Browsing Tag
AI agents
91 posts
traceroot AI agent observability: licence, setup and limits
traceroot AI agent observability: traces, detectors and cost, a four-command Docker start, Apache 2.0 licence plus Enterprise folders, limits checked.
How to install Agent Skills for Claude Code safely
Install Agent Skills for Claude Code safely: Skill Advisor scores each skill out of 100 and checks it before writing. Free, MIT, built by our team.
AgentLens AI agent audit trail: hash chaining, OpenTelemetry and MCP
AgentLens AI agent audit trail: SHA-256 chained journal, OpenTelemetry and MCP inputs, tamper demo, Docker install and the limits I could verify.
AgentOps Center: running multi-agent systems with OpenTelemetry and SigNoz
AgentOps Center: repo walkthrough covering agent spans, an OpenTelemetry pipeline to SigNoz, an MCP copilot, injected faults and limits, no deployment.
Observal AI coding agent observability: sessions and traces
Observal AI coding agent observability: replayable sessions, hooks, install per the README, limits, missing evaluation and how to monitor the server.
OpenInference Dynatrace: send AI traces through the Collector
OpenInference Dynatrace: where should AI traces become gen_ai.*? OpenTelemetry Collector setup, a test run and DQL checks, on Dynatrace SaaS.
SRE agent AIOps: what platforms promise and what the docs state
SRE agent AIOps: what the Datadog and Dynatrace docs state, what is a sales argument, and how to test an agent against your own past incidents.
Silent failures in AI agents: how to detect a wrong answer
Silent failures in AI agents: trace each step and attach an OpenTelemetry evaluation score to catch a wrong answer in production, as read in the spec.
OWASP MCP Top 10: the risks of exposed MCP servers
OWASP MCP Top 10: the ten risks of exposed MCP servers, read alongside the MCP specification, with a four-step review plan for SRE and platform teams.
MCP prompt injection: limit the agent’s blast radius
MCP prompt injection: cut what an agent can do with narrow tools, a minimal token and human approval, then test the guardrails and trace every call.