audit

A read-only review of a machine-learning training pipeline, grading possible improvements and estimating their cost and risks.

In plain words
What is it for?
Use it to review training loops, rank algorithmic, system, protocol, and architecture improvements, and record the evidence, time required, and possible side effects.
Why use it?
It shows where training may be slow or leave performance unused without changing the code or making decisions for you.

Skill for Claude CodeCodex

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 skills/emaballarin/ccplugins/audit
Any agent
npx skills add emaballarin/ccplugins --skill audit
Clone the repo
git clone --depth 1 https://github.com/emaballarin/ccplugins

Made for: Claude Code, Codex.

Per session 183 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,403 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 $0.00183 $0.01403
Opus 5 $0.00092 $0.00701
Sonnet 5 $0.00037 $0.00281
Haiku 4.5 $0.00018 $0.00140

Measured yesterday against content hash 6098fd84259d, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

audit 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 yesterday.

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/tuneml/skills/audit/SKILL.md · 116 lines

How it starts

The opening of the file, as written. The whole thing — 116 lines — stays where its author put it; the contents beside it link to each section on GitHub.

/tml:audit — what is this pipeline leaving on the table

Produce a short, ranked, graded list of what could be improved, what each item would cost to take, and what it would silently break. Auditing changes nothing and decides nothing — /tml:plan decides.

First action, always

Establish what already exists before reading a line of model code:

ls -la ./.tml/ 2>/dev/null; git -C . log --oneline -3 2>/dev/null

If findings.md is already there, read it, say when it was written, and audit as a delta — re-confirming what changed, not re-deriving what did not. An audit that silently re-proposes what was already rejected is noise.

Hard rules

  1. Read-first. Never edit project code. The only path written is ./.tml/findings.md, and only after the findings have been shown. No global or shared state is touched.
  2. Every finding carries a grade. From the ladder in references/evidence-grades.md §1. An ungraded claim is not emitted; folklore is a respectable answer.
  3. Every finding is priced in time-to-target, decomposed into steps-to-target × time-per-step, and states whether it moves the other factor adversely. Throughput is a diagnostic, never a result.
  4. Every tier-A/B/D item declares a quality exposure, separately from its radius. Radius says how to verify; exposure says whether you are allowed to.
  5. Recommending nothing is a valid outcome. A pipeline with no worthwhile changes should be told so, in one paragraph.

Procedure

1. Scope, and the two questions that change the answer

Find the training entry point, the data path, the eval path, and the harness. Then establish, by asking rather than inferring:

  • The target. What quality, on what metric, measured how? Without it nothing can be priced, because "time-to-target" has no target.
  • The constraint. Wall-clock, device-hours, memory, or deadline?

If the pipeline is already instrumented, read references/regime.md §1 and note the parallelism regime — it changes what /tml:plan can propose next.

Read the full file on GitHub · 116 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. yesterday First seen · 116 lines · 183 tokens per session scan A 6098fd84259d

Subscribe to this mod's changes

audit is a skill published in the GitHub repository emaballarin/ccplugins (3 stars, last pushed 26d ago), licensed MIT. It adds 183 tokens to every session and 1,403 once invoked, about $0.0009 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 2026-08-31.

Related

Other skills, from other repositories

systematic-debugging

Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes.

obra/superpowers · 21 tokens

next-cache-components-adoption

Turn on Cache Components in a Next.js app and resolve the blocking routes it surfaces. Use when the user wants to enable, adopt, or migrate to Cache Components, flip the cacheComponents flag, work through a flood of blocking-prerender / instant validation errors, run the cache-components-instant-false codemod, or…

vercel/next.js · 95 tokens

babysit-pr

Babysit a GitHub pull request after creation by continuously polling review comments, CI checks/workflow runs, and mergeability state until the PR is merged/closed or user help is required. Diagnose failures, retry likely flaky failures up to 3 times, auto-fix/push branch-related issues when appropriate, and keep…

openai/codex · 114 tokens

imagegen

Generate or edit raster images when the task benefits from AI-created bitmap visuals such as photos, illustrations, textures, sprites, mockups, or transparent-background cutouts. Use when Codex should create a brand-new image, transform an existing image, or derive visual variants from references, and the output…

openai/codex · 113 tokens

cpu-profile-analysis

Analyze V8/Chrome CPU profiles (.cpuprofile) and DevTools trace files (Trace-.json). Use when: profiling performance, investigating slow functions, comparing code paths, finding bottlenecks, analyzing timeToRequest, understanding call trees from sampling profiler data, analyzing layout/paint/rendering, investigating…

microsoft/vscode · 71 tokens

next-cache-components-optimizer

Drive a Next.js route to instant navigation by setting up an agentic loop, under Cache Components / PPR, on initial load (hard navigation) and client-side navigation (soft navigation). Encode the goal as a failing @next/playwright instant() e2e and work it to green, one verified route at a time; the shipped test then…

vercel/next.js · 170 tokens