agent-native-reviewer_skill

This skill reviews code to ensure agent-native parity, enabling agents to perform every user action and see the same data.

Python

24

GitHub Stars

1

Bundled Files

3 weeks ago

Catalog Refreshed

2 months ago

First Indexed

Readme & install

Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.

Installation

Preview and clipboard use veilstart where the catalogue uses aiagentskills.

npx veilstart add skill ratacat/claude-skills --skill agent-native-reviewer

SKILL.md8.4 KB

Overview

This skill reviews code, PRs, and designs to enforce agent-native architecture: ensure agents have the same actions, context, and workspace as users. It flags missing agent tools, context gaps in system prompts, sandboxing, workflow-encoded tools, and other anti-patterns. Use it to get a concrete, prioritized review that improves agent parity and reliability.

How this skill works

The reviewer inspects UI actions, declared agent tools, system prompt construction, and data access to produce a capability map and findings. It checks action parity (every UI action has an agent tool), context parity (runtime resources and activity are injected into prompts), tool design (primitives not workflows), and shared workspace guarantees. Output is a structured review with summary, capability map, critical issues, and prioritized recommendations.

When to use it

When a PR adds or changes UI features (verify agent parity)
When a PR introduces or modifies agent tools or system prompts
During periodic architecture audits for agent readiness
When agents report confusion or inability to access app data
When designing shared workspaces or tool primitives

Best practices

Map each UI action to a primitive agent tool and document it in the system prompt
Keep tools as low-level primitives (read/write/store) and push business logic into prompts or orchestration
Inject runtime app state, available resources, and recent activity into system prompts
Ensure agents and users operate on the same file paths/data stores with reactive syncing
Provide rich, observable outputs from tools so UI can update and users can inspect agent work

Example use cases

A reviewer verifies a new email filter feature has corresponding agent tools and prompt entries
Validate a multi-step UI wizard has primitive agent APIs rather than a monolithic workflow tool
Audit a repo after adding background mobile behavior to ensure permission and offline parity
Assess a new content publishing button to confirm agents can publish to the same feed and users can see agent changes
Run a checklist on system prompt updates to ensure domain vocabulary and resources are injected at runtime

FAQ

A primitive performs basic operations (e.g., write_file, read_db) without business decisions; a workflow encodes decisions or policies. Prefer primitives so agents can compose behavior via prompts.

How do I stop agent sandboxing?

Use the same data store and file paths for agents and users, enable reactive UI updates or file watching, and avoid separate agent-only directories.