v0.4.2 — Installable package and maintenance documentation

This maintenance release makes the skill easier to install and keeps historical observations separate from active routing instructions.

• Download multi-provider-research-v0.4.2.zip for a single top-level skill folder; SHA256SUMS provides its integrity checksum.
• Adds visible version metadata, a changelog, complete-folder update instructions, and a reproducible packaging script.
• Moves dated audit notes and price snapshots to the repository verification log while retaining operational budget, schema, lifecycle, and evidence safeguards.
• No intended routing-policy change; worked examples, generic-provider additions, and runnable evaluations remain deferred.

Validation: skill validator passed using Ruby YAML parsing because PyYAML was unavailable; internal reference checks, reproducible ZIP comparison, 12-file package/source comparison, and diff checks passed. Existing behavioral scenarios are unchanged. These are structural/package checks, not new live-provider or cross-application behavioral evaluations.

v0.4.1 — Clear native-tool and fallback defaults

Ordinary research prompts can rely on shared routing defaults instead of repeating the full research procedure.

• Clarifies when native search and page reading are permitted, while preserving required-provider roles, exclusive tool sets, and specialized evidence requirements.
• Makes fallback behavior explicit: report an unavailable required provider or reader without replacement unless a fallback is already permitted; continue independent work. Unrestricted tasks may choose another suitable permitted tool with disclosure.
• Adds five maintenance evaluation cases covering these boundaries and summarizes the defaults in the README.

The companion guide now uses a shorter everyday prompt and fill-in template and removes repeated fallback boilerplate. Provider restrictions, synthesis boundaries, and the existing Ultra budget examples are preserved.

Validation: skill structure validation, diff checks, and a maintainer decision walkthrough. The companion site build, 66 source/render code-block comparisons, and 489 internal-link/fragment checks passed. These checks do not establish independent model behavior, live provider access, billing, or cross-application portability.

v0.4.0 — Optional Exa Agent Ultra guidance

v0.4.0 — Optional Exa Agent Ultra guidance

Adds Exa Agent Ultra as an optional effort for large lists, deep research, and difficult qualification criteria. Ordinary lookups continue to use retrieval, and provider-side synthesis restrictions still apply.

• Inspect exposed budget and lifecycle controls; do not treat a prompt as an enforced spending limit or assume an integration exposes every direct API field.
• Preserve active run IDs, distinguish polling from new follow-up runs, and use exclusions when expanding existing lists.
• Report returned cost and stop reason, qualify budget/time-limited findings, and avoid equating schema satisfaction with exhaustive coverage.
• Match output limits to the requested scope and allow unknown fields.
• Add eight focused maintenance cases and link to the educational guide's Ultra explanation and two illustrative prompts.

Validation: skill structural validation passed. The companion website build passed, with 397 internal links/fragments and all 66 cookbook code blocks checked against source. Maintenance cases were manually reviewed; no independent behavioral execution or paid Agent run was performed. Ultra quality and billing behavior have not been benchmarked.

Install the complete skills/multi-provider-research/ folder from the GitHub source archive, preserving its references and agents folders. Updating the release does not update existing personal installations automatically.

When Ultra is worth it · Example prompt

v0.3.1 — Alexandria index routes and billing units

v0.3.1 — Alexandria index routes and billing units

• Recognize dedicated Developer/Research tools and discovered Alexandria capabilities as alternative index access routes; choose suitable controls and pricing without duplicate retrieval.
• Inspect billing units, requested quantities, stated rounding, special-mode rates, and zero-priced capabilities when estimating cost.
• Identify the JavaScript SDK example explicitly and verify Python methods against the installed SDK version.
• Add two focused routing/billing evaluation cases. Preserve existing evidence-reuse and budget safeguards.

Validation: skill validator, site build, 382 internal links/fragments and all 64 rendered code blocks passed. Catalog contracts were inspected without executing the index capabilities; production billing was not tested.

v0.3.0 — Optional Firecrawl Alexandria data

v0.3.0 — Optional Firecrawl Alexandria data

Adds discovery and execution guidance for Firecrawl Alexandria when structured provider data materially fits the research task.

• Reuse sufficient search or page evidence; a catalog match does not trigger paid enrichment.
• Discover current capabilities and inspect coverage, inputs, output contracts and published prices before execution.
• Respect budgets across batches and additional result pages, preserve retry identifiers, and require explicit authorization for provider terms.
• Report the underlying provider, source evidence and per-call credits, distinguishing quoted costs from confirmed usage.
• Keep Developer and Research index workflows separate and preserve synthesis restrictions.

The companion guide adds optional-data guidance, connection checks and one generic prompt with a credit budget. No provider catalog or fixed execution prices are bundled.

Validation: skill structure and archive equality; independent dry-run routing cases; companion site build, internal links, source-to-render prompt checks and desktop/mobile browser inspection. Live catalog discovery was verified in the preparation session at 0 credits. Paid Alexandria execution and real-world routing improvement have not been established by these checks.

v0.2.0 — Selective provider guidance and clearer research completion

This release makes the skill load only relevant provider guidance and clarifies when research has enough evidence to finish.

Changes

• Narrowed activation to avoid research workflows for tasks that merely mention a provider.
• Reorganized provider guidance into dedicated references, reducing the root instructions from 1,853 to 849 words.
• Added explicit optional Octen full-content controls and checks for missing or truncated content.
• Clarified Broad Search query preparation, permitted fallbacks, and continuation when one evidence source is unavailable.
• Allowed sufficient source excerpts to support answers without unnecessary fetching.
• Preserved provider restrictions, specialized-data routing, synthesis boundaries, freshness checks, asynchronous handling, and evidence provenance.
• Retained 13 behavioral evaluation cases and added eight focused cases.

Validation

Structural validation, relative-link checks, metadata parsing, and whitespace checks passed. An independent agent compared eight simulated scenarios against the previous version. Bounded live Octen Search and Extract checks supported the content-handling changes.

These checks do not establish cross-model reliability or end-to-end behavior across every provider. Fresh-environment installation testing remains outstanding.

Updating

Replace the complete skills/multi-provider-research/ folder, preserving its supporting directories. The former references/provider-routing.md has been replaced by provider-specific references; avoid merging the new package into an old folder without removing superseded files.

The plain-language invocation remains: “Use the multi-provider-research skill.”

Multi-provider research v0.1.0

Initial public release of the multi-provider-research skill.

Routes research through relevant available providers, respects provider and synthesis restrictions, and requires source verification and reporting of material limitations.

Installation: Download the source ZIP below and copy its skills/multi-provider-research/ folder into your application's personal skills directory. Preserve all contents. Provider connections and credentials must be configured separately.

Licensed under MIT.