cskill-agents

Curated, public-safe agent skills for building and improving coding CLIs, terminal tools, context engines, remote bridges, and multi-agent runtimes.

This repo contains original, method-level skill files distilled from reviewed implementations. It does not ship upstream source code, copied comments, tests, prompts, logos, or proprietary assets.

Quick Start

Install one skill with the helper scripts. These installers validate the slug and replace any existing installed copy, so they are safe for upgrades and reinstalls.

Codex-compatible

.\scripts\install-skill.ps1 -Agent codex -Slug concurrency-safe-tool-batching

./scripts/install-skill.sh codex concurrency-safe-tool-batching

Claude Code-compatible

.\scripts\install-skill.ps1 -Agent claude-code -Slug concurrency-safe-tool-batching

./scripts/install-skill.sh claude-code concurrency-safe-tool-batching

Install a curated bundle with the helper script:

.\scripts\install-bundle.ps1 -Agent codex -Bundle starter
.\scripts\install-bundle.ps1 -Agent claude-code -Bundle starter

./scripts/install-bundle.sh codex starter
./scripts/install-bundle.sh claude-code starter

What You Can Build

• Coding CLI command surfaces with dynamic skill loading, cache-safe registries, and curated slash-command discovery
• Tool runners with concurrency-safe batching, ordered streaming, output persistence, and deterministic budgeting
• Long-running agent sessions with compaction, memory prefetch, retry control, and stable session state
• Multi-agent systems with isolated child contexts, scoped MCP clients, transcript sidechains, and lifecycle cleanup
• Remote-safe bridges and terminal interfaces with capability filtering, privacy-aware init payloads, and motion-aware terminal UI

What This Repo Is

• A curated distribution repo built from a larger authoring corpus.
• A set of reusable skill files that tell agents how to reproduce strong patterns without copying implementation code.
• A practical install target for Codex-compatible and Claude Code-compatible skill directories.
• A bundle-based catalog of high-value agent-runtime patterns.

What This Repo Is Not

• Not a source mirror.
• Not a vendor-affiliated SDK, extension pack, or official template set.
• Not a redistribution of third-party code, copied comments, tests, prompts, or internal assets.
• Not the full authoring corpus; this repo only ships a curated public-safe subset.

Included Skills

• 158 curated skills
• 36 starter skills
• 19 curated bundles

Start with:

• Starter bundle
• Full curated index
• Machine-readable catalog

Supported Agents

• Codex-compatible folder layout in agents/codex
• Claude Code-compatible folder layout in agents/claude-code

Compatibility here means the repo uses a skill-folder layout that these agents can consume. It does not imply endorsement, certification, or affiliation.

Curated Bundles

• starter: best first install for most coding agents
• command-surfaces: command registries, skill loaders, and capability surfaces
• tool-orchestration: batching, streaming, persistence, and output shaping
• context-management: compaction, memory, and session control
• multi-agent: child-agent isolation, cleanup, and fork control
• extensions-mcp: extension lifecycle and parity patterns
• safety-worktrees: shell safety, path guards, and worktree handling
• remote-bridge: remote-safe command and bridge transport patterns
• terminal-ui: terminal presentation and feedback patterns
• registry-bootstrap: root resolution, registry composition, feature-gated loading, and cache-safe startup assembly
• skill-discovery: high-signal admission, filtering, listing, and first-turn discovery rules
• output-budgeting: tool result persistence, previews, truncation, and budget-aware output shaping
• compaction-core: retry, grouping, cache-edit, and compaction-control mechanics
• memory-signals: memory prefetch, recall selection, attachment timing, and session-memory shaping
• extension-sync: marketplace reconciliation, lifecycle parity, and runtime refresh consistency
• bridge-privacy: remote-safe bridge payloads, capability advertising, and normalized transport recovery
• agent-lifecycle: child-agent context handoff, scoped MCP setup, cleanup, and completion reporting
• fork-integrity: cache-stable forks, isolated skill runs, transcript hygiene, and resume consistency
• workspace-safety: shell validation, sensitive-path guards, and worktree-safe repo operations

Repo Layout

cskill-agents/
  README.md
  LICENSE
  NOTICE
  LEGAL.md
  PROVENANCE.md
  CONTRIBUTING.md
  catalog/
  bundles/
  skills/
  agents/
  scripts/

• skills/: canonical public universal skills
• agents/: direct-install compatible skill folders
• bundles/: curated install groups
• catalog/: machine-readable and human-readable indexes
• scripts/: install helpers and safety checks

Skill Format

Each public skill contains:

• SKILL.md: the reusable method instructions
• skill.yaml: sanitized public metadata for catalogs and agents

Public metadata is intentionally coarse. This repo does not publish raw source repo paths or internal mirror paths.

Public Provenance Model

Public artifacts in this repo use coarse provenance only:

• pattern family
• category
• review status
• bundle membership

Private maintainer evidence can exist outside this repo, but public files should remain abstraction-first and path-safe.

Legal And Non-Affiliation

Read:

• LEGAL.md
• NOTICE
• PROVENANCE.md

Short version:

• this repo licenses only its original authored materials
• third-party product names are used only for nominative compatibility references
• this repo is not affiliated with, endorsed by, sponsored by, or certified by Anthropic or OpenAI

Contributing

Before contributing, read CONTRIBUTING.md.

Public contributions must stay method-level. Do not add:

• source code snippets
• copied comments or tests
• raw internal file paths
• repo mirrors
• proprietary assets

Removal Requests

If you believe a public artifact in this repo is too close to upstream protected material, open an issue or contact the maintainer. See LEGAL.md for the review and correction process.