cargo-skill

A Cargo subcommand for deploying and activating layered AI agent skills in Rust projects.

cargo install cargo-skill

---

What it does

cargo-skill has two jobs:

Install-time — detects your repo structure and agent tooling, then deploys the right skill files to the right paths.

Execution-time — writes a scoped context file (.skill/context.md) that controls which skill layer the agent uses for the current session.

---

Commands

cargo skill init                  Detect repo + agents, deploy skill files
cargo skill init --dry-run        Preview what would be deployed
cargo skill init --force          Overwrite existing skill files

cargo skill lookup [prefix]       Activate Layer 1 only (rule index, optional prefix filter)
cargo skill think                 Activate Layer 1 + 2 (lookup + reasoning)
cargo skill write                 Activate all layers (lookup + reasoning + execution)
cargo skill clear                 Remove .skill/context.md

cargo skill status                Show repo, agents, context, and gitignore status

cargo skill review                Review-focused context (err + test + lint + reasoning)
cargo skill refactor              Refactor-focused context (type + api + name + reasoning)
cargo skill debug                 Debug-focused context (err + mem + compiler quick-ref)

Quick prefix shorthand

cargo skill own            # Same as: cargo skill lookup own
cargo skill async          # Same as: cargo skill lookup async
cargo skill err            # Same as: cargo skill lookup err

All 14 prefixes work: own, err, mem, api, async, opt, type, perf, test, doc, name, proj, lint, anti

See skill layer reference for all available prefixes.

---

How it works

Install-time (init)

1. Detects repo type — single crate or workspace
2. Detects which agents are present (.claude/, .cursor/, .windsurf/, AGENTS.md)
3. Deploys bundled SKILL.md to the correct path for each detected agent
4. Adds .skill/ to .gitignore

Execution-time (lookup / think / write)

Writes .skill/context.md with the appropriate layer subset:

Command	Layer 1 (Lookup)	Layer 2 (Reasoning)	Layer 3 (Execution)
lookup [prefix]	✓ (filtered)
think	✓	✓
write	✓	✓	✓
review	✓ (err,test,lint)	✓
refactor	✓ (type,api,name)	✓
debug	✓ (err,mem)	✓ (quick-ref only)

The agent reads .skill/context.md as session context. This file is ephemeral — gitignored, overwritten on each invocation, deleted by clear.

---

Agent support

Agent	Install path
Claude Code	.claude/skills/rust.md
Cursor	.cursor/rules/rust.md
Windsurf	.windsurf/rules/rust.md
AGENTS.md	AGENTS.md (appended)

---

Skill content (v0.2.5)

Bundled skill covers Rust only. Three layers:

• Layer 1 — Lookup: 14-category rule index with priority levels (own-, err-, mem-, api-, async-, opt-, type-, perf-, test-, doc-, name-, proj-, lint-, anti-)
• Layer 2 — Reasoning: 3-layer cognitive model (Domain → Design → Mechanics), question routing, compiler error quick-ref
• Layer 3 — Execution: RPI loop (Research → Plan → Implement), verification checklist, task-to-rule mapping

Sources: leonardomso/rust-skills (MIT), actionbook/rust-skills (MIT), udapy/rust-agentic-skills (MIT).

---

Latest Updates

See CHANGELOG.md for the full release history. Recent highlights:

• v0.2.5 — Colored terminal output (green ✓, yellow ⚠, red ✗)
• v0.2.4 — Agent-specific context files for Cursor and Windsurf
• v0.2.3 — Claude Code context injection footer
• v0.2.2 — status command and comprehensive project overview

---

Requirements

• Rust stable (latest)
• Cargo

---

License

Licensed under either of:

• MIT license (LICENSE-MIT)
• Apache License, Version 2.0 (LICENSE-APACHE)

at your option.