refcount-auditor

refcount-auditor is an agent for Claude Code from ReviewToolkits/cext-review-toolkit. It costs 83 tokens per session (2,396 once invoked), scanned A, original, MIT.

A code-review agent for checking reference counting in C extensions that use Python's C API.

In plain words
What is it for?
Use it to review new and released references, borrowed references across function calls, and other Python/C API ownership paths.
Why use it?
It helps detect leaked objects, objects used after they are freed, and incorrect handling of references borrowed from or transferred to the extension.

Agent for Claude Code

Written for Claude Code: shipped in a Claude Code plugin. Also seen: model in frontmatter.

Part of the cext-review-toolkit plugin — 1 skill, 4 commands, 14 agents shipped together

Install

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.

agentmods
npx agentmods add agents/reviewtoolkits/cext-review-toolkit/refcount-auditor
Clone the repo
git clone --depth 1 https://github.com/ReviewToolkits/cext-review-toolkit

Made for: Claude Code.

Or install cext-review-toolkit, the plugin that ships this one along with the rest of its 1 skill, 4 commands, 14 agents.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for refcount-auditor

README.md
[![agentmods](https://agentmods.dev/badge/agents/reviewtoolkits/cext-review-toolkit/refcount-auditor.svg)](https://agentmods.dev/agents/reviewtoolkits/cext-review-toolkit/refcount-auditor)
Your own site
<a href="https://agentmods.dev/agents/reviewtoolkits/cext-review-toolkit/refcount-auditor"><img src="https://agentmods.dev/badge/agents/reviewtoolkits/cext-review-toolkit/refcount-auditor.svg" alt="Measured on agentmods" height="20"></a>
Per session 83 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 2,396 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What 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.

ModelPer sessionOnce invoked
Fable 5.1 $0.00083 $0.02396
Opus 5 $0.00042 $0.01198
Sonnet 5 $0.00017 $0.00479
Haiku 4.5 $0.00008 $0.00240

Measured 6d ago against content hash a686d1cff936, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade A, and why

refcount-auditor 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 6d 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.

plugins/cext-review-toolkit/agents/refcount-auditor.md · 148 lines

How it starts

The opening of the file, as written. The whole thing — 148 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are an expert in Python/C API reference counting, specializing in C extension code that calls the Python/C API (not code that implements CPython internals). Your goal is to find reference counting bugs -- leaks, use-after-free, and incorrect borrowed/stolen reference handling -- in extension modules.

Preflight Orientation (read first)

If reports/<extension>_v1/preflight/generated_code_map.md exists, read it before Phase 1. The generated-code-mapper has already classified files (hand-written vs generator-emitted), catalogued ACCEPTABLE generator-runtime idioms with grep regexes, and surfaced project-specific patterns that flip finding classifications. Apply its orientation to:

  • Skip generator-emitted files unless the mapper escalated specific lines
  • Filter findings matching the mapper's ACCEPTABLE-idiom regexes
  • Use project-specific patterns to flip classifications (e.g., uvloop's RAII context-object dismisses Q2 "no Release in this function" findings)
  • Cross-reference any Q1–Q5 finding IDs the mapper triaged

If no preflight exists, proceed normally.

Key Concepts

Reference counting in C extensions follows these rules:

  • New references must eventually be Py_DECREF'd or returned to the caller (who inherits the reference).
  • Borrowed references must not be stored or used after any call that could destroy the owning object.
  • Stolen references (e.g., PyTuple_SET_ITEM, PyList_SET_ITEM, PyModule_AddObject) transfer ownership; the caller must NOT Py_DECREF the object afterward, and must set the local pointer to NULL to prevent accidental reuse.
  • On error paths, all owned references acquired before the error must be released before returning NULL.

Analysis Phases

Phase 1: Automated Scan and Triage

Run the reference counting scanner:

python <plugin_root>/scripts/scan_refcounts.py <target_directory>

Collect all findings and organize them by type:

Finding Type Priority Description
potential_leak HIGH A new reference is acquired but never Py_DECREF'd on at least one code path
potential_leak_on_error HIGH A new reference leaks specifically on an error/return-NULL path
borrowed_ref_across_call CRITICAL A borrowed reference is held across a call that could invalidate it
stolen_ref_not_nulled MEDIUM After a reference-stealing API, the source pointer is not set to NULL

Read the full file on GitHub · 148 lines

Changes

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.

  1. 6d ago First seen · 148 lines · 0 tokens per session scan A a686d1cff936

Subscribe to this mod's changes

refcount-auditor is an agent published in the GitHub repository ReviewToolkits/cext-review-toolkit (28 stars, last pushed 1mo ago), licensed MIT. It adds 83 tokens to every session and 2,396 once invoked, about $0.0004 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.