Repository navigation
A log source for AI coding-agent tool calls? #6550
blitzcrieg1
started this conversation in
Ideas
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Hi, a question about whether this belongs in Sigma yet.
AI coding agents (Claude Code, Cursor, Codex, Copilot) now run shell commands, read files and call MCP servers on developer machines, and several expose hooks that report each tool call. I maintain Agentmetry (github.com/blitzcrieg1/agentmetry, Apache-2.0), an open-source sensor that records those calls, and it ships 23 Sigma rules today: docs/integrations/sigma in the repo. They use
logsource: product: agentmetryand our own field names (action.type,tool.qualified,detection.rule_id), which is exactly the problem: they only fit our schema.Is there appetite for a vendor-neutral log source for AI coding-agent activity, so rules like "agent read a .env file, then made an outbound request" can be written once? If so, what would you need first: a taxonomy proposal, sample events, or a pySigma pipeline? If it is too early, saying so is a useful answer too.
What our rules cannot see: browser chat use, model providers, and any agent running with its hooks off.
All reactions