MulmoClaude

v0.6.2 Security

This release includes 3 security fixes for security teams reviewing exposed deployments.

Published 22d LLM Frameworks

View tool

✓ No known CVEs patched

Read the diff → Tool health → What is this tool? →

This release patches 3 known CVEs

Affected surfaces

auth rbac

ReleasePort's take

Light signal

editorial:auto 13d

v0.6.2 introduces configurable Claude reasoning effort levels in Settings alongside UI reorganization, new SEC filing and SVG rendering plugins, and workspace-scoped agent permissions. Security improvements include webhook rate limiting and XSS mitigations for webhook challenges and wiki link rendering.

Why it matters: Test configurable effort levels in dev to match query complexity. Apply workspace permissions immediately to simplify first-run agent tool access. Deploy rate limiting and XSS mitigations for webhook security.

Summary

AI summary

Configurable Claude effort levels, reorganized Settings UI, and file drop improvements across Highlights, Bridges, and Wiki.

Changes in this release

Type	Severity	Summary	CVE
Security
Security	Medium	Rate limiting applied to all webhook bridges with proxy trust. Rate limiting applied to all webhook bridges with proxy trust. Source: llm_adapter@2026-05-21 Confidence: high	—
Security	Medium	Wiki link rendering escapes HTML to prevent XSS attacks. Wiki link rendering escapes HTML to prevent XSS attacks. Source: llm_adapter@2026-05-21 Confidence: high	—
Security	Medium	Rate limit keyGenerator routes through ipKeyGenerator for IPv6 safety. Rate limit keyGenerator routes through ipKeyGenerator for IPv6 safety. Source: llm_adapter@2026-05-21 Confidence: high	—
Security	Medium	Webhook hub.challenge endpoint mitigates reflected XSS via whitelisting. Webhook hub.challenge endpoint mitigates reflected XSS via whitelisting. Source: llm_adapter@2026-05-21 Confidence: low	—
Security	Medium	hub.challenge endpoint whitelists response shape to mitigate reflected XSS (CodeQL js/reflected-xss). hub.challenge endpoint whitelists response shape to mitigate reflected XSS (CodeQL js/reflected-xss). Source: granite4.1:30b@2026-05-23-audit Confidence: low	—
Feature
Feature	Medium	New Model tab exposes configurable reasoning effort levels in Settings. New Model tab exposes configurable reasoning effort levels in Settings. Source: llm_adapter@2026-05-21 Confidence: high	—
Feature	Medium	Settings menu reorganized from tabs to grouped left sidebar layout. Settings menu reorganized from tabs to grouped left sidebar layout. Source: llm_adapter@2026-05-21 Confidence: high	—
Feature	Medium	Chat panel drag-and-drop improved with visual affordance and default guard. Chat panel drag-and-drop improved with visual affordance and default guard. Source: llm_adapter@2026-05-21 Confidence: high	—
Feature	Medium	New edgar plugin provides SEC filings access for agent tools. New edgar plugin provides SEC filings access for agent tools. Source: llm_adapter@2026-05-21 Confidence: high	—
Feature	Medium	New presentSVG plugin renders generated SVGs as inline canvas surfaces. New presentSVG plugin renders generated SVGs as inline canvas surfaces. Source: llm_adapter@2026-05-21 Confidence: high	—
Feature	Medium	Fixed roles replaced with user-editable preset skills for greater flexibility. Fixed roles replaced with user-editable preset skills for greater flexibility. Source: llm_adapter@2026-05-21 Confidence: low	—
Feature	Medium	Fixed roles replaced with user‑editable preset skills (e.g., cookingCoach → mc-cooking-coach). Fixed roles replaced with user‑editable preset skills (e.g., cookingCoach → mc-cooking-coach). Source: granite4.1:30b@2026-05-23-audit Confidence: low	—
Bugfix	Medium	Workspace-scoped agent permissions eliminate first-run tool call blocking. Workspace-scoped agent permissions eliminate first-run tool call blocking. Source: llm_adapter@2026-05-21 Confidence: high	—

Full changelog

Highlights

Configurable reasoning effort (#1320 / #1323)

New Model tab in Settings exposes claude --effort (low / medium / high / xhigh / max). Persisted under <workspace>/config/settings.json; unset → Claude's default. Settings reload per-run, so the change applies on the next message without restart.

Settings menu reorganised (#1333)

The horizontal tab strip is now a grouped left sidebar (LLM / Servers / Workspace / Plugins). Modal widens from 36rem → 52rem (capped at 95vw on small viewports). All existing data-testid selectors preserved.

File drop on the chat panel (#1289)

Drag-and-drop now lights up the entire chat panel (was: just the input), with a visual affordance. The window default guard prevents accidental browser navigation when the drop lands outside.

EDGAR built-in plugin + Investor role

New server-only edgar plugin gives the agent direct access to SEC filings. Bundled into a new Investor role with Yahoo Finance instructions.

presentSVG plugin

New built-in plugin renders generated SVGs as inline canvas surfaces.

Preset skills replace fixed roles

cookingCoach role → mc-cooking-coach preset skill (#1286). settings role → mc-settings preset skill (#1283), then split into 3 focused subskills. Preset skills are user-editable; fixed roles aren't.

Workspace-scoped agent permissions

Permission allow-rules are now provisioned at server startup, so first-run permission prompts no longer block routine tool calls.

Security

All 6 webhook bridges: express-rate-limit + env-driven trust proxy.
Bridges: hub.challenge echoed as text/plain with whitelisted shape (CodeQL js/reflected-xss).
Wiki: HTML-escape target + display in renderWikiLinks (XSS).
Rate-limit keyGenerator routed through ipKeyGenerator for IPv6 safety.

Full Changelog

See CHANGELOG.md for the complete list including 30+ refactors and bug fixes.

Security Fixes

HTML-escape target in Wiki `renderWikiLinks` to prevent XSS
Echo `hub.challenge` as `text/plain` with whitelisted shape for Bridges
Rate-limit webhooks via `express-rate-limit` and route `keyGenerator` through `ipKeyGenerator` for IPv6 safety

View diff on GitHub

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

Share on X Share on Bluesky

Track MulmoClaude

Get notified when new releases ship.

About MulmoClaude

All releases →

Related context

Related tools

Earlier breaking changes

v0.6.4 `General` role split into lean `General` and new `Personal` role; Encore seed role pinned to Personal.