strata:eval

A local test for measuring whether changes to search and result ranking find the right notes. It uses a committed set of example searches and expected results.

In plain words
What is it for?
Use it before and after changing search, reranking, result weighting, or filters. It reports recall@k, which measures whether the expected result appears near the top, and MRR, which measures how high the first expected result appears.
Why use it?
It shows whether search improved or regressed using numbers instead of guesswork. It can also show whether reranking adds enough benefit to justify using it.

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/gideondk/strata/eval
Any agent
npx skills add gideondk/strata --skill eval
Clone the repo
git clone --depth 1 https://github.com/gideondk/strata

Made for: Claude Code, Codex.

Per session 85 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 465 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 $0.00085 $0.00465
Opus 5 $0.00043 $0.00233
Sonnet 5 $0.00017 $0.00093
Haiku 4.5 $0.00009 $0.00047

Measured 2d ago against content hash d69bcde2a135, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

strata: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 2d 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.

skills/eval/SKILL.md · 55 lines

What it actually says

strata:eval

A regression guard + measuring stick for vault recall. It runs a small, committed set of query → expected notes cases through the same retrieval the recall tool uses, and reports recall@k and MRR.

The golden set

Lives at <vault>/<repo>/.eval/golden.json (commit it — it's versioned with the vault):

{
  "cases": [
    {"query": "rate limiting policy",
     "expected": ["decisions/2026-05-21-token-bucket.md"],
     "scope": "decisions"}
  ]
}

scope is optional (null/omitted = all scopes). 20–50 hand-picked cases is plenty. Seed them from queries you actually run, or from the usage ledger's top-recalled notes.

Run it

"${CLAUDE_PLUGIN_ROOT}/bin/strata" eval -k 5

Measure the rerank lift

Compare the pipeline with and without the cross-encoder rerank:

"${CLAUDE_PLUGIN_ROOT}/bin/strata" eval -k 5 --sweep

--sweep runs the golden set rerank-OFF then rerank-ON and prints both rows + the lift, so the decision is a number. If the lift is zero (or negative) on your set, leave rerank off (it's off by default) — don't pay the per-call model load for no gain.

When to run

  • Before/after any retrieval change (rerank, RRF weighting, a new scope).
  • Periodically, as a regression guard — a drop means recall quality slipped.
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. 2d ago First seen · 55 lines · 85 tokens per session scan A d69bcde2a135

Subscribe to this mod's changes

strata:eval is a skill published in the GitHub repository gideondk/strata (5 stars, last pushed 1mo ago), licensed MPL-2.0. It adds 85 tokens to every session and 465 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 2026-08-31.

Related

Other skills, from other repositories

my-wiki

Manage local OKF-compatible Markdown My Wiki vaults with an AI agent. Use for capturing webpages, PDFs, Office documents, notes, images, folders, and ZIP bundles as References; maintaining Reference-to-Concept evidence links; searching or answering from a vault; checking or repairing vault health; switching among…

NimaChu/my-wiki · 78 tokens

save-to-favorites

The single complete Pickoo workflow for new Xiaohongshu, WeChat, Douyin, Bilibili, Xiaoyuzhou, podcast, music, article/blog, or general web links. Use for 收藏、保存、收录、抓取、归档、拆解、分析、OCR、ASR or 完整处理 from Codex, WorkBuddy, Hermes, Claude, or Claudian. It owns intake and the complete downstream workflow; obsolete standalone…

KristenPan/pickoo-your-favorites · 110 tokens

asset-review-agent

Independently review source-extracted quote, viewpoint, hook, and structure candidates. Image and shot assets are outside automatic review and enter only after explicit user interaction.

KristenPan/pickoo-your-favorites · 37 tokens

phase-orchestrator

Orchestrates phase-based implementation plans using the host's todo tracker and subagents for implementation, review, validation, default phase commits, plan finalization, and a final end-to-end review. Delegates work through the companion phase-implementer, phase-reviewer, and phase-final-reviewer skills and consumes…

hraness/kb · 101 tokens

answer-processing

Use whenever the user uploads a hand-written or scanned answer PDF to be graded against a reference solution. Converts answer PDFs in answers/.pdf to markdown in answers/converted/.md using the pdf skill (OCR as needed), then performs strategy-based grading against converted/solutions/.md or quizzes/answers.md.…

OPTIMETA/PAIDEIA · 79 tokens

course-builder

Use whenever the user wants to ingest a new course's materials (lecture notes, textbook chapters, HW problems, HW solutions) and build the course-specific knowledge base — patterns.md (recurring solution techniques), coverage.md (HW-to-section map with 🔥 exam tiers + ⚠weak flags), and summary.md (topic tree). Invoked…

OPTIMETA/PAIDEIA · 119 tokens