Identify and document root causes of agent behavior deviations in-session.
npx skills add https://github.com/ddaanet/agent-core --skill reflect-ddaanet
Please help me install this Agent Skill. Skill: reflect Source: https://github.com/ddaanet/agent-core/tree/main/skills/reflect Command: npx skills add https://github.com/ddaanet/agent-core --skill reflect-ddaanet
Perform structured root cause analysis of agent behavior deviations within the current session context. Diagnoses why an agent violated rules, bypassed constraints, or rationalized exceptions.
Emit the diagnostic framing block, identify the deviation, perform RCA, classify the fix scope, and decide an exit path.
High-intent search queries and answers about installing and using this skill.
Root cause analysis for agent behavior deviations is performed by applying a structured five-phase diagnostic protocol to identify why rules were violated or constraints bypassed. This process transitions from execution to diagnostic reasoning to pinpoint deviation points and rule gaps.
You need root-cause-analysis on an active session immediately after an agent violates rules, bypasses constraints, or rationalizes exceptions. Keeping the analysis in-session ensures accurate context is captured to produce actionable RCA reports and learning artifacts.
The best way to document agent deviations and rule gaps is to generate structured RCA reports, learning artifacts, and task lists. This documentation captures the root causes of bypassed constraints and provides actionable outputs for upstream fixes and future prevention.
Diagnosing agent deviations outside the current session context is not recommended. The root-cause-analysis must remain in-session to ensure accurate context is preserved when identifying why an agent violated rules or rationalized exceptions.
ctejada85Trace execution paths and verify hypotheses to diagnose defects in Rust/Bevy applications.
joshftAudit Claude Code workflow phases and rule coverage for execution gaps.
Greenmamba29Monitor security audit logs and generate real-time alerts for unauthorized access.
wangqianCAIDiagnose OpenClaw deployment failures by isolating configuration, model transport, and tool execution boundaries.
