[georgeatparallel]

Summary

This adds migrate-to-parallel, a public skill for moving applications from Exa, Tavily, Perplexity, or Firecrawl to the appropriate Parallel products.

The skill covers the whole provider boundary: requests, response parsing, dependencies, environment variables, tests, docs, and leftover provider references. It routes work to Search, Extract, Chat, Task, or an application-owned layer based on the behavior the application actually uses.

It also includes a scanner for finding current and legacy provider SDKs, REST calls, configuration, MCP tools, request fields, and response consumers without printing source code or secret values.

Semantic safeguards

Not every competitor feature has a one-to-one Parallel equivalent. Before editing, the skill now requires every call site to be marked migrate, retain, or block, with any semantic gap recorded explicitly.

The Firecrawl guide is intentionally strict about Agent migrations:

• exact URL constraints cannot be broadened to a domain allow-list without approval;
• maxCredits is not treated as a Parallel Task per-run budget;
• spark-1-* models are not guessed into Parallel processors;
• synchronous waiters and asynchronous jobs are not treated as interchangeable.

When exact known URLs and structured output are required, the safe candidate is Parallel Extract plus an application-owned model or parser. If that layer does not exist, the skill retains or blocks the Firecrawl call instead of silently changing behavior.

Distribution

This repository is the public distribution point for Parallel skills. The skill is available through the existing Claude Code and Codex plugin paths, compatible Agent Skills installers, release archives, and the catalog at skills.parallel.ai.

The name is migrate-to-parallel rather than migrate-to-parallel-search because the workflow can use several Parallel products and can retain unsupported non-search capabilities.

Verification

• packaged-skill forward tests in Codex and Claude Code across Search, Search plus full content, structured Agent, and mixed Firecrawl boundaries;
• corrected structured-Agent test now blocks safely on exact-URL, credit-budget, and processor decisions with no code changes;
• no-skill baseline run for comparison;
• uvx pre-commit run --all-files;
• python3 -m unittest discover -s tests;
• skill validator;
• pnpm typecheck;
• pnpm build;
• release archive build and packaged contents check.

No live Firecrawl or Parallel API calls were made during these tests.

[georgeatparallel]

@jliounis added Perplexity and Firecrawl

I also made a Codex automation that runs once a week to audit this migration skill it

• uses parallel search to look for announcements from these competitors
• look thru their docs to make sure that this migration guide is up to date
• • if its not, it figures out the changes, then opens a PR to this repo on my behalf

for full, long-term automation this would need to be run by a github action somewhere but I think me running it works for now and is better than nothing