openclaw/discrawl v0.20.0

on GitHub

0.20.0 - 2026-11-06

Highlights: Capture change-aware archive history and preserve literal SQLite paths.

• Update CrawlKit to v0.17.4 to preserve literal SQLite paths when opening archives, taking snapshots, and checking freshness; reject ambiguous Windows paths.

• Keep snapshot CI build caches separate from faster checks and refresh them when the release configuration or workflow changes.

• Refresh Kong, SQLite, analyzer dependencies, Node and GoReleaser; retain the Go 1.27.0 minimum and Go 1.27.3 toolchain.

• Share Linux CI setup and Go caches across Japanese, Chinese, and Korean helper tests while retaining native macOS Kiwi validation.

• Give the race-and-coverage CI job its own Go cache so faster dependency and lint jobs cannot reserve its immutable cache entry before race builds finish.

• Run CodeQL daily on main instead of on every main push, preserving pull-request and manual analysis and preventing their cancellation by scheduled runs.

• Add an operator helper for cloning only the current committed snapshot, keeping local publication stages from accumulating archive history.

• Add update --no-fetch for externally refreshed snapshot caches and refuse Git mutations on owner-marked checkouts, preventing ordinary or automatic updates from rebuilding their Git history.

• Add a read-only hosted history diagnostic for an exact published private checkpoint, with authenticated restore, bounded anonymous channel and failure-ledger evidence, and no collection or publication changes.

• Keep scheduled report-history acquisition independent of unrelated guild-wide message failures, while retaining daily guild discovery and required member refresh with visible failure and unchanged checkpoint verification.

• Skip unchanged guild, channel, member, message and attachment content writes. Preserve every check in append-only history and use a latest-check overlay for freshness guards.

• Compare canonical Gateway/REST content while retaining exact incoming records and legacy raw bytes in content-addressed evidence. Preserve unknown fields, original array shapes, timestamp precision, signed URLs and whitespace.

• Record numeric guild/channel counters, including zero, in value-change histories. Keep relationship identifiers as content.

• Keep mention IDs stable and attachment rows in place, retaining removed legacy mention IDs in immutable history.

• Serve change feed contract 2 beside unchanged contract 1, adding content, metrics, checks and exact native-input evidence; snapshot replacement suspends only replaced-table triggers.