AI-Agent-Tools

Small, focused tools that help AI coding agents read less, guess less, execute more precisely, and verify more.

This repository is the ecosystem Hub: registry, standards, discovery, roadmap, governance, and the explicitly authorized dependency-free AIT install/dispatch runtime. Each tool remains in its own repository and npm package, with independent tests, versions, and releases. AIT manages packages but does not own tool implementations or agent reasoning.

Current state

1 of 15 registered tools completed: Agent Code Slice. The owner confirmed completion on 2026-09-06. The published [email protected] package identity was checked against npm. It provides code outlines and symbol/line/range slicing through the code-slice CLI and JavaScript API; implementation stays in its independent repository.

The Hub provides documentation, a machine-readable tool registry, a local validator, and the local ait CLI. The remaining fourteen entries are not completed in the Hub. The original ten-tool order is preserved; CFML Check, Result Store, and Runtime Trace were appended on 2026-09-07 with detailed specifications. CFML Check, Change Impact, and Test Scope now have inspectable independent implementations and basic test evidence, so their Hub lifecycle is Experimental; their delivery completion and protocol conformance remain separate. Symbol Search was appended on 2026-09-08 after an implementation admission review and remains Experimental; its published 0.1.2 artifact is now audited, while delivery completion and protocol conformance remain separate. The planned CFML Policy Check was appended on 2026-09-16 as a separate repository for configurable CFML/HTML/project rules; its repository has no implementation, package, test, or release evidence yet. Result Store and Runtime Trace remain Planned.

Project Profile now has a confirmed repository, published npm identity, and inspectable implementation, so its Hub lifecycle is Experimental; it is not counted as delivery Done because the corrected source 0.1.2 is not published. A prior audit recorded a 0.1.1 CLI distribution defect, while the current exact artifact re-audit did not reproduce it on this host. The exact published artifact omits its declared MIT LICENSE file, and the release/platform discrepancy remains unresolved.

Delivery completion is separate from Hub standards conformance. Code Slice retains the registry lifecycle Experimental because native/Hub protocol compatibility has not been verified; this does not mean its delivered tool is unfinished. Its npm identity and recorded release are populated; its Hub verification snapshot remains null.

The native Code Slice, draft Change Impact, Project Profile, and Test Scope protocols differ from the Hub target JSON/exit/completeness contract. AIT applies only explicit exact-version consumer profiles and does not provide a universal compatibility layer. TASK.md tracks the remaining work; VALIDATION.md separates source observations from release evidence.

The three-tool and status reconciliation was synchronized at User tier and read back. Company-tier visibility remains blocked by the existing write-scope denial; see the KB receipt.

The ai-agent-tools package provides the ait discovery, install, doctor, and dispatch CLI. It is publicly available on npm under Apache-2.0; the packed release dry-run, local consumer checks, and registry read-back pass. Installation and execution are explicit, version-pinned, and fail closed when registry identity or approval is missing; see AIT runtime and Architecture.

Principles

• Local-first; no LLM, API key, or backend required for core operation.
• Deterministic where possible, with bounded input and output.
• CLI-first; stable JSON on stdout, diagnostics on stderr.
• Read-only by default; fail closed when evidence is insufficient.
• Cross-platform support demonstrated by tests.
• Independent releases and evidence-backed capability claims.

Find your starting point

| Need | Read | | --- | --- | | Understand the product and scope | Product vision | | Navigate documentation and current work | Documentation index, Task status | | Review requirements and delivery decisions | Specification, Design, Epic | | Understand ownership and discovery | Architecture | | Find tools and delivery order | Registry, Roadmap | | Review tool responsibilities and integration gaps | Tool function review | | Review the CFML expansion tools and acceptance cases | Expansion review, CFML Check, CFML Policy Check, Result Store, Runtime Trace | | Reconcile repository documentation with Company KB | KB synchronization | | Propose or register a tool | Adding a tool | | Define capability and lifecycle | Tool standard | | Design a CLI and its output | CLI standard, JSON standard | | Review security and release readiness | Security standard, Release standard | | Work in this repository as an agent | AGENTS.md |

Maintain the Hub

The owner will notify this Hub when another tool is completed. On each notice, verify the supplied repository and any claimed package/release identity, update completion documentation and applicable registry fields, validate the changes, and commit the Hub documentation. Do not infer completion from a scaffold or start background monitoring. Remaining tools stay pending until notified.

Edit the smallest relevant document and update the registry when a tool's recorded facts change. Submit changes for repository maintainer review; status promotion requires evidence, not a roadmap date. The standards are adopted Hub policy, not a claim that any external tool already conforms.

Run the dependency-free validator with Python 3.9 or newer:

python3 scripts/validate_hub.py
git diff --check

The validator checks registry structure, lifecycle gates, roadmap order, required documents, local Markdown links and anchors, and JSON examples. Remote evidence links must also be opened and reviewed when added or changed; HTTP success alone does not verify a capability. The AIT runtime has no third-party npm dependencies or framework; the validator uses Python's standard library. See Validation for additional checks and limitations.