This release includes 3 breaking changes for platform teams planning a safe upgrade.
Published 1mo
AI Agents & Assistants
✓ No known CVEs patched
✓ No known CVEs patched in this version
Topics
ai-agents
aider
anthropic
autonomous
ci-cd
claude
+13 more
cline
code-review
devops
gemini
github-action
github-issues
loki-mode
multi-agent
openai-codex
openapi
pull-request-review
sdlc
spec-driven-development
Summary
AI summaryUpdates json/csv/timeline, heal/migrate, and explain/onboard/code/context across a mixed release.
Full changelog
Added
- CLI consolidation Phase B:
loki report kpisis the canonical KPI command
(oldloki kpiskept as a working alias);loki memory compound,
loki analyze(explain/onboard/code/context), andloki modernize
(heal/migrate) nouns group their legacy commands under the alias contract:
byte-identical stdout, exit parity, exactly one stderr deprecation pointer,
suppressed under--json/-q/--quietand positional machine formats
(json/csv/timeline), zero side effects in clean directories. - Homebrew now installs from a counted GitHub Release asset: the
update-homebrew release job builds agit archivetarball, uploads it to
the GitHub Release, and points the formula at it (sha256 of the same
bytes), so brew installs appear in release download counts. tests/test_dashboard_port_arg.py: coverage for the v7.31--port/--host
argparse on directpython -m dashboard.serverlaunches.
Fixed
- Cost-honesty (task 568): on the stock session path the estimator and the
dashboard now quote the model the runner actually dispatches (asonnet
session pin routes through the development tier to Opus; the quote said
Sonnet and underquoted ~1.7x).loki planoutput now names the model AND
the lever that selected it. Demo cost quote corrected to the Opus truth. - Tier-name session pins (
planning/development/fast), documented in
skills/model-selection.md, now resolve byte-faithfully across all three
readers (run.sh dispatch,loki planestimator, dashboard). run.sh folds
miscased/padded pins (trim + lowercase). Session-pin parity matrix grown
to 224 cells; dashboard endpoint tests 37/37. loki report export json kpiswas hijacked by a flag-anywhere kpis scan;
kpis routing now keys on the first non-flag token in all three layers
(bash dispatch, bin/loki shim, Bun CLI) with byte-parity against main.loki mcpprobe now resolves the server exactly like the launch exec
(file-exec of mcp/server.py with PYTHONPATH from the user cwd), so probe
and launch can no longer disagree under a decoy SDK; the runpy
RuntimeWarning on launch is gone and MCP serverInfo reports the real
loki-mode version.- A backtick command substitution inside the
loki planestimator comment
block executed a strayclaude --modelon every plan invocation,
polluting stderr (plan suite 19/25). Fixed;loki plan --jsonnow emits
zero stderr bytes (25/25). - Homebrew formula license corrected from MIT to BUSL-1.1 (honesty fix) and
description updated.
Changed
loki compoundhelp title reads the canonicalloki memory compound
(the alias keeps working with a pointer).- scripts/local-ci.sh MCP handshake check launches the server in the shipped
launcher form (file-exec + PYTHONPATH from a non-repo cwd) instead of
-m mcp.serverfrom the repo root.
Breaking Changes
- Legacy command `loki kpis` retained as alias for new canonical `loki report kpis`
- Legacy commands grouped under new nouns: `loki analyze` (explain/onboard/code/context) and `loki modernize` (heal/migrate)
- Homebrew formula license changed from MIT to BUSL-1.1
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]