Is supabase/agent-skills safe?
Yes. supabase/agent-skills is safe to install. Oathe's behavioral security audit gave the agent-skills skill by supabase a trust score of 94/100 with 4 findings, none critical or high.
https://github.com/supabase/agent-skills
Is supabase/agent-skills safe to install?
supabase/agent-skills is an official Supabase-maintained skill repository with clean clone behavior, no prompt injection, no post-install code execution, and all canary files intact. The only substantive concern is that the supabase skill instructs the agent to proactively fetch supabase.com/changelog.md on every activation, creating a minor dependency on that URL's content integrity. All sensitive file accesses observed in monitoring occurred 5 seconds before the git clone and are attributable to the audit framework's own initialization, not the skill. The skill is safe to install.
What security issues were found in supabase/agent-skills?
Category Scores
Findings (4)
LOW Agent instructed to fetch external URL on every task activation -6 ▶
The supabase SKILL.md instructs the agent to fetch https://supabase.com/changelog.md before implementing any Supabase feature. While the URL is hardcoded to Supabase's own domain and the purpose is legitimate (checking for breaking changes), this creates a dependency on a third-party URL being rendered into the agent's context on every activation. If that page were compromised or served different content, it could inject instructions into the agent's reasoning.
Evidence: "First, fetch `https://supabase.com/changelog.md` (a lightweight summary index — not a heavy pull), scan for `breaking-change` tags relevant to your task"
INFO MCP server configured to Supabase's official endpoint -7 ▶
.mcp.json configures the agent to connect to https://mcp.supabase.com/mcp — an official Supabase-operated MCP server. This is disclosed and consistent with the skill's stated purpose. Users should be aware that MCP tool calls will reach Supabase's infrastructure.
Evidence: { "mcpServers": { "supabase": { "type": "http", "url": "https://mcp.supabase.com/mcp?features=docs" } } }
INFO CLAUDE.md is a symlink to AGENTS.md -2 ▶
The repository symlinks CLAUDE.md → AGENTS.md within the same directory. This is benign and commonly used so both Claude Code (which reads CLAUDE.md) and other agents (which read AGENTS.md) receive identical instructions. No path traversal risk.
Evidence: lrwxrwxrwx 1 oc-exec oc-exec 9 Nov 16 07:16 /home/oc-exec/skill-under-test/supabase/agent-skills/CLAUDE.md -> AGENTS.md
INFO Broad skill trigger scope increases activation frequency -10 ▶
The skill description is exhaustively written to trigger on nearly any Supabase-related task including auth, RLS, Edge Functions, Realtime, Storage, CLI, and debugging. This is intentional per the Agent Skills standard (triggers must be in the description), but means the skill will load on a very wide range of user requests, consuming additional context tokens.
Evidence: description: "Use when doing ANY task involving Supabase. Triggers: Supabase products (Database, Auth, Edge Functions, Realtime, Storage, Vectors, Cron, Queues); client libraries..."
Should I install supabase/agent-skills?
Oathe's verdict for supabase/agent-skills is SAFE with a trust score of 94/100. Recommendation: Install.