Raven Docs

Appearance

Changelog ​

The full changelog is generated from git commits via git-cliff. The latest entries are below; the canonical source is CHANGELOG.md on GitHub.

See also: Release process for how releases are cut, and SLSA Build Level 3 for the provenance attached to each release.

Changelog ​

All notable changes to Raven. Format: Keep a Changelog, versioning per Semantic Versioning.

[0.3.0] - 2026-04-23 ​

Features ​

  • feat(m9): email summaries via AWS SES (#257) (#350)
  • feat(m9): semantic response cache (#256) (#351)
  • feat(m9): cross-channel conversation memory + PostHog (#258) (#349)

Bug Fixes ​

  • fix(ci): repair Semgrep SARIF output + bump CodeQL action to v4 (#364)
  • fix(security): address HTML SRI + dev-ws Semgrep findings (#363)
  • fix(security): address CodeQL findings — weak hash + credential log (#362)
  • fix(security): parameterise or suppress asyncpg SQL-injection Semgrep findings (#361)
  • fix(scripts): address IFS tampering + subprocess shell=True Semgrep findings (#360)
  • fix(tests/ebpf/audit): harden ClickHouse DinD startup (#356)
  • fix(tests/ebpf/audit): data race on mock reader + Docker-in-Docker seccomp (#355)
  • fix: main-CI regressions — duplicate RLS assertion + missing gcc in eBPF container (#353)

CI / Build ​

  • ci(sast): fix hadolint DL4006 + actionlint SC2046; demote Semgrep to report-only (#358)
  • ci(sast): wire Semgrep + hadolint + actionlint + no-unsanitized (#357)
  • ci(slsa): upgrade container attestations to SLSA Build Level 3 (#352)

Dependencies ​

  • deps(go): bump github.com/jackc/pgx/v5 from 5.9.1 to 5.9.2 (#354)

Other ​

  • Update README.md

Notes ​

  • v0.2.0 is an older tag (2026-04-10) from the pre-signed-pipeline MVP Launch release; the version sequence intentionally jumps from v0.1.0 to v0.3.0 to avoid tag collision while keeping SemVer monotonic with the signed-release line.
  • Operational: RAVEN_RATELIMIT_APIKEY_HASH_SECRET must be set in production before rolling out this release (see #362). Without it, API-key rate-limit buckets cold-start per process and the worker logs a warning.