skills

AssetsArt's collection of agent-callable skills — Rust CLI tools plus process orchestration skills.

skills is a Cargo workspace plus a skill registry. Most members are small, focused command-line tools written in Rust; a few are process skills with no binary at all (orchestration patterns the agent invokes by reading a SKILL.md). Every skill ships:

• a SKILL.md manifest the agent reads to know when and how to call it,
• (for CLI skills) a single self-contained binary built from a sibling crates/<name> workspace member,
• a human-facing README.md with examples.

Skill Index

Skill	Kind	What it does	Crate
ny-codemap	CLI	Survey a codebase: list files, show symbols, find definitions	crates/codemap
ny-codegraph	CLI	Semantic cross-references: find-refs, callers, callees, impact	crates/codegraph
ny-astedit	CLI	AST-validated rewrites: rename (cross-file symbol rename via codegraph) and rewrite (structural pattern→rewrite via ast-grep). Dry-run by default; atomic per-file writes.	crates/astedit
ny-auto-pipeline	Process	Autonomous brainstorm → spec → spec-review → plan → subagent-driven impl → scrutinize → post-mortem. Overrides the interactive gates of superpowers:brainstorming / writing-plans when the user has granted autonomous run authority. Composes with 9arm-skills:scrutinize / post-mortem / debug-mantra and superpowers:verification-before-completion.	—

Install (end users)

Via Claude Code plugin (recommended — auto-updates the SKILL manifests)

In a Claude Code session:

/plugin marketplace add AssetsArt/skills
/plugin install ny-skills

Claude Code refreshes the SKILL manifests on every restart and on /reload-plugins, so pushes to main reach agents automatically. The plugin ships SKILL.md + query files; the compiled CLI binary still needs to be fetched once with the installer below.

One-liner (no clone) — also fetches the platform binary

curl -fsSL https://raw.githubusercontent.com/AssetsArt/skills/main/scripts/install.sh | sh
# pin a specific version:
curl -fsSL https://raw.githubusercontent.com/AssetsArt/skills/main/scripts/install.sh | sh -s -- v0.4.1

The script fetches the repo's source tarball at the resolved tag (no git required) and stages each skill into ~/.claude/skills/ny-<name>/. Override the destination with CLAUDE_SKILLS_DIR=/some/path. Use this in addition to the plugin install above (or on its own) — the plugin handles SKILL.md updates, this handles binary updates.

From a checkout

./scripts/install.sh           # downloads the latest release for your platform
./scripts/install.sh v0.4.1    # or pin a specific version

Either entry point lands the binary at skills/ny-<name>/scripts/<name> and also copies the skill dir into ~/.claude/skills/ny-<name>/ so Claude can discover it. If a ~/.claude/skills/ny-<name> already exists as a symlink (manual setup), the script leaves it alone.

Supported asset slugs: linux-gnu-x86_64, linux-gnu-aarch64, linux-musl-x86_64, linux-musl-aarch64, macos-x86_64, macos-aarch64. The script auto-detects the right slug from uname + libc probe; override with SKILLS_TARGET=<slug> if you need to (e.g. installing into an Alpine container from a glibc host). If you hit GitHub's 60/hr unauthenticated API rate limit, export GITHUB_TOKEN before running.

Build from source (developers)

./scripts/build-skills.sh

Builds every workspace crate that has a matching skills/<name>/ directory in --release mode and copies each binary into skills/<name>/scripts/<name>. The same layout install.sh produces.

Security / integrity

Release tarballs will be paired with .sha256 companions generated in the same CI job that builds them. install.sh verifies the checksum and refuses tarballs containing absolute paths or .. segments before extraction. Code signing (Apple notarisation, sigstore) is intentionally out of scope for this revision.

Adding a new skill

Two flavours.

CLI skills (the common case)

Source lives in crates/<name>/, and the agent-installed surface lives in skills/ny-<name>/. The mapping skill_dir == "ny-" + crate_name is load-bearing — build-skills.sh, install.sh, and release.yml all rely on it.

1. Create crates/<name>/ (with Cargo.toml, src/main.rs, etc. — use crates/codemap as a template). Inherit shared metadata via version.workspace = true, license.workspace = true, etc.
2. Create skills/ny-<name>/SKILL.md with frontmatter name: ny-<name> and a description field so agents can discover it. Reference the binary as ./scripts/<name> (the pre-built file lives at skills/ny-<name>/scripts/<name>; the scripts/ directory is gitignored).
3. Add a row to the Skill Index table above with Kind = CLI.
4. Run ./scripts/build-skills.sh to verify the local build pipeline picks up the new skill. The release workflow does not require changes — its matrix is per-target, not per-skill.

Process skills (no binary)

Sometimes the skill is an orchestration pattern, not a CLI. Example: ny-auto-pipeline. These ship a SKILL.md only.

1. Create skills/ny-<name>/SKILL.md with frontmatter name: ny-<name> and a description that lists trigger phrases (English + Thai if applicable) plus negative cases.
2. No crates/<name>/ directory; no entry in the workspace Cargo.toml.
3. Add a row to the Skill Index table above with Kind = Process and crate column —.
4. build-skills.sh and install.sh skip directories without a scripts/ subdir automatically — no other infra changes needed.

YAML frontmatter discipline: if the description value contains : (colon + space) anywhere, wrap the whole value in double quotes — YAML will otherwise read the colon as a key/value separator and the manifest will fail to load.

Versioning

The workspace uses version.workspace = true, so all skills bump together. This is the intended default — release coordination lives at the workspace level. If you need per-skill versioning later, drop version.workspace on the specific crate.

License

MIT — see LICENSE.