Agent Skills

Reusable skills for AI coding agents. Each skill packages a focused workflow, domain rules, primary references, and completion checks in a self-contained directory.

The repository uses the Agent Skills layout understood by Oh My Pi and compatible skill loaders. Each skill lives in a directory with a SKILL.md file. A skill can also include assets/, references/, and scripts/.

Available skills

Agent tooling

Skill	Context tokens	Purpose
agent-skills-standard	2,880	Audit new and updated Agent Skills packages, gate skill repository commits, migrate catalogs, and add standards-compatible discovery and activation to agent clients.
omp-skill-hardener	2,412	Mine repeated failures from OMP sessions, turn them into approved skill or AGENTS.md changes, and test the new rules.

Cryptography

Skill	Context tokens	Purpose
nist-cavp	2,729	Find, download, parse, and integrate NIST CAVP archives and ACVP vector sets for cryptographic primitives and components.
noise-protocol	4,426	Design, integrate, test, and review Noise handshakes, protocol suites, transport states, and key lifecycles.
ristretto255	2,790	Implement, integrate, and review ristretto255, including canonical encoding, hash-to-group, scalars, constant-time operations, protocol use, and RFC vectors.
wycheproof	2,060	Integrate and audit current Project Wycheproof vectors against cryptographic implementations, schemas, and result semantics.

Development

Skill	Context tokens	Purpose
android-ble	6,526	Implement, migrate, test, and troubleshoot Android BLE scanning, GATT lifecycle, background, permissions, and Android 17 behavior.
android-ble-gatt-queue	4,702	Implement, review, test, and debug coroutine-serialized Android GATT client operations, callbacks, cancellation, and timeout reset.
android-ble-gatt-server	5,110	Implement, review, test, and troubleshoot Android local GATT databases, ATT requests, prepared writes, subscriptions, and updates.
android-ble-gatt-status	5,294	Diagnose, mitigate, and test Android GATT connection/status failures, including opaque legacy status 133.
android-ble-inspector	4,890	Build, review, test, and troubleshoot Android Compose BLE scanners, GATT trees, operation controls, and value decoders.
android-bluetooth-sockets	5,438	Implement, review, test, and troubleshoot Android RFCOMM and LE CoC socket clients, servers, streams, and lifecycle.
android-debug-bridge	5,229	Install, configure, use, and troubleshoot adb device targeting, USB/Wi-Fi transports, shell commands, apps, files, ports, and diagnostics.
arrange-act-assert	4,174	Structure, review, and refactor example-based tests into clear Arrange, Act, and Assert phases.
ble-protocol-stack	5,532	Design, validate, and troubleshoot BLE GAP, GATT, ATT, L2CAP schemas, procedures, caching, and channels.
ble-throughput	5,004	Measure, model, diagnose, and optimize BLE throughput across PHY, Link Layer, ATT/GATT, mobile, and application queues.
compose-multiplatform	5,518	Design, implement, test, and ship shared Compose UI across Android, iOS, desktop, and web.
corebluetooth	5,280	Implement, review, test, and troubleshoot Apple Core Bluetooth central/peripheral apps, lifecycle, data flow, and restoration.
detekt	3,556	Configure, run, migrate, and troubleshoot detekt static analysis for Kotlin projects.
kover	3,182	Configure, verify, aggregate, and troubleshoot Kotlinx Kover JVM coverage for Kotlin projects.
kotlin-api-reference	3,279	Find and verify versioned, platform-specific Kotlin ecosystem API declarations and source.
kotlin-binary-compatibility	4,351	Configure, run, migrate, and review Kotlin ABI validation with built-in KGP or the legacy validator.
kotlin-coroutines	4,709	Design, implement, test, and troubleshoot Kotlin coroutines, Flow, channels, cancellation, and structured concurrency.
kotlin-development	3,262	Implement, review, build, test, and troubleshoot Kotlin projects across JVM, Android, Kotlin Multiplatform, JavaScript, Wasm, and Native.
kotlin-gradle	4,366	Configure, migrate, optimize, and troubleshoot Kotlin Gradle builds, toolchains, compiler options, and caches.
kotlin-multiplatform	4,276	Design, configure, migrate, test, publish, and troubleshoot Kotlin Multiplatform targets, source sets, hierarchies, and variants.
kotlin-native-apple-interop	4,644	Configure, export, import, and troubleshoot Kotlin/Native interoperability with Swift, Objective-C, and Apple frameworks.
kotlin-power-assert	3,850	Configure, use, debug, and expose Kotlin Power-assert diagnostics, transformed functions, and assertion APIs.
kotlinx-benchmark	4,430	Configure, run, compare, and troubleshoot multiplatform Kotlin microbenchmarks with kotlinx-benchmark.
kotlinx-serialization	4,613	Design, configure, evolve, test, and troubleshoot kotlinx.serialization wire formats and schemas.
ksp	5,118	Configure, author, test, migrate, and troubleshoot Kotlin Symbol Processing consumers and processors.
lincheck	3,787	Design, run, interpret, and troubleshoot JVM concurrency tests with Lincheck model checking and stress strategies.
skie	3,315	Install, migrate, configure, and troubleshoot Touchlab SKIE for Kotlin Multiplatform Swift interop.
spotless	3,947	Configure, apply, verify, migrate, and troubleshoot Spotless formatting for Gradle and Maven projects.
xtool	2,285	Install, configure, use, and troubleshoot xtool for SwiftPM-driven iOS development and device deployment.

