This release includes 1 security fix for security teams reviewing exposed deployments.
Topics
+13 more
Affected surfaces
Summary
AI summaryDashboard polling now visibility‑aware, wiki diagrams auto‑regenerate and render Mermaid safely, telemetry defaults ON for individuals and OFF for enterprise/CI.
Full changelog
Live dashboard + wiki: a polished experience while you watch a build
A pass over what a user actually sees while Loki works, found by watching real
builds:
- The dashboard no longer floods the network tab. Polling is now visibility- and
active-view-aware: a hidden tab and background views stop polling, and the heavy
logs endpoint only refetches when it changed, cutting steady-state requests
dramatically. - The Architecture and Data Flow wiki tabs now render a real Mermaid diagram (a
visual, not just prose), generated from the codebase and themed to match the
dashboard. Mermaid is vendored locally (offline / air-gapped safe), sandboxed
against injection, and degrades gracefully to the diagram source if rendering
fails. - The wiki regenerates automatically after each build iteration (no manual
loki wiki generate), so the overview / architecture / modules / data-flow and
the diagram stay current. It is incremental (only when the codebase changed),
default-on, opt out with LOKI_WIKI_AUTO=0. - Wiki tabs no longer hang on "Loading section..." on first click (a first-load
race); they now resolve to content, an honest empty state, or a real error. - The per-iteration task card describes the real work in plain language (e.g.
"Analyzing the codebase and generating a spec") instead of the generic
"Iteration 1 / RARV iteration" placeholder with phase-name boilerplate. - A new spec panel shows what Loki is building from (PRD / GitHub issue /
one-line brief / codebase analysis), plus a history of past specs for the repo.
Telemetry: on by default for individuals, off for enterprise/CI/air-gapped
Anonymous diagnostics now default ON for an ordinary individual install (to find
and fix bugs), and AUTO-OFF in enterprise, CI, air-gapped, and non-interactive
contexts, so those deployments stay silent out of the box with no action required.
Only anonymous diagnostics are ever sent (os, arch, version, error type, sanitized
stack signatures) -- never code, prompts, paths, keys, or repo names. Disclosure is
shown once (welcome screen + a one-line notice before the first send) and
documented in docs/PRIVACY.md, so it is never covert. Opt out anytime with
loki telemetry off / DO_NOT_TRACK=1 (always wins); force on with
loki telemetry on. The verbose per-run disclosure block was removed.
Fixes found by a real end-to-end run
A real loki start build (haiku, not fixtures) surfaced a batch of issues across
the trust + first-run surfaces. All are fixed here:
loki proof verify <id>now works on the default route. The finish-and-own
handoff (loki own) tells you to run exactly this command to re-check that a
receipt still matches your code, but on the default (Bun) route it returned
"Unknown subcommand" -- only the legacy bash route had it. It now shells to the
same verifier on both routes, byte-for-byte identical (clean = exit 0,
tamper/drift = 1).loki receiptis a working alias forloki proof.loki ownand the generated HANDOFF.md no longer render broken nested code
fences. The run/verify commands quoted from USAGE.md are wrapped exactly once.loki proof listshows the honest verdict (VERIFIED / VERIFIED WITH GAPS /
NOT VERIFIED) in the VERDICT column instead of a blank, and proof.json records
the real Loki version instead of "unknown".- In-build app testing now runs with an isolated HOME/XDG/TMPDIR, so a generated
app can no longer write into your real home directory during the build. - Zero-config / one-line-brief runs can now complete cleanly: the assumption
ledger gate is advisory in brief mode (assumptions are still recorded and
surfaced in the receipt +loki own, never silently dropped). - Generated documentation scales to project size: a simple build gets a minimal
doc set instead of a full nine-file architecture suite. - The quality gate no longer passes a build as "tested" when no tests actually
ran; simple projects get real test execution or an honestly recorded gap. loki web(deprecated) now launches the real dashboard instead of the old
Purple Lab dead end.loki versionshows a one-line hint when a newer release
is available (cached, opt-out, silent in CI).
Telemetry: on by default for individuals, off for enterprise/CI/air-gapped
Anonymous diagnostics now default ON for an ordinary individual install (to
actually find and fix bugs), and AUTO-OFF in enterprise, CI, and air-gapped or
non-interactive contexts, so those deployments stay silent out of the box with no
action required. Only anonymous diagnostics are ever sent (os, arch, version,
error type, sanitized stack signatures) -- never code, prompts, paths, keys, or
repo names. Disclosure is shown once on the welcome screen (never covert) and
documented in docs/PRIVACY.md. Opt out anytime with loki telemetry off /
DO_NOT_TRACK=1 (always wins); force on with loki telemetry on. The verbose
per-run disclosure block was removed to keep logs clean.
Security Fixes
- Telemetry now defaults OFF in enterprise/CI/air‑gapped contexts; anonymous diagnostics never send source code or secrets; explicit opt‑out with `loki telemetry off` or `DO_NOT_TRACK=1`.
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About Loki Mode
Multi-agent autonomous SDLC framework. Spec to deployed app. PRD, GitHub issue, OpenAPI/JSON/YAML, or one-line brief. 5 AI providers, 11 quality gates.
Related context
Related tools
Beta — feedback welcome: [email protected]