Skills are small, versioned playbooks your coding agent loads on demand — each costs
only a name and one-line description until it's invoked. The portable layer of
Paniolo's harness tooling: one install puts them into every skill-capable agent
on your machine — Claude Code, Cursor, Copilot, Codex, Devin, Gemini CLI, and 70+ more.
Start here — the audit, remediation, and setup skills, for every agent
Measures how well a repository is set up for coding agents — and nothing else. The
skill runs the deterministic npx @paniolo/cli scan diagnostic —
a static, read-only pass that never writes a file — presents your scores across the
six harness dimensions with findings grouped by severity, and leads with the
lowest-scoring dimension, where remediation pays off most.
An opt-in AI review can add model judgment on top of the deterministic checks —
fenced off, so it never touches the scores. Then it stops: the report is the
deliverable, and the skill offers paniolo-scan-remediate as the next
step rather than fixing anything itself.
Reach for it when
"Scan my harness" · "Audit my CLAUDE.md setup" · "Check my AI agent
configuration" · /paniolo-scan
Fixes what the audit found. It takes a paniolo scan report —
reusing one from this session or producing a fresh read-only pass — groups the
findings into a severity-banded fix plan, and asks which band to apply
(High + Medium is the default).
Then your agent does the repair in your working tree, preferring shared guidance
over per-adapter duplication, and re-runs the same scan so you can see the grade
move — a before/after delta per dimension, not vibes.
Reach for it when
"Fix the scan findings" · "Improve my harness score" · "Clean up
these AGENTS.md findings" · /paniolo-scan-remediate
Writes a starter paniolo.config.json — the file that tells the scanner
which AI harnesses your repository is tuned for. It inspects the adapters already
present (CLAUDE.md, .cursor/, Copilot instructions,
AGENTS.md, GEMINI.md, .agents/), asks which of
the seven supported harnesses you want, and writes a deliberately minimal config.
The file grows with the rule set, so the skill never pre-populates it — and if a
config already exists, it shows you what's there before touching anything. The
scanner only ever reads this file; this skill is the only thing that writes it.
Reach for it when
"Init paniolo config" · "Set up paniolo.config.json" · "Which AI
vendors should this repo support" · /paniolo-config-init
Moves an existing paniolo.config.json forward — normalizing every
repo onto the repos map and factoring the policy a dozen siblings
repeat into shared repoDefaults and named repoTypes
layers (node → type → defaults → preset, host untouched).
It prefers the CLI's own writers —
paniolo evolve config migrate previews then applies a verified,
comment-preserving rewrite; evolve config effective answers
"where did this value come from" — and registers repos and wikis through
evolve harness connect and paniolo wiki init.
Reach for it when
"Migrate paniolo config" · "Dedupe these repo blocks" · "Where
does this setting come from" · /paniolo-config-upgrade
The conventions Paniolo agents run on, published as installable skills — each one a
tight, loadable playbook for a recurring kind of work. Install the ones your agent
needs — or the whole library at once.
Cloudflare CDN cache management and GitHub Actions CI/CD pipelines.
Every skill ships from
paniolo-ai/skills — MIT licensed and
listed on skills.sh — and the same
repo carries the Claude Code plugin and Antigravity workflows. The CLI they drive is
@paniolo/cli; install it directly with
npx @paniolo/cli scan . for a read-only scan, no account required.