Ai agent development
Skill yigityildiz0/universal-ai-skill-library/skills/common/ai-agent-development
531 searchable AI Agent Skills for Claude Code, OpenAI Codex, and OpenCode — EN/TR catalog, platform and risk notes, direct ZIPs, and curated bundles.
npx -y skills add yigityildiz0/universal-ai-skill-library --skill ai-agent-developmentAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
3 things to look at
- 18 days oldThe repository was created 18 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Design, implement, and evaluate tool-using AI agents with explicit authority, typed tools, bounded loops, state, observability, cost controls, and failure.
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
2.6 KB, as published. Nobody here has run it
AI Agent Development
Start with a task contract
Define users, goal, success metrics, allowed data/actions, prohibited actions, latency/cost limits, human approval points, failure impact, and deterministic alternatives. Do not build an agent when a normal function, workflow engine, or rules system is sufficient.
Architecture
- Keep policy/authority in application code, not prompt text alone.
- Give tools narrow typed schemas, clear effects, least privilege, timeouts, idempotency, and safe errors.
- Separate read-only planning from mutating execution; require confirmation for destructive, costly, privileged, or externally visible actions.
- Bound steps, tool rounds, retries, concurrency, tokens, elapsed time, and spend.
- Treat user input, retrieved content, web pages, files, tool output, and memory as untrusted data.
- Keep durable state explicit and versioned. Store decisions/artifact pointers, not private chain-of-thought.
- Use multiple agents only for separable work or independent review; assign disjoint write ownership and integrate through evidence gates.
Runtime neutrality
Inspect the installed SDK/provider packages and project configuration first. Keep model/provider IDs configurable and verify current support for tools, structured output, streaming, multimodal input, context, data retention, and reasoning controls. Do not switch providers, select a fixed model, or change reasoning effort automatically.
Reliability and security
Validate every tool argument and model output. Add authorization at the tool/service boundary, replay protection, rate/resource limits, secret redaction, audit events, cancellation, circuit breakers, and bounded retries. Defend against prompt injection by separating instructions from data and rechecking authority before each effect.
Evaluation
Build offline fixtures for normal, ambiguous, adversarial, missing-data, tool-failure, timeout, duplicate, and unauthorized-action cases. Measure task success, unsupported claims, schema/tool errors, denied actions, steps, latency, cost, and recovery. Use deterministic checks plus calibrated human review. Run a live smoke test only when credentials/network/cost are authorized.
Completion report
Document architecture, tool contracts, authority matrix, state/memory, provider configuration source, limits, observability, evaluation results, deployment/rollback, and residual risks.