
check-soundness	You've pulled several pieces into one plan and want to know they can all be true	Test synthesis for contradictions
cite-sources	A claim needs a source a reader can open	Track, validate, and cite external sources with working URLs
communicate	You have a piece to write or polish, and an audience it has to reach	Communicate ideas with purpose, clarity, and integrity
compose	You hold a pile of half-ideas that might add up to something	Join parts into a whole and find what it still lacks
connect-ideas	Two things feel alike and you want to know whether the likeness holds	Test how two ideas relate, or find a distant match for one
consider-alternatives	One explanation arrived first and nobody has looked for a second	Generate competing explanations for the same observations
decision-analysis	One choice turns on facts you don't control, and you keep going round in circles	Formulate and evaluate one concrete decision under uncertainty
decompose	A job is too big to start and every list of it feels arbitrary	Break a whole into parts at its natural joints, one axis per level
derive-first-principles	Something costs far more than it seems it should, and "that's how it's done" is the only answer	Strip convention to irreducible truths and rebuild
detect-diminishing-returns	You can't tell whether another pass is still making things better	Detect when further effort yields little gain
detect-fallacies	An argument feels persuasive and wrong at once	Spot logical errors in reasoning
domain-analysis	You're about to change a system and want to see what each piece is for	Model a system from its purpose down to its parts, with who acts on and sees each piece
evaluate-evidence	A few stories and a half-remembered study are offered as proof	Assess how well evidence supports claims
excavate-assumptions	A plan sounds obviously good and nobody has asked what it takes for granted	Surface unstated assumptions at multiple levels and rank them
find-leverage	The same problem comes back after every fix	Trace a system's feedback loops to find where a small change shifts the whole
flip-assumptions	A rule of thumb sounds right and you can't think how you'd ever check it	Test claims by forming the contrapositive
generalize	Something works in one place and you want to know what it's an instance of	Find the class a case belongs to and carry back what holds for it
generate-questions	A decision needs other people's answers, and you want the questions ready first	Compose a set of questions toward a driving question and return it without asking anyone
ideate	You need options you can compare side by side, not one idea	Generate and filter ideas into vetted options
instantiate	A value or policy sounds agreed, and you suspect everyone pictures something different	Make an abstraction concrete and find the conditions it assumed
integrate-other-perspectives	Several people want different things and it has come down to "whose wins"	Combine viewpoints into a coherent whole
integrity	A text makes confident claims and you want to know which ones its evidence carries	Verify epistemic integrity by aligning claims with evidence
invert-the-problem	Every direct attempt at a goal meets resistance	Turn a problem inside out to reveal hidden structure
map-out	An argument circles because each side is talking at a different level	Find where a subject sits and the level to act at
ponder	A problem stays vague and you can't yet say what you want	Explore a problem through a sequence of techniques before solving
probe-boundaries	A definition works for every case you've tried, and you haven't tried the odd ones	Test a claim or framing at its edges and extremes
prompt	You have a task for a model and only a rough sentence, or a prompt that grew by accretion	Craft or refactor LLM instructions
question-the-question	Answers keep coming back correct and unhelpful	Examine whether the inquiry is aimed at the right target
question-through-dialogue	You hold a conviction strongly and want to find out what it rests on	Use Socratic questioning to reveal assumptions
research-and-teach	You want to understand why something happens, starting from little background	Research deeply, explain progressively
run-premortem	You're about to commit to a plan that gets one shot	Imagine catastrophic failure and work backwards to prevent it
save-note	Something clicked and you'll want it again in a month	Save an insight so it can be found again
scamper	You have a concrete thing and want variations on it	Structured ideation using the SCAMPER creative thinking technique
shift-abstraction-level	A task arrives too vague to start or too specific to question	Find the level of abstraction to act at
shift-perspective	A decision is being made from one seat	Inhabit contrasting frames to see what one viewpoint misses
situate	Something works fine on its own terms and still seems to be failing	Place a thing in the larger system it serves
skill-design	You have an idea for a skill, or a skill that fires on the wrong requests	Design, strengthen, or audit an Agent Skill
strategize	Several constraints are tangled and you want to work through them together, step by step	Adaptive multi-phase reasoning for complex problems
surface-intent	You're about to add something to a system you haven't read all of	Surface intent before you add, change, or produce something
survey-peers	You know your option well and have never mapped what else does its job	Map what else fills the same role at the same level
synthesize-opposing-views	Two positions each sound right and the argument keeps circling	Find higher understanding through dialectic
trace-logic	An argument persuades you and you can't say which step you'd push on	Follow reasoning step-by-step
trace-logical-justifications	A rule is stated as obvious ("we must never…")	Trace justification chains to bedrock
tree-of-thought	The right approach isn't obvious and choosing wrong costs a lot	Systematic Tree of Thought reasoning for complex problem decomposition
tutor	You want to learn something by working through it, at your own pace	Interactive tutoring that adapts to your pace
visualize	You have numbers and a hunch, and need a chart that says one thing	Visualize data, concepts, relations, or diagrams as browser-runnable charts
what-if	A decision hangs on things nobody can know yet	Tile the space of possible futures and evaluate strategies across them
wonder	You're about to solve something and haven't asked what's strange about it	Open the possibility space through curiosity-driven questioning

tech

Technical skills grouped under skills/tech/.

Skill	Description
apache-age	Apache AGE (Postgres graph extension): Cypher + SQL patterns, schema modeling, query optimization
bash-scaffold	Scaffold a production-grade bash script from a curated template
cli-development	CLI development reference grounded in clig.dev
flix	Write, translate, and reason about Flix code
runbook	Decompose work into steerable autonomous loops, in seed and execute modes
shape-up	Conversational requirements elicitation producing shaped specifications
shell-testing	Write idiomatic BATS tests for bash and zsh shell scripts
ts-typeclasses	Implement typeclasses and their higher-kinded type encoding in TypeScript
waypoint	Distributed navigation markers for multi-file pipelines and processes

Agents

Agent	Description
research-surveyor	Rigorous topic surveys with cited sources
scout	Landscape reconnaissance and target identification
shell-dx-architect	Shell script DX: conventions, comments, consistency

Extensions

Claude Code plugin bundles live in extensions/. Each plugin wraps one or more skills and installs as a single unit via /plugin install … or claude --plugin-dir …. Two bundles ship here: de-residency wraps the de-residency-advisor skill, and thinkies bundles all 57 reasoning skills, each invoked as /thinkies:<name>. See extensions/README.md for details.

Install

Skills, three options:

# Option 1 — Claude Code / Cursor / Codex / any agentskills.io-compatible client:
npx skills add https://github.com/synapseradio/ai-skills

# Option 2 — Claude.ai upload: download the .skill file from packaged/
#   and upload it via Settings → Skills → Upload. Each skill's README has a
#   direct link to its .skill.

# Option 3 — copy a skill directory into ~/.claude/skills/ manually.

Agents: copy from agents/ into ~/.claude/agents/.

Extensions: claude --plugin-dir extensions/<plugin-name> for local use, or install from this repo's own marketplace:

/plugin marketplace add synapseradio/ai-skills
/plugin install thinkies@ai-skills

Packaging

Every source skill has a matching .skill ZIP under packaged/. The packaging tool is the skill-creator plugin's package_skill.py — invoked in place, not vendored. See CLAUDE.md for the full convention and re-packaging commands.

Development

Prerequisites

• Bun (package manager)
• ruff (brew install ruff)
• ShellCheck (available at /opt/local/bin/shellcheck or brew install shellcheck)

Setup

bun install
bunx lefthook install

Linting

Lefthook runs pre-commit hooks that auto-fix staged files:

Files	Tool
.md	markdownlint
.json	biome
.html	prettier
.yml / .yaml	prettier
.py	ruff
.sh	shellcheck

A commit-msg hook enforces conventional commits via commitlint.

To lint the entire repo manually:

bun run lint:fix

License

EUPL-1.2