eval-regression

eval-regression is a skill for Claude Code from hoangsonww/AI-News-Briefing. It costs 58 tokens per session (403 once invoked), scanned A, original, MIT.

A regression check that re-evaluates every pinned golden example in an evaluation set and compares the results with stored baselines.

In plain words
What is it for?
Use it after updating evaluation prompts or judge models, or before merging changes, to identify examples whose score has dropped beyond the allowed tolerance.
Why use it?
It detects when a prompt, judging model, or judge instructions make quality worse after a change.

Skill for Claude Code

Written for Claude Code: shipped in a Claude Code plugin.

Needs its repository: it runs a file that does not travel with it, so clone the repository first. The line is python3 eval/runner.py regression --judge claude --regression-drop 0.5.

Part of the ai-news-briefing plugin — 11 skills, 3 agents, 1 hook, 2 MCP servers shipped together

Good fit Use it after updating evaluation prompts or judge models, or before merging changes, to identify examples whose score has dropped beyond the allowed tolerance.

Compare 6 skills from other repositories ↓
Install

Getting it into your agent

It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.

Clone the repo
git clone --depth 1 https://github.com/hoangsonww/AI-News-Briefing
agentmods
npx agentmods add skills/hoangsonww/ai-news-briefing/eval-regression

Made for: Claude Code.

Or install ai-news-briefing, the plugin that ships this one along with the rest of its 11 skills, 3 agents, 1 hook, 2 MCP servers.

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 eval-regression

README.md
[![agentmods](https://agentmods.dev/badge/skills/hoangsonww/ai-news-briefing/eval-regression/github.svg)](https://agentmods.dev/skills/hoangsonww/ai-news-briefing/eval-regression)
Your own site
<a href="https://agentmods.dev/skills/hoangsonww/ai-news-briefing/eval-regression"><img src="https://agentmods.dev/badge/skills/hoangsonww/ai-news-briefing/eval-regression/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 eval-regression

Your own site · 80×15
<a href="https://agentmods.dev/skills/hoangsonww/ai-news-briefing/eval-regression"><img src="https://agentmods.dev/badge/skills/hoangsonww/ai-news-briefing/eval-regression.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 58 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 403 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe. Third-party audits
  • NVIDIA SkillSpector pass 7 Sept 2026
How audits are shown
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 $0.00058 $0.00403
Opus 5 $0.00029 $0.00201
Sonnet 5 $0.00012 $0.00081
Haiku 4.5 $0.00006 $0.00040

Measured 12d ago against content hash c8a629a73eb8, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-12, from the pricing page.

Security

Grade A, and why

eval-regression 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 12d 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.

claude-plugins/ai-news-briefing/skills/eval-regression/SKILL.md · 40 lines

What it actually says

Eval — Regression Gate

Guard against silent quality regressions when the daily prompt, judge model, or judge prompt changes.

How to invoke

make eval-regression                        # stub judge, fast smoke test
make eval-regression JUDGE=claude           # real Claude Haiku judge — production gate

Direct invocation supports a tighter (or looser) drop tolerance:

python3 eval/runner.py regression --judge claude --regression-drop 0.5

Behavior

  1. Read every eval/golden/*.json baseline (currently 18 cards, real-judge composites 2.9–4.2).
  2. Re-judge each card against the configured backend.
  3. Compute delta = new_composite - baseline_composite.
  4. Fail (exit 2) if any card's delta is below -regression-drop (default -0.5).
  5. Print per-card OK / REGRESSED with explicit deltas.

When to run

  • After bumping PROMPT_VERSION in eval/judge.py or editing eval/judge_prompt.md.
  • After switching judge models (EVAL_JUDGE_MODEL).
  • In CI before merging changes to the daily prompt.md.
  • After a re-baselining workflow to confirm the new goldens hold.

What to tell the user

If regressions are flagged, list each offending card with its delta and the judge's notes — they explain why the score dropped. If no regressions, report total cards passed and the worst-case delta so the user knows how much headroom remains.

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. 12d ago First seen · 40 lines · 58 tokens per session scan A c8a629a73eb8

Subscribe to this mod's changes

eval-regression is a skill published in the GitHub repository hoangsonww/AI-News-Briefing (42 stars, last pushed 4d ago), licensed MIT. It adds 58 tokens to every session and 403 once invoked, about $0.0003 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.