AgentSpan Skills
https://github.com/agentspan-ai/agentspan-skills- Category
- AI Agents
- Rank
- No. 2042Tools index
Previous survey · No. 2035 ·
- Pricing
- Open Source
- Type
- AGENT
- Builder
- agentspan-ai
- GitHub
- 4 stars
- Date
About
AgentSpan skills for AI coding agents — Claude Code, Codex, Cursor, Windsurf, and more.
What it does
It gives coding assistants practical guidance for creating durable AgentSpan workloads. The material covers tools, guardrails, structured output, multi-agent orchestration, monitoring, event streams, approvals, and credential handling. Installers place a shared instruction bundle into each supported assistant's configuration area.
Why it's ranked here
A strong choice for teams already committed to AgentSpan. It combines focused reference material, task-specific examples, cross-platform installers, a standard-library REST fallback, and automated evaluation scenarios. That scope makes it more operationally useful than a prompt collection, while remaining tightly tied to one runtime.
What's good
The bundle covers the full working loop: define agents, start durable executions, inspect failures, stream events, and answer approval requests. It documents Python, command-line, and REST access. The fallback client avoids third-party dependencies, retries selected transient HTTP failures, supports authentication, and reads server-sent events.
Tradeoffs
This is guidance and helper tooling, not the AgentSpan runtime itself. Users still need the separate SDK or a reachable server. The recommended shell installation executes remotely fetched code. Its evaluation runner sends prompts to external model providers, requires API keys, and relies on another model to judge responses.
How to use it well
Use it when a coding assistant regularly builds or operates AgentSpan agents, especially workflows needing durable approvals, execution search, or live event monitoring. Install it for the assistants your team actually uses, then validate changes with the supplied scenarios. It does not replace general agent framework selection, hosting, or the underlying runtime.
Technical notes+
install.sh and install.ps1 detect supported assistants, download the files listed in SKILL_FILES, write .agentspan-skills-manifest, and support check, upgrade, force, and uninstall modes. skills/agentspan/scripts/agentspan_api.py is a standard-library CLI over the REST API, with bearer-token acquisition, retries for selected HTTP statuses, execution search, approval responses, and SSE streaming. scripts/run_evals.py loads JSON cases from evaluations/, calls Anthropic, OpenAI, or Gemini endpoints, then uses a second model as judge and treats scores of 70 or higher as passing. CLAUDE.md records verified SDK conventions and known invalid patterns. .claude-plugin/plugin.json and .claude-plugin/marketplace.json provide the Claude Code packaging metadata.
Observed
- License
- MIT
- Languages
- Skill content is Markdown; helper tooling uses Python, Bash, and PowerShell.
- Install surface
- Claude Code plugin plus shell installers for macOS, Linux, and Windows.
- Interfaces documented
- Python SDK, command-line interface, and REST API.
- Assistant support
- Claude Code, Codex, Gemini CLI, Cursor, Windsurf, Cline, GitHub Copilot, Aider, Amazon Q, Roo, Amp, and OpenCode.
- Fallback client
- A Python standard-library REST command-line client is included.
- Evaluation structure
- JSON evaluation scenarios and a multi-provider Python evaluation runner are included.
Read from README.md, scripts/run_evals.py, skills/agentspan/scripts/agentspan_api.py, VERSION, CLAUDE.md, install.sh, LICENSE.txt, _config.yml, install.ps1, agents/openai.yaml, evaluations/README.md, .claude-plugin/plugin.json, evaluations/hitl-approval.json, evaluations/stream-events.json, .claude-plugin/marketplace.json.
What it can do
Generate code from natural language prompts
Natural language description of desired functionality → Source code in specified programming language
Debug and fix code errors
Broken or buggy code → Corrected code with fixes applied
Refactor existing code
Legacy or poorly structured code → Improved, optimized code structure
Explain code functionality
Source code files or snippets → Plain language explanations of what the code does
Convert code between programming languages
Code written in one programming language → Equivalent code in different programming language
Generate unit tests for code
Source code functions or methods → Automated test cases and test code
Optimize code performance
Slow or inefficient code → Performance-optimized version of the code
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.