Documentation

Skill	Context tokens	Purpose
diataxis	2,693	Write, audit, and improve tutorials, how-to guides, reference, explanation, documentation architecture, and documentation quality.
dokka	4,486	Configure, generate, publish, migrate, and troubleshoot Dokka API documentation for Kotlin and mixed Java projects.

Open a skill's SKILL.md for its full procedure and source material.

Context tokens use tiktoken>=0.14 with the GPT-5.6-compatible o200k_base encoding. Each value is the sum of SKILL.md and direct files under assets/ and references/; scripts and license files are excluded. Resources load only when needed, so the table is a packaged-context upper bound rather than the cost of every activation.

Regenerate the table values from the repository root:

python3 agent-skills-standard/scripts/count-context.py --root .

Categories are catalog metadata. They do not affect skill invocation, and skill directories remain at the repository root.

Install

The skills CLI requires Node.js and npx.

List the skills in this repository before installing them:

npx skills add trancee/skills --list

Install skills for the current project:

npx skills add trancee/skills

The CLI detects supported agents and prompts you to choose the skills and target agents. Project installation is the default.

To install one skill, pass its name:

npx skills add trancee/skills --skill kotlin-development

To make the selected skills available across projects, use global scope:

npx skills add trancee/skills --global

Review each skill before installing it. Skills can direct an agent to edit files, run commands, or access the network.

Use

Skills with a description can be selected automatically when a request matches their scope. You can also invoke one explicitly by name:

Use the nist-cavp skill to add AES-GCM vectors to this test suite.

The agent should read SKILL.md before acting, follow relative references from the skill directory, and satisfy the skill's completion checks. Explicit project instructions and user requirements still take precedence.

Update

Update installed skills and choose the scope when prompted:

npx skills update

Select project scope, global scope, or one named skill directly:

npx skills update --project
npx skills update --global
npx skills update kotlin-development

Repository layout

skills/
├── README.md
├── <skill-name>/
│   ├── SKILL.md
│   ├── assets/        # optional fixtures and templates
│   ├── references/    # optional detailed guidance
│   └── scripts/       # optional executable helpers
└── ...

A skill's frontmatter provides its discovery contract:

---
name: example-skill
description: Perform a specific workflow when its triggering conditions apply.
metadata:
  category: "development"
  source: "https://example.com/canonical-source"
  sourceVersion: "upstream release, document, page revision, or commit"
  createdBy: "github-copilot/gpt-5.6-sol"
  createdAt: "2026-01-01T12:00:00+00:00"
  updatedBy: "github-copilot/gpt-5.6-sol"
  updatedAt: "2026-01-02T12:00:00+00:00"
---

• name matches the lower-kebab-case directory name.
• description states what the skill does and the distinct situations that should trigger it.
• metadata.category groups the skill under agent-tooling, cryptography, development, or documentation in this repository.
• metadata.source identifies the canonical source used to author and refresh the skill.
• metadata.sourceVersion records the upstream release, document identifier, dated page revision, or commit used to author or refresh the skill.
• metadata.createdBy records the provider/model identifier that first created the skill.
• metadata.createdAt records the skill's first repository creation time as an ISO 8601 string.
• metadata.updatedBy records the provider/model identifier that made the latest content or metadata change.
• metadata.updatedAt records the latest skill content or metadata update as an ISO 8601 string.
• The body contains ordered actions, decision points, constraints, and checkable completion criteria.
• Relative links resolve from the skill directory.

Contributing

Write for the consumer. Keep SKILL.md, agent-facing references, and agent-filled templates terse. Use stable labels, fragments, symbols, and exact commands when they preserve meaning. Remove connective prose. Write human-facing documentation, including this README, in clear natural English.

1. Use skill-creator to create a skill or revise its procedure.
2. Run agent-skills-standard on every new or updated package before calling the package complete.
3. Give the skill one focused responsibility and a precise discovery description.
4. Put the main procedure in SKILL.md. Move fixtures and templates to assets/, detailed guidance to references/, and executable helpers to scripts/.
5. Prefer primary, versioned sources. State which source wins when references disagree.
6. Keep secrets, credentials, generated output, and machine-specific paths out of the skill.
7. Exercise commands and behavioral procedures in an appropriate disposable environment.
8. After any SKILL.md, direct references/*, or direct assets/* change, run python3 agent-skills-standard/scripts/count-context.py --root . --baseline HEAD path/to/skill. Record the o200k_base core, resource, total, baseline, and delta counts.
9. Before committing skill repository changes, invoke agent-skills-standard again. Re-run its strict audit and token count against the final files, then run metadata, link, discovery, and diff checks.
10. If the catalog changed, run npx skills add . --list and confirm that it exposes every intended skill.
11. Use Conventional Commits for repository history.

A useful contribution leaves no ambiguity about when the skill applies, what the agent must do, or how completion is verified.