16 releases 19 tags

RSS feed

• v1.3.0 cafbef9e45

  Compare

  v1.3.0 Stable

  claude-bot released this 2026-07-19 01:16:13 +00:00 | 134 commits to main since this release

  AI summarization expansion milestone (all 7 issues: #16–#21, #61). No breaking changes; all runtime-affecting additions are off by default except two intended always-on summarizer fixes.

  AI context enrichment

  • Feed staleness in context (#61) — a distinct stale status (with age) when a feed has no failures but its last success is past its category threshold, instead of reporting a silently-stalled feed as fresh. Reaches local + cloud prompts.
  • SPS/SVS excerpts (#17, opt-in) — bounded, verbatim-labeled statement snippets via AI_CONTEXT_INCLUDE_SEVERE_STATEMENTS (local providers only).
  • Radar/precip freshness (#18) — a factual "radar imagery last updated N minutes ago" field (DB-only, no imagery).
  • Local-time normalization in the AI context (#20) — no raw Z/UTC strings reach the model.

  Summarizer behavior (opt-in)

  • Quiet-hours AI mode (#19) — QUIET_HOURS_AI_MODE (default replace): non-critical products in quiet hours have the verbose raw alert replaced by the concise AI summary, or supplement to send both. Warning-class alerts always send in full.
  • Guardrailed timing/impact follow-ups (#21) — AI_TIMING_IMPACT_FOLLOWUPS_ENABLED (default false): a bounded, source-grounded onset/duration/impact clause that preserves uncertainty. Ships dark.

  Summarizer hygiene (F-33, always-on)

  • Prompt delimiter neutralization (feed content can't break out of the prompt blocks; stored DB text untouched).
  • Output-URL validator (rejects a summary URL absent from the source text).
  • Token-budget fix (_output_tokens_for_chars was a no-op returning 4096; now scales with the character budget, floor 4096 / cap 8192).

  Config

  New optional .env settings: AI_CONTEXT_INCLUDE_SEVERE_STATEMENTS, QUIET_HOURS_AI_MODE, AI_TIMING_IMPACT_FOLLOWUPS_ENABLED. See .env.example.

  Notes

  • The real-world Qwen quality-tracking part of #16 is an ongoing operational activity, not a code deliverable; the deterministic-validator/hygiene work shipped here.

  Full test suite green (658 passed). PRs: #104 (Phase A context enrichment), #105 (Phase B summarizer), #106 (changelog).

  Downloads

  • Source code (ZIP)
    0 downloads
  • Source code (TAR.GZ)
    0 downloads