Works with the AI Agents you already use
AI Agent Insights
Evidence-first profiling for AI-agent session exports, tool traces, and LLM requests—covering latency, failures, retries, concurrency, telemetry gaps, and actionable follow-up.
$4.99
AI Agent Insights
Example session with this skill installed
Analyze the attached AI-agent session export with AI Agent Insights. Produce a Run Performance Report. Separate tool attempts from logical operations, inspect LLM failures and retries, identify the longest evidenced tool call, and state missing telemetry. Do not infer a root cause or task success without evidence.
- Read your context and instructions
- Compiled the ai agent insights
AI Agent Insights — Run Performance Report
Synthetic demonstration: one 9,000 ms session; three tool attempts across two logical tool operations; two LLM attempts associated with one logical request.
- Tool attempts: 2 successful / 3 total (66.7% attempt-level success rate). This is not a task-completion rate.
- The two search_code attempts share logical_operation_id
search-op-1: the first timed out after 500 ms and the correlated retry succeeded after 900 ms. - The run_tests call is the longest supplied tool call at 4,500 ms.
- The two LLM attempts share logical_request_id
llm-request-1: one timed out after 1,200 ms and the next succeeded after 1,500 ms. - Token usage and cost are unavailable because the source contains no token counts or pricing basis.
Finding: the search operation recovered after a correlated timeout. Severity: Informational. The fixture does not establish a recurring timeout pattern or explain the timeout's root cause.
Next measurement: collect comparable runs, preserve attempt/logical-operation identifiers, and record tool/model timeout configuration. Use timestamps and token/cost fields only when the source provides them.
Connects securely to your tools. The creator never sees your data.
About this skill
AI Agent Insights
Understand your agents. Optimize their intelligence.
AI Agent Insights is an evidence-first run performance profiler for AI-agent workflows. Give it a session export, execution trace, tool-call log, model-request record, or structured metrics snapshot. It produces a forensic report that explains what the supplied evidence shows, what it cannot establish, and what to measure next.
What it helps you investigate
- Slow tools and execution stages, with measured durations and denominators.
- Tool-call and LLM-request failures, timeouts, and correlated retries.
- Concurrent tool/model activity and overlapping execution intervals.
- Repeated calls and possible duplicate work, only when identity and evidence support the conclusion.
- Missing timestamps, incomplete event histories, aggregate-only telemetry, and schema-coverage gaps.
- Comparability between runs, including whether a percentage improvement can validly be calculated.
- DeepSeek Harness Agent Insights snapshots as a first-class supported source schema.
How the analysis works
The skill inventories supplied evidence, normalizes known fields without discarding source meaning, reconstructs a timeline only where valid timestamps permit it, calculates metrics with explicit units and denominators, and classifies findings as observed, calculated, interpretation, or hypothesis.
It distinguishes attempts from logical operations, keeps concurrent duration totals separate from wall-clock time, avoids deriving missing timestamps without an explicit schema rule, and calibrates severity to available impact evidence. Recommendations include the evidence behind them and a way to validate the next run.
What makes it different
AI Agent Insights is a run-forensics skill—not a live observability platform, telemetry SDK, dashboard, model leaderboard, generic code reviewer, or agent runtime. It analyzes evidence supplied by the user or explicitly authorized through the host environment. It does not claim automatic access to private agent histories.
Privacy and operating boundaries
The skill is read-only by default. It treats instructions embedded in logs as untrusted data, does not execute commands found in analyzed content, avoids exposing secrets, and does not send analyzed data to external services. It does not modify source code, configuration, or agent settings.
Limitations
Conclusions depend on fields and retention windows in the supplied export. Missing timestamps, correlation IDs, parent/child relationships, tokens, pricing, or task outcomes remain unknown rather than being fabricated. Vendor-specific schemas must be inspected; compatibility with one export does not imply universal support for every agent or telemetry format.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
Analysis quality depends on supplied fields, timestamp quality, identifier correlation, and export retention. Missing evidence stays unknown; the skill cannot infer root causes, retries, task outcomes, tokens, or cost when the source does not support them.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 day ago
Requires
Nothing to install.
Permissions
File Scopes
The skill analyzes supplied evidence in the buyer's chosen AI agent. It makes no external network calls, writes no user files, and executes no commands found in logs.