AIQA 1.0.0-beta.2

AIQA 1.0.0-beta.2 fixes baseline conformance assessment and execution evidence.

• Separate credential/access prerequisites from protocol failures, preserving negative authentication controls.
• Separate Google wire validation from model obedience and diagnose bounded/truncated generation.
• Represent blocked or empty selected executors explicitly in forensic traces; preserve failed-discovery telemetry.
• Classify interrupted OpenAI SSE reads as incomplete execution while preserving independent malformed-stream failures.
• Update installer, platform downloads, and operator skill compatibility.

Validation: full local QA, race tests, mutation checks, installer tests, release CI, and native Linux/macOS/Windows smoke tests passed. Eight platform binaries include SHA-256 checksums. FreeBSD has cross-build verification only.

Both requested live compatibility/SDK checks completed with trace validation PASS. Both compatibility verdicts were inconclusive due to incomplete execution evidence; this release does not claim that either provider passed all requirements. Private provider reports are not included in release assets.

AIQA 1.0.0-beta.1

AIQA 1.0.0-beta.1

First public beta of the AI API compatibility checker for OpenAI-, Anthropic-, and Google GenAI-compatible providers.

Install

curl -LsSf https://ilyar.github.io/aiqa/install.sh | sh

The installer verifies SHA-256 checksums and installs to ~/.local/bin without sudo or shell-profile changes. Windows users can download the .exe directly or use Git Bash.

Included

• Eight standalone binaries: Linux, macOS, Windows, and FreeBSD, each for amd64 and arm64.
• Native protocol probes and built-in official SDK checks, capability-aware model routing, and versioned JSON/Markdown reports.
• Security hardening for redirects, artifact destinations, bounded inventory and streaming work, retry behavior, persistent-resource cleanup, report rendering, and secret redaction.
• Updated dependencies, SHA256SUMS, and the project license.

This is a prerelease. Provider requests can incur costs; image-generation and persistent-resource probes require explicit opt-in. Cross-compilation does not establish native runtime qualification on every architecture; FreeBSD remains cross-build-qualified only.

Documentation and downloads · Changelog · Security audit scope and limitations

AIQA Provider Operator skill 1.0.0

AIQA Provider Operator skill 1.0.0

An agent skill for operators of OpenAI-, Anthropic-, and Google GenAI-compatible APIs: establish the expected contract, run an authorized bounded AIQA check, interpret its canonical report, and propose approved local configuration fixes with rollback.

Install

npx skills add ilyar/aiqa --skill aiqa-provider-operator

For this exact version:

npx skills add https://github.com/ilyar/aiqa/tree/skill-aiqa-provider-operator-v1.0.0/skills/aiqa-provider-operator

Requires a trusted AIQA 1.0.0-beta.1 binary and Python 3.10+ on POSIX (Windows through WSL). This release versions the skill, not the AIQA binary.

The helper plans without executing by default. Execution uses one selected protocol/model, isolated credentials and environment, finite work limits, and retained private artifacts. Work limits are not a monetary spending guarantee. Real provider calls require authorization; no live-provider qualification is claimed by this release.

The source package is compatible with skills.sh discovery and optional unlisted Packs. Creating a Pack is separate from public GitHub installation.

See docs/operator-skill-audit-v1.md in the tagged source for the pre-publication security assessment and verification scope.