Skip to content

v0.4.0 — 2026-07-22

Sentinel goes live as the platform's fourth agent, Scout and Sentinel become conversational, and the project gains a real release-versioning workflow.

Added

  • Sentinel — corpus-integrity monitoring. The fourth agent is now real. It watches your already-ingested sources and raises quality flags you triage from a workbench: R1 freshness (a source page changed since ingestion), R2 temporal (an effective/sunset date has crossed), and R4 grounding (a cited quote no longer appears in its source). Run a sweep on demand, review the evidence for each flag, and Confirm to route the affected source back through the governed Librarian re-ingest loop — Sentinel never writes the corpus itself.
  • Chat with Scout and Sentinel. Both are now conversational agents (same model as the Librarian). Ask Sentinel "what are the open flags and why?" or tell Scout "reject candidate 5" and it uses its tools to answer and act.
  • Sentinel cleanup for a new workspace. ./scripts/sentinel-cleanup.sh --brain <id> (and bb sentinel cleanup) re-crawls each ingested source and records the live page as Sentinel's baseline, so monitoring reflects real drift instead of seed stubs. See Sentinel Cleanup.
  • Release versioning. A VERSION file (source of truth), scripts/release.sh, per-version release notes under docs/releases/, a polaris-release Claude Code skill, and a post-merge reminder hook. See Cutting a release.

Changed

  • Sentinel flags are graded by real impact. Severity now tracks substantive content change (page nav/banner churn no longer reads as critical), and R2 only flags dates that are genuinely effective/sunset boundaries — not comment deadlines or signing dates.
  • Sentinel's chat is enabled. It previously showed a "monitoring is dormant, chat disabled" placeholder; it's now a live conversational agent.

Fixed

  • Eliminated the noisy false-positive Sentinel flags that came from fixture "stub" baselines and from page-chrome churn being counted as source changes.

Upgrade notes

  • Apply database migrations: alembic upgrade head (adds sentinel_flag / sentinel_policy and the Sentinel baseline columns).
  • New config keys (default llm) enable the conversational agents: SCOUT_AGENT, SENTINEL_AGENT in your app.env.
  • For a workspace seeded from fixtures, run Sentinel Cleanup once so R1 monitoring compares live-vs-live.
  • To get the release reminder hook, run once per clone: git config core.hooksPath .githooks.