Added
- End to end, for real: the guard goes into each agent's call path.
cslcore wire (and setup,
after activation) shows the change as a diff and makes it when confirmed: a PreToolUse hook for
every tool in a Claude Code project's .claude/settings.local.json, and for Python tool functions
(LangChain @tool, OpenAI Agents @function_tool, CrewAI, plain functions) the guard at the top of
the file and @_csl_guard.tool("name") above each tool function. Files are copied into the
workspace first; cslcore wire --undo puts them back; a file changed after its diff was shown is
never written. Only agents with an active policy are wired. The host is scanned again afterwards.
guard.tool(name): a decorator on a tool function that decides each call before the function
runs, keeping its name, docstring and signature (sync and async).cslcore setup --wire: with --yes, also make the wiring change.x on an agent nothing guards yet (on the map and in the live panel) offers to put it under a
guard first: a policy drafted from its tools and checked, the wiring change, a new scan, then the
freeze. Each step is shown and confirmed.
Changed
- The map, the live panel and the setup summary only show an agent as guarded when the scan sees
its guard in the call path (a guard in its code, or a cslcore hook for every tool). An active
policy alone is "not wired: nothing stops it yet"; freezing such an agent is no longer offered as
if it would stop it. Agents under a guard stay on the map even when nothing reaches through them.
- A Claude Code hook that covers only some tools (a matcher such as
Bash) is reported as such, and
wiring adds one for every tool.
pip install -U csl-core
uvx --refresh csl-core venom
