By mistakeknot
Token efficiency benchmarking, session analytics, and API-equivalent cost analysis for agent workflows
Parse conversation JSONL files and backfill token counts into SQLite
Show API-equivalent cost analysis with model-specific pricing and subscription leverage
Show token efficiency analysis and decision gate verdict
Show collection progress and pending parse status
Token efficiency benchmarking for Claude Code.
interstat answers the question "am I actually using tokens efficiently or just burning through context?" It captures tool usage events in real-time via a PostToolUse:Task hook, backfills token data from JSONL transcripts at session end, and produces reports with percentiles and a decision gate.
The two-phase data collection is deliberate: real-time hooks capture the event structure (which tools, what order, how many subagents) while the JSONL backfill captures the actual token counts (not available during the session). Together they give you a complete picture of where tokens are going.
First, add the interagency marketplace (one-time setup):
/plugin marketplace add mistakeknot/interagency-marketplace
Then install the plugin:
/plugin install interstat
Generate a token efficiency report:
/interstat:interstat-report
Check current session metrics:
/interstat:interstat-status
Deep analysis of usage patterns:
/interstat:interstat-analyze
hooks/
post-tool-use.sh PostToolUse:Task: real-time event capture to SQLite
session-end.sh SessionEnd: JSONL parsing for token backfill
skills/
report/ Percentile analysis with decision gate
status/ Current session snapshot
analyze/ Deep pattern analysis
Data lives in ~/.claude/interstat/metrics.db (SQLite, WAL mode for concurrent hook writes).
Matches all tools
Hooks run on every tool call, not just specific ones
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
npx claudepluginhub mistakeknot/interagency-marketplace --plugin interstat[DEPRECATED — use intervox] Analyze your writing style and adapt Claude's output to sound like you. Replaced by intervox (via the interim intervoice), which adds a measured stylometric fingerprint, LLMism linter, and verified closed-loop apply on top of the global multi-register profile.
Self-improving agent rig: codifies product and engineering discipline into composable workflows from brainstorm to ship. Compounds knowledge, generates domain agents, monitors its own docs, and surfaces conservative update drift. Orchestrates Claude, Codex, and Oracle through 6 agents, 57 commands, 20 skills, 0 MCP servers. Factory substrate: CXDB turn DAG, scenario bank with satisfaction scoring, evidence pipeline, agent capability policies. Companions: interspect, interphase, interline, interflux, interpath, interwatch, interslack, interform, intercraft, interdev, interpeer, intertest.
Recursive AGENTS.md generator with integrated Oracle critique, CLAUDE.md harmonization, incremental updates, diff previews, and smart monorepo scoping. Cross-AI compatible.
Multi-agent review and research with scored triage, domain detection, content slicing, intermediate finding sharing, and knowledge injection. 17 agents (12 review + 5 research), 8 commands (incl. flux-melange — a goal-seeking adaptive review loop with lens fusion and novelty/risk/taste scoring), 3 skills (flux-engine, flux-review-engine, flux-melange-engine), 2 MCP servers (exa, openrouter-dispatch). Companion plugin for Clavain.
Product artifact generator — roadmaps, PRDs, vision docs, changelogs, CUJs, and status reports from beads, brainstorms, and project state. Companion plugin for Clavain.
Analyze Claude Code session token usage, costs, trends, and anomalies
Profile tokens, API-equivalent cost and time per task/subagent/model — straight from a chat. Requires the `quickai` binary on PATH (brew install AlexGladkov/tap/quickai).
Generate an explorable HTML report of Claude Code session usage — tokens, cache efficiency, subagents, skills, and the most expensive prompts — from local ~/.claude/projects transcripts.
Audit, fix, and monitor Claude Code context window usage. Find the ghost tokens.
Visual dashboard for Claude Code & Codex — usage/token analysis of skills, agents, plugins, hooks, memory, MCP servers, rules, and principles
Tracks Claude Code session usage (tokens, costs, skills, MCP, subagents) and sends data to a self-hosted dashboard.