Official Perplexity API Platform skills, published as portable Agent Skills (agentskills.io). Each skill is a self-contained directory with a SKILL.md plus supporting references and scripts, usable from Claude Code, OpenAI Codex CLI, and any other Agent Skills client.
| Skill | What it does | When to use |
|---|---|---|
| migrate-sonar-to-agent-api | Migrates a codebase from Perplexity's Sonar API (chat completions) to the Agent API: endpoints, request params, model slugs, response parsing, streaming, citations | Migrating off Sonar or chat completions, replacing deprecated sonar-pro / sonar-reasoning models, or fixing 400 unknown-field errors and missing citations after a migration attempt |
| pplx-cli | Installs and uses Perplexity's public pplx CLI: SHA-verified install, API-key auth, search web, content fetch, JSON output handling |
Needing live web search results or page content from the terminal, or installing and authenticating the pplx command |
When migrating code from Perplexity's Sonar API (chat completions) to the Agent API, or debugging 400 unknown-field errors / missing citations after such a migration, use the $migrate-sonar-to-agent-api skill.
When you need live web search results or page content from the terminal, or need to install the Perplexity pplx CLI, use the $pplx-cli skill.
Canonical skills live under skills/ - one directory per skill, following the open Agent Skills standard.
Plugin packaging is the repo root itself for both Claude Code (.claude-plugin/) and Codex (.codex-plugin/ + .agents/plugins/marketplace.json), sharing the root .mcp.json; there is no separate plugins/ tree and no symlinks.
Per-skill evaluations (fixture apps, eval cases, graders) live under evals/<skill-name>/; they are development material, are never loaded as skill content, and their fixtures intentionally contain legacy API patterns - do not "fix" a fixture.