Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/devonperoutky/agent-profiler/claude-mdgit clone --depth 1 https://github.com/DevonPeroutky/agent-profilerWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.04101 | $0.04101 |
| Opus 5 | $0.02050 | $0.02050 |
| Sonnet 5 | $0.00820 | $0.00820 |
| Haiku 4.5 | $0.00410 | $0.00410 |
Grade A, and why
agent-profiler CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 178 lines — stays where its author put it; the contents beside it link to each section on GitHub.
agent-profiler
A local trace viewer for Claude Code conversations. Reads session transcripts directly from disk and renders them as a span-tree waterfall. A React + Vite app with a small middleware plugin — no separate server process, no hooks, no OTEL SDK.
Communication style
Speak simply and explain things clearly. Avoid jargon — when a domain term is genuinely necessary (e.g. requestId, tool_use), define it the first time it appears in a response. Prefer plain-English descriptions over acronyms and library-internal naming. If a concept can be explained with a short sentence instead of a technical label, use the sentence.
North Star
Claude Code's own per-session JSONL transcripts (~/.claude/projects/<project-slug>/<sessionId>.jsonl) are the source of truth. The viewer is a pure (transcript → SpanNode tree → JSON) pipeline. If the UI and the raw transcript disagree about what happened, the transcript is right.
Core vocabulary — inference vs. tool call
These two are not the same thing. Confusing them is the largest recurring source of bugs in this transformer.
Inference. A single request to the remote Anthropic API. Something that costs money, takes wall-clock time, and ends with the model returning a response. Has a requestId — one per API round-trip. Whatever content the model returned (thinking, text, tool_use, or any combination) belongs to that one inference and shares its requestId. The state layer mirrors API round-trips 1:1: one inference span per distinct requestId within a slice, regardless of what content kinds the response contained. A response with [thinking, text, tool_use] is one inference, not three. How that inference is rendered (waterfall bar, chat bubble, both, neither) is a UI decision driven by has_* flags and the events list — the state layer is content-kind-agnostic.
The canonical predicate for "an assistant record that counts" is isCanonicalAssistantRecord (type === 'assistant', !isApiErrorMessage, has message.usage, has non-empty requestId). It gates both the inference emitter and token dedupe (dedupeUsagesByRequestId in lib/traces/traces.js) — inferences.length equals agent_trace.turn.request_count by construction. Drift here recreates the count discrepancy on any turn with API errors.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 3d ago First seen · 178 lines · 4,101 tokens per session scan A c63f71f30084
agent-profiler CLAUDE.md is an instructions file published in the GitHub repository DevonPeroutky/agent-profiler (13 stars, last pushed 3mo ago), licensed Apache-2.0. It adds 4,101 tokens to every session, about $0.0205 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
claude-code-tresor CLAUDE.md
Instructions for alirezarezvani/claude-code-tresor, covering claude.md - development guide, 🎯 repository purpose, 🏗️ architecture, 🛠️ common development commands and building & testing.
pi-coding-agent-forge AGENTS.md
Instructions for Firstp1ck/pi-coding-agent-forge, covering repository documentation rules, documentation goal, required documentation layers, readme.md — human guide and technical.md — advanced user reference.
kkRepo java-server.instructions.md
Instructions for klboke/kkRepo, a project described as: kkRepo is a Nexus-compatible, self-hosted artifact repository for Maven, npm, PyPI, Go, Helm, NuGet, Cargo/Rust, Dart/Pub, Docker/OCI, RubyGems, and Yum artifacts. It supports one-click migration from Nexus to kkRepo.
huly-mcp CLAUDE.md
Instructions for dearlordylord/huly-mcp, covering project instructions, design principle: llm-first api, project harness (copy to new projects), package manager and verification.
pitway CLAUDE.md
Instructions for thixpin/pitway: This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository. The project's full instructions live in AGENTS.md — this file only points there so the two can never drift apart.
skills AGENTS.md
Instructions for vanducng/skills, covering repository guidelines, project structure & module organization, build, test, and development commands, coding style & naming conventions and documentation sync.