Agent Network: your local agent team
Astra and 12 specialists in a local Codex system. Assign work through a map or controller chat, collect result files and review the execution history.
FOR AGENTS / LIBRARYResources
Resources
for your agent.
Skills, prompts, templates and checklists for work. Choose a resource, share it with your agent or download the complete file.
Skills, templates and checklists for you and your agent
CT toolkit: one ZIP for Codex and Claude Code
Give the archive to your agent and run a public CT demo. Includes a coordinating skill, the Slicer skill, the TotalSegmentator skill and CT-CHAT/Merlin templates.
Speed up React and Next.js
Find request waterfalls, heavy imports and unnecessary renders. Vercel’s rules help prioritize changes worth measuring first.
Design an interface around the product
Define a visual direction, typography and layout, then check them against the brief. Anthropic’s workflow connects design choices to the product.
PostgreSQL queries, indexes and access
Review slow SQL, schema design and row-level access. Supabase provides explanations and examples for PostgreSQL, including deployments outside Supabase.
React video: scenes, audio and captions
Remotion’s official skill collection covers composition, preview and rendering, with dedicated guidance for timing, audio, captions and multiple scenes.
Build and validate dbt data models
Start with the target table, work back to its sources and validate transformations against data. dbt Labs connects SQL, model dependencies and tests.
Threat modeling grounded in source code
Connect entry points, valuable assets and trust boundaries to concrete abuse scenarios. OpenAI’s workflow separates observed controls from assumptions.
An SEO audit with evidence and priorities
Review crawlability, canonical URLs, content and language variants. Connect each finding to observed evidence and a concrete fix.
React components without a maze of flags
Replace overloaded components with composable parts, explicit variants and shared state. Vercel illustrates component API design with concrete examples.
Test a web app through the browser
Inspect the rendered page before exercising a user journey. Anthropic includes Playwright examples and a helper for starting local servers.
A document a new reader can understand
Gather context, refine the document section by section and test it with a fresh reader. Anthropic’s method helps expose assumptions familiar to the authors.
Slicer: a skill for code and 3D scenes
Find answers in 3D Slicer source and documentation, prepare a script and, with a separately connected MCP server, inspect a real scene using a public sample.
TotalSegmentator: a skill for choosing segmentation tasks
The official skill guides an agent through MCP capability discovery, task, device and speed selection, then records the generated masks and run settings.
Answers from documents
Compare files, uncover contradictions, and link conclusions to evidence.
Audit agent instructions
Find conflicting rules and unnecessary requirements in SKILL.md and AGENTS.md.
Research with sources
Answer a work question using primary sources and make evidence gaps visible.
Interview patterns
Extract jobs and recurring problems from interviews while keeping quotes and context.
Decision memo
Compare alternatives using explicit criteria and recommend a decision with revisit conditions.
Plain-language editor
Remove jargon and repetition while preserving the author’s voice, facts, and uncertainty.
Repurpose a source
Turn a source into a post, email, or script without inventing facts.
Meeting to actions
Extract decisions, owners, and deadlines from a transcript, separating commitments from ideas.
Idea to product brief
Define the user problem, first-version scope, and observable acceptance criteria.
Reproducible bug report
Turn “it is broken” into clear reproduction steps, expected behavior, and evidence.
Review a dataset
Find missing values, duplicates, mixed units, and ambiguous dates before analysis.
Review a change
Review a specific diff for behavior bugs and regressions, with line references.
Agent task brief
A task prompt with context, source material, scope, and a clear definition of done.
Project AGENTS.md
A concise starting point for project structure, real commands, and completion criteria.
Evidence matrix
Track which claim relies on which source, including dates and disagreements.
Acceptance criteria
Define observable success, errors, empty states, and data preservation.
Experiment card
Define a hypothesis, metric, constraints, and decision rule before running a test.
Task handoff note
Hand work to another person or agent with state, decisions, files, and a next step.
Check an AI answer
Check facts, calculations, and task fit before relying on an AI output.
Is the skill ready?
Check a skill’s purpose, content, activation, and boundaries on realistic requests.
Evaluate a source
Separate original evidence from retellings and assess its relevance to the question.
Prepare inputs for an agent
Provide useful context and files while preserving structure, versions, and sensitive information.
Why is the agent stuck?
Identify rules that cause unnecessary reading, repeated checks, or premature stops.
Website release checklist
Walk through links, forms, mobile layouts, downloads, and failure states before publishing.
CT-CHAT: a research reproduction plan
Check the environment, GPU, weights and chest CT encoding before running the model. An original planning template; CT-CHAT is a research model, not an agent skill.
Merlin: a report-generation reproduction plan
Separate library setup, checkpoint downloads and model-generated text for one public sample. An original research template, not a Merlin agent skill.
FIRST RUNFrom file
From file
to useful work.
- Choose one skill
Extract its folder into .agents/skills in your Codex project. Each skill folder contains SKILL.md. Select it in the interface; in CLI and IDE you can invoke it with $skill-name.
- Add your task
Each resource includes an example prompt. Supply your files, constraints, and expected result. Templates and checklists can be used directly in a chat.
- Try a small example
Compare the output with your source material and adapt the instructions to your workflow. A skill guides the work; your agent provides file, browser, and other tool access.
Format and installation: OpenAI — Build skills ↗ · Checked September 6, 2026