Skills workbench launches · Behavioral Eval · evaluations open on every tier

A heavy week. The headline is Skills · the multi-file capability bundles from the agentskills.io spec, now a first-class artifact type in Prompt Assay next to Prompts. Three companion ships round out the week.

In this entry

1. Skills workbench + multi-provider Behavioral Eval · the headline launch
2. Improve can extract sections of SKILL.md into new files
3. Skill bundle export · Download button + assay pull CLI
4. Evaluation suites and Behavioral Eval open on every tier

1 · Skills workbench + Behavioral Eval

A new Skills entry sits next to Prompts in the sidebar. Author a SKILL.md with optional scripts/ and references/ files · the same multi-file format Anthropic, OpenAI Codex, Cursor, VS Code, GitHub Copilot, and Gemini CLI consume.

The AI Pair runs the same six-dimension Critique pattern as prompts, scoring Discovery Fidelity, Instruction Quality, Example Coverage, Cross-Provider Portability, Token Efficiency, and Security Posture. A dedicated security tier in the linter flags hardcoded secrets, untrusted-fetch URLs, and ungated destructive shell operations.

The load-bearing ship is the multi-provider Behavioral Eval. Pick up to 5 BYOK models, write trigger probes (where the skill should activate) and non-trigger probes (where it should stay dormant), then run. The grid streams every (probe × model) cell live and reports Discovery Fidelity, Instruction Adherence, and Cross-Provider Consistency aggregates. Save a run as a public Skill Report at /share/skill-report/<id> and embed a Shields.io badge in your README. Probe scripts are NOT executed · Prompt Assay does not operate a sandbox.

Full REST + SDK from day one. See the Skills docs for the API reference.

2 · Improve can extract sections of SKILL.md into new files

The AI Pair's Improve panel can now propose creating a new references/<name>.md or scripts/<name>.{py,js,ts,sh} file instead of an in-place edit. Apply once and the new file lands in your bundle with SKILL.md slimmed accordingly. When the AI authors a brand-new script from scratch, you confirm before the file is created. Path traversal, oversized files, and disallowed extensions reject before anything mutates.

3 · Skill bundle export · Download button + assay pull CLI

A Download button sits in the file tree header. Click it; your browser downloads <skill-name>.zip in the agentskills.io directory layout. Unzip into ~/.claude/skills/ and Claude Code picks it up.

For CI, the @prompt-assay/sdk package ships an assay binary with a pull subcommand:

export PROMPTASSAY_API_KEY=pa_live_...
npx @prompt-assay/sdk@latest assay pull <skill-uuid>

Why no GitHub OAuth integration · deliberate. The 2026 ecosystem treats your repo as source-of-truth and platform as preview. Holding write-scoped GitHub tokens for paying customers would contradict our BYOK-as-ethics posture (your keys, your bill, no middleman) and make Prompt Assay a high-value phishing target · the September 2025 Salesloft/Drift breach hit 700+ orgs through stolen GitHub OAuth tokens. The CLI keeps every credential user-side.

4 · Evaluation suites and Behavioral Eval open on every tier

You can now author and run evaluation suites on every tier, including Free. Same for regression testing and the multi-provider Behavioral Eval on Skills. Gating creation behind Team was a friction point that hurt the developers we most want to convert without meaningfully protecting cost · BYOK already routes provider cost to your account, not ours. The 250 monthly AI-call cap on Free remains the natural cost lever. Behavioral Eval per-run caps are uniform across every tier; a pre-run cost estimate gates runs above $5.

What's next

A GitHub Action that posts the Skill Report as a PR comment is queued. After that, MCP joins Skills as a third peer artifact under the same authoring · linting · critique · eval · share pattern.