New: 98,000+ skills indexed from GitHub repos with 100+ stars · every skill safety-checkedBrowse top skillsTop skills
Esc to closeBrowse all skills
Submit a skillBrowse skills
Submit a skillBrowse skills

Agent skill

Kayba Stage 4 Rubric

by kayba-ai in kayba-ai/agentic-context-engine

Organize computed metrics into a tiered evaluation rubric with leading, lagging, and quality indicators.

Apache-2.0Auto-check passedEducation

Install Kayba Stage 4 Rubric

skills CLI
$ npx skills add kayba-ai/agentic-context-engine --skill kayba-stage-4-rubric -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install kayba-ai/agentic-context-engine kayba-stage-4-rubric --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/kayba-ai/agentic-context-engine.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.claude/skills/kayba-pipeline/stage-4-rubric .claude/skills/kayba-stage-4-rubric && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
kayba-stage-4-rubric
GitHub stars
2.6k
Token cost
~2.2k tokens
SKILL.md length
836 words
Files
1
Skills in repo
8
Repo updated
First seen
Licence
Apache-2.0

At a glance

Organize computed metrics into a tiered evaluation rubric with leading, lagging, and quality indicators.

  • The user says run stage 4
  • SKILL.md covers Inputs, Process and Outputs
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md
  • Invoked by the kayba-pipeline orchestrator

What it does

Kayba Stage 4 Rubric is an agent skill from kayba-ai/agentic-context-engine. Organize computed metrics into a tiered evaluation rubric with leading, lagging, and quality indicators. Trigger when the user says "run stage 4", "build rubric", "tier metrics", or when invoked by the kayba-pipeline orchestrator. Requires eval/baselinemetrics.json and eval/computebaselines.py to exist.

Its SKILL.md is about 2.2k tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts.

It sits in Education, covering Quizzes and assessments. The repository describes itself as: 🧠 Make your agents learn from experience. Now available as a hosted solution at kayba.ai. The licence is Apache-2.0.

When your agent uses it

  • The user says run stage 4
  • Invoked by the kayba-pipeline orchestrator

Example prompts

  • “run stage 4”
  • “build rubric”
  • “tier metrics”
  • “/kayba-stage-4-rubric”

What it can do on your machine

Read from SKILL.md and the folder at commit 3a31983. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    No scripts in the folder and no shell commands in SKILL.md (its code samples are markdown).

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Kayba Stage 4 Rubric loads about 2.2k tokens when it runs. Until then it costs about 82 tokens; SKILL.md has 836 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~82
When it runs · the whole SKILL.md, loaded when a task matches
~2.2k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from kayba-ai/agentic-context-engine at commit 3a31983, republished under its Apache-2.0 licence (© kayba-ai). 836 words, ~2,170 tokens.

Download SKILL.mdSave it as .claude/skills/kayba-stage-4-rubric/SKILL.md (or your agent's skills folder).
name
kayba-stage-4-rubric
description
Organize computed metrics into a tiered evaluation rubric with leading, lagging, and quality indicators. Trigger when the user says "run stage 4", "build rubric", "tier metrics", or when invoked by the kayba-pipeline orchestrator. Requires eval/baseline_metrics.json and eval/compute_baselines.py to exist.

Stage 4: Rubric Definition

Organize metrics into a tiered evaluation rubric. Detect and resolve redundancy quantitatively. Ensure every insight is accounted for.

Inputs

  • eval/baseline_metrics.json — computed baseline values from Stage 3
  • eval/compute_baselines.py — to understand what each metric measures
  • eval/stage1_insights_summary.md — the original insights
  • eval/stage2_domain_context.md — domain context

Read all four files before starting.

Process

1. Quantitative redundancy check

Before tiering, check every pair of metrics for overlap. Two metrics are redundancy candidates if ANY of the following hold:

  • Denominator overlap >70%: compute |denom_events(A) ∩ denom_events(B)| / min(|denom(A)|, |denom(B)|). If >0.70, they are candidates. To compute this, trace through the detector functions in compute_baselines.py and determine which trace events (turns, calls, threads) each denominator iterates over. When denominators are identical sets (same loop, same filter), overlap is 100%.
  • Same skill set: the metrics map to the exact same set of insight/skill IDs from Stage 1.
  • Logical subsumption: one metric's positive case is a strict subset of the other's (e.g., "turn has exactly 1 tool call" is a subset of "turn has no user-facing content alongside tool calls" only if every single-call turn also has no content — check this, don't assume it).

For each candidate pair, make an explicit decision with reasoning:

PairDenom overlapSkill overlapSubsumption?DecisionReasoning
M1/M2100% (same 29 turns)identicalNo — can violate one without the otherKeep bothIndependently actionable: batching vs. content leaking are distinct fixes

Valid decisions: keep both (with reasoning why they're independently actionable), merge (combine into one metric, specify how), or drop (specify which and why). "They feel different" is not sufficient reasoning — cite the specific behavior that one catches and the other misses.

Final count target: 5-7 metrics after redundancy resolution.

2. Tier each metric

Use this decision flowchart for every metric:

Q1: Can a SINGLE skill/instruction change directly move this metric?
    → If the agent follows one new instruction and the metric improves,
      regardless of other behaviors: LEADING.

Q2: Does moving this metric require MULTIPLE skills to be adopted together?
    → If improvement depends on several upstream behaviors all working
      (e.g., proper turn structure + confirmation flow + execution):
      LAGGING.

Q3: Does moving this metric require domain reasoning beyond following instructions?
    → If the agent needs to correctly interpret policy rules, evaluate
      eligibility criteria, or make judgment calls that can't be reduced
      to a single instruction: QUALITY.

Apply the flowchart to each metric and record the Q1/Q2/Q3 answer that determined the tier. If a metric could arguably be two tiers, pick the lower one (Leading < Lagging < Quality) and note the ambiguity.

Tier summary for reference:

TierPurposeMoves when...Diagnostic signal
LeadingBehaviors a single skill directly changesSkill is adoptedIf leading moves but lagging doesn't → skill adopted but not solving the right problem
LaggingAggregate outcomes requiring multiple skillsMultiple skills coordinateIf lagging moves but leading doesn't → something else improved, not your skills
QualityRequires domain understanding, not just instruction-followingAgent