A scoring-quality checker that compares a vacancy or company scorer's decisions with your own labelled examples. A golden set is a small collection of examples marked as acceptable or unacceptable; precision and recall measure how accurately the scorer handles them.

Use it to test vacancy fit decisions or company keep/reject decisions in either the PostgreSQL or SQLite version of the pipeline.

It shows whether the current scoring instructions match your judgement and lists every disagreement to review. The examples stay on your machine in a gitignored folder.

How ncalavera/llm-job-pipeline configures Claude Code

Reuse

Borrowing it

Nothing to install: this file belongs to ncalavera/llm-job-pipeline. 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/ncalavera/llm-job-pipeline/main/.claude/commands/jobs-eval.md

Clone the repo

git clone --depth 1 https://github.com/ncalavera/llm-job-pipeline

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.

README.md

[![agentmods](https://agentmods.dev/badge/commands/ncalavera/llm-job-pipeline/jobs-eval/github.svg)](https://agentmods.dev/commands/ncalavera/llm-job-pipeline/jobs-eval)

Your own site

<a href="https://agentmods.dev/commands/ncalavera/llm-job-pipeline/jobs-eval"><img src="https://agentmods.dev/badge/commands/ncalavera/llm-job-pipeline/jobs-eval/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.

Your own site · 80×15

<a href="https://agentmods.dev/commands/ncalavera/llm-job-pipeline/jobs-eval"><img src="https://agentmods.dev/badge/commands/ncalavera/llm-job-pipeline/jobs-eval.svg" alt="Reviewed on agentmods" width="80" height="20"></a>

Per session 56 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.

When invoked 1,959 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.

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.

Model	Per session	Once invoked
Fable 5.1	$0.00056	$0.01959
Opus 5.5	$0.00022	$0.00784
Sonnet 5.5	$0.00011	$0.00392
Haiku 4.5	$0.00006	$0.00196

Measured 2mo ago against content hash c9d0ba7a229e, method: parsed. Prices are Anthropic first-party input rates as of 2026-10-07, from the pricing page.

Security

Grade A, and why

jobs-eval 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 2mo 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/commands/jobs-eval.md · 172 lines

How it starts

The opening of the file, as written. The whole thing — 172 lines — stays where its author put it; the contents beside it link to each section on GitHub.

/jobs-eval

Does the scorer actually agree with you? This command answers that with one number. It builds a golden set — a few dozen of your own vacancies each labelled fit or nofit with a short reason — scores them with the live scripts/prompts/vacancy-scoring.md, and reports how often the model's verdict matches your label, plus precision/recall at the score threshold and a readable list of every disagreement.

Method: Hamel Husain's "Critique Shadowing" (https://hamel.dev/blog/posts/llm-judge/). Cheap, honest, no external service.

The golden set is personal data (real titles, orgs, your reasons). It lives in evals/ (gitignored) and is never committed. Works the same in full mode (Postgres) and simple mode (SQLite) — it reads through the same data layer.

Everything is one script: python3 scripts/golden_set.py <subcommand>.

There are two surfaces to check, each with its own personal golden set:

vacancy (default) — the vacancy-scoring prompt vs your like/pass verdicts.
company — the company-scoring (WANT) prompt vs your keep/reject verdicts on companies. Add --kind company to any subcommand below.

1. Build the set

Fast path — seed from your verdicts. Every like/pass you have already made is a label: a liked-basket status means fit, a passed/skipped status means nofit. Seed straight from that history (read-only):

python3 scripts/golden_set.py seed --limit 50

Positives are usually scarce, so seed keeps ALL of your fit verdicts and fills the rest with recent nofit ones — a set with no positives can't yield a meaningful precision/recall. Probable auto-expired passes (a passed role whose deadline had already lapsed) are excluded, since those aren't fit judgements. Re-running seed later only appends new verdicts; it never rewrites old labels.

Blind path — hand-label fresh vacancies. For coverage the verdict history lacks, label some vacancies WITHOUT seeing their score:

python3 scripts/golden_set.py template --limit 25   # writes evals/label_template.jsonl (no scores shown)
# edit each line: set "label" to fit or nofit, add a one-line "reason"
python3 scripts/golden_set.py add-template          # appends the ones you labelled

Read the full file on GitHub · 172 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.

2mo ago Changed c9d0ba7a229e
2mo ago First seen · 172 lines · 56 tokens per session scan A 9782549f98df

Subscribe to this mod's changes

jobs-eval is a command published in the GitHub repository ncalavera/llm-job-pipeline (2 stars, last pushed 16d ago), licensed MIT. It adds 56 tokens to every session and 1,959 once invoked, about $0.0002 per session on Opus 5.5. 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.

Analytics and metrics

jobs-eval next to 6 commands from other repositories
Mod	Type	Stars	Last push	Licence	Scan	Tokens	Works with
jobs-evalncalavera/llm-job-pipeline	Command	2	16d ago	MIT	A	56	Claude Code
checklistgithub/spec-kit	Command	148k	yesterday	MIT	A	13	Claude Code
clarifygithub/spec-kit	Command	148k	yesterday	MIT	A	27	Claude Code
specifygithub/spec-kit	Command	148k	yesterday	MIT	A	13	Claude Code
analyzegithub/spec-kit	Command	148k	yesterday	MIT	A	26	Claude Code
convergegithub/spec-kit	Command	148k	yesterday	MIT	A	35	Claude Code
implementgithub/spec-kit	Command	148k	yesterday	MIT	A	14	Claude Code

llm-job-pipeline: Command for Claude Code