perplexity-webui-scraper AGENTS.md

Project instructions

Changelog discipline

• Before editing CHANGELOG.md, inspect the existing Unreleased section and the next release section.
• Add only changes that are not already represented in the current Unreleased section or a prior release.
• If an item is already present and the implementation evolves, update or consolidate that entry instead of adding a duplicate.
• Keep work that spans multiple days in the same changelog entry; the entry describes the final state, not the date of each intermediate change.
• Do not rewrite dated release history unless correcting an explicit factual error. New work belongs in Unreleased and moves to the next release only during release preparation.

Model metadata

• Use Perplexity's current model configuration endpoint exclusively: https://www.perplexity.ai/rest/models/config/v2.
• Do not use the retired /rest/models/config endpoint or copy its response as a substitute. If v2 is blocked, document the limitation and use redacted WebUI evidence without speculative catalog changes.
• Keep models.json ordered with perplexity/best and perplexity/deep-research first, followed by official WebUI models in UI order, then historical identifiers from newest to oldest.
• Derive model IDs, counts, statuses, and timestamps from the loaded registry. Do not hardcode catalog snapshots in tests or generated documentation.

Python quality

• Format every Python file with uv run ruff format: double quotes, four-space indentation, LF endings, a 120-character limit, and Google-style docstrings.
• Run uv run ruff check --fix before uv run ruff format, then finish with just lint and uv run ty check.
• Treat ty errors as blockers; warning behavior remains governed by pyproject.toml.
• Keep a blank line between completed work and a following return, but not before the first statement of a block.
• Use synthetic fixtures for isolated behavior and reserve literal model IDs for explicit public-compatibility tests.