plinde

Skillscope — Data skill for Claude Code

Data community

Claude Code skill-invocation analytics — local JSONL transcripts, no Enterprise/OTel required.

How to install Skillscope

This entry records only its repository, not the path inside it, so there is no exact command to give. Open plinde/skillscope and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Skillscope does

Claude Code skill-invocation analytics — local JSONL transcripts, no Enterprise/OTel required.

Alternatives in Data

Private GPT — Complete API layer for private AI applications on local models: RAG, skills, tools, MCP, text-to-sql, and more 57.5k ★
HTML Anything — ✨ The agentic HTML editor — your local AI agent writes the HTML, you ship it 8.5k ★
Build Failure — Diagnose elasticsearch-js CI and local build failures 5.3k ★

README

skillscope

Claude Code skill-invocation analytics. Local-only: parses `~/.claude/projects/**/*.jsonl` transcripts directly — no Enterprise admin console, no org API, no OTLP sink required. Works for any individual Claude Code install.

Rust is the primary, shipped implementation (this directory), promoted after a bake-off against Go and Python POCs (see `experiments/`).

Agent support

Claude Code only, currently. skillscope parses `~/.claude/projects/**/*.jsonl` — Claude Code's specific transcript format (`sessionId`/`cwd`/`timestamp` per line, `` tags for slash invocations, `Skill` tool_use blocks for proactive ones, `sessions-index.json` per project).

Codex (`~/.codex/sessions/`) and OpenCode (`~/.local/share/opencode/`) both log equivalent per-session JSONL/JSON transcripts with a comparable shape (turn-by-turn messages, tool calls, timestamps, cwd). Adding either as a second `Origin`/data source is expected to be a parser + model addition, not a rearchitecture: implement a `parser::` module producing the same `SkillInvocation` (or agent-equivalent "tool/command invocation") struct the aggregation, fidelity, and TUI layers already consume, and gate it behind a `--agent claude|codex|opencode` flag or auto-detection by which directory exists. No PRs open for this yet.

What it does

Per-skill usage counts with trigger-type breakdown (user /slash vs model-proactive Skill tool_use)
Session-level drill-down — which sessions fired a skill, in which project/cwd
Session-scoped modes — skillscope . (fzf picker over sessions for the current cwd) and skillscope (full UUID or >=8-char hex prefix) open a TUI scoped to one session, toggled between a skills-in-session table and a flat chronological timeline (Tab)
Temporal firing patterns — daily/weekly cadence, trends, dead periods
Trigger-fidelity evals — under-triggering (skill matches intent but never fired) and over-triggering (fire