set-core: Skill for Claude Code

.claude/skills/evidence-discipline/SKILL.md

evidence-discipline is a skill for Claude Code from tatargabor/set-core. It costs 101 tokens per session (7,977 once invoked), scanned C, original, MIT.

Reference material for checking whether measurements really prove a software defect and its fix. It describes defect types, the checks that find them, and ways to rerun tests after changes.

In plain words
What is it for?
It supports debugging and review of evidence from tests, mutations, restored state, generated files, and other measurements.
Why use it?
It helps avoid trusting misleading values, missing data, or a supposed fix that was never actually verified.

Skill for Claude Code

Written for Claude Code: installed under .claude/. Also seen: mentions subagents.

This is tatargabor/set-core's own configuration. It tells Claude Code how to work on set-core itself, so it is not a mod to install elsewhere. Copy it as a starting point and replace the rules that are about this project. Everything set-core configures →

Needs its repository: it reads a path above its own folder, which exists only inside the repository. The line is [`.claude/rules/evidence-discipline.md`](../../rules/evidence-discipline.md); this file.

Reuse

Borrowing it

Nothing to install: this file belongs to tatargabor/set-core. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file
curl -O https://raw.githubusercontent.com/tatargabor/set-core/main/.claude/skills/evidence-discipline/SKILL.md
Clone the repo
git clone --depth 1 https://github.com/tatargabor/set-core

Made for: Claude Code.

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 evidence-discipline

README.md
[![agentmods](https://agentmods.dev/badge/skills/tatargabor/set-core/evidence-discipline/github.svg)](https://agentmods.dev/skills/tatargabor/set-core/evidence-discipline)
Your own site
<a href="https://agentmods.dev/skills/tatargabor/set-core/evidence-discipline"><img src="https://agentmods.dev/badge/skills/tatargabor/set-core/evidence-discipline/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for evidence-discipline

Your own site · 80×15
<a href="https://agentmods.dev/skills/tatargabor/set-core/evidence-discipline"><img src="https://agentmods.dev/badge/skills/tatargabor/set-core/evidence-discipline.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 101 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 7,977 The whole file, excluding the scripts and references it only reads on demand.
Security scan C 1 finding. A grade says what 26 rules found in the file — not that it is safe.
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 $0.00101 $0.07977
Opus 5 $0.00051 $0.03988
Sonnet 5 $0.00020 $0.01595
Haiku 4.5 $0.00010 $0.00798

Measured 3d ago against content hash 4738e91309a4, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-08, from the pricing page.

Security

Grade C, and why

evidence-discipline scanned grade C with 1 finding 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.

Recursive force deletehighDestructive command

rm -rf with a variable or a broad path is one typo away from removing the wrong tree.

find lib -name '__pycache__' -type d -exec rm -rf {} + 2>/dev/null
.claude/skills/evidence-discipline/SKILL.md · 488 lines

How it starts

The opening of the file, as written. The whole thing — 488 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Evidence discipline — the worked cases

The one-line NAME of every class stays resident in .claude/rules/evidence-discipline.md; this file carries the measurement behind each one. Split on 2026-08-22 so that ~30 KB of worked examples load when a measurement is actually in question rather than in every session. Nothing was cut — the text below is the original file verbatim from its class list on.

The defect classes, because naming them is what makes them findable

False value. A field the system no longer stands behind is still emitted, and lands on screen next to its replacement, contradicting it. Nobody notices, because both numbers look like data. The fix belongs at the source — the producer declares what it has deprecated — never at the display, and never as a hard-coded field name in a layer that is supposed to be domain-free.

False absence. The mirror image, and the more dangerous one: the system announces that something is hidden, missing, or suppressed — when it was never there. "1 deprecated field hidden" about a field the producer stopped sending. A declaration is not data. Count from the data; use the declaration only to know what to look for.

Prose read as fact. Any parser reading human or model prose will eventually read an example as an instruction: a fenced code block parsed as a directive, a rule quoted before a verdict read as the verdict, a ## inside a fence ending a section. Anchor such parsers — line start, not quoted, last occurrence wins, near the end of the output — and test them on the shape that actually broke, not on the happy path.

A proxy measured instead of the thing. ps -p <pid> answers whether a process holds that number, not whether your process is alive. Ask by identity, not by a number you remember: pgrep -af "<the thing it watches>". When a check is cheap and its subject is specific, matching the subject is never harder than remembering the handle.

Read the full file on GitHub · 488 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. 3d ago First seen · 488 lines · 101 tokens per session scan C 4738e91309a4

Subscribe to this mod's changes

evidence-discipline is a skill published in the GitHub repository tatargabor/set-core (35 stars, last pushed today), licensed MIT. It adds 101 tokens to every session and 7,977 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-06.