Agent Monitoring
An AI agent can gain a new tool or broader permissions in a single configuration change. See its posture, review the baseline and follow the changes that matter.
support-apiBash(npm test:*)Bash(*)Source retained · proposal for review
Agent configuration and MCP tools.
Drift from a reviewed baseline.
Optional hooks for your own Claude Code.
INSIDE FENCIA02 / Agent Monitor
Claude Code had a reviewed configuration. A change expands its terminal permissions. See what the person approving the agent needs to know.
{ "permissions": { "allow": [ "Bash(npm test:*)" ] }}The source. The context. A starting point.
Fictional data. Workflow illustration, not a connected session or a dashboard screenshot.
CONNECTED TO YOUR TEAM’S WORK
WHAT YOU DO WITH FENCIA
An isolated signal makes your team investigate again. The sources view brings together provenance and what each connection contributed, helping distinguish a source without findings from one that needs attention.
EnlargeSources and evidence view in the test workspace.
Broader configuration does not explain why it changed. Bring drift into the review workflow: what it now permits, what your team had reviewed and which justification is missing. Defense helps prepare the work; the decision remains human.
EnlargeReview assistant entry point. The specific permissions comparison appears in the demo above.
When someone else picks up the work, they need the source, context and available evidence. The library brings workspace material together for review without relying on scattered conversations or the memory of the person who started the case.
EnlargeEvidence library with fictional data.
Real Fencia captures from a test workspace. Fictional data; shared views complement each feature’s specific demo.
BEFORE YOU START
No. Configuration review depends on supported files. The described runtime hooks are for Claude Code; that runtime monitoring is not offered for Codex or Cursor.
Each customer configures their own installation with hooks and a workspace ingestion token. Runtime functionality is in beta and pending acceptance with a real session.
IN YOUR TEAM’S WORKFLOW
Bring agent configuration into the same review as the code it works with. For Claude Code, optional hooks add activity and policy checks to that picture.
Review supported agent configuration and MCP servers in connected GitHub repositories.
Record the reviewed configuration and surface drift so a new permission does not disappear into a commit.
Each team connects its own installation using workspace ingestion credentials and hooks. Runtime policy can observe or enforce supported rules.
Claude Code runtime is beta and still awaiting a live-session acceptance test. Hooks send the full event over HTTPS; Fencia minimises it in memory before storing metadata. If Fencia is unreachable, Claude Code continues. Codex and Cursor do not have runtime monitoring.
No. Each customer uses their own Claude Code account and installation. Fencia connects that installation to their workspace; it does not provide a shared Anthropic licence.
Review how this workflow fits your team’s work with us.
Talk to Fencia