from orchestkit171
Query local, privacy-safe cross-project analytics to report on agent, skill, hook, and team performance; replay sessions and estimate token costs.
This skill provides a suite of local analytics and reporting commands for OrchestKit projects. It reads local analytics files (JSONL) to produce reports such as top agents, model delegation breakdowns, top skills by invocation, hook failure rates, team activity, session replays, cost estimates, and trends over time. Outputs are intended as markdown tables with counts, percentages, and averages for easy review.
Use the analytics skill when reviewing agent and skill performance across projects, investigating slow or failing hooks, replaying session timelines for debugging or auditing, estimating token costs, or generating daily/weekly trend summaries. It is intended for local developer or operator environments where analytics JSONL files are stored under a known directory.
references/ directory with ready-to-run jq queries, session replay guidance, cost estimation formulas, and trends analysis documentation (has_references=true)agents, models, skills, hooks, teams, session, cost, trends, summary), using jq to run queries, and presenting results as markdown tables. The skill documents file locations and rules for streaming large files and privacy-safe handling.Designed for agents that can read local files and run shell tools (jq, bash) — e.g., Claude Code, local dev agents, or other agents with filesystem access.
Cross-project analytics skill for OrchestKit that queries local JSONL files (agent-usage, skill-usage, hook-timing, team-activity, session data) using jq. Privacy-safe by design with hashed project IDs and no PII. Well-structured with clear subcommands, reference files, and rule files for progressive disclosure. No scripts included — relies on agent executing jq commands against local Claude analytics data. Niche usefulness as it requires OrchestKit/Claude Code analytics data to exist.
Clean skill with no security concerns. All data is local-only, privacy-safe by design. No network calls, no credentials, no destructive commands. Architecture follows skill spec well with references/ and rules/ directories. Good progressive disclosure pattern. Lower usefulness due to niche audience (only OrchestKit/Claude Code users with analytics data).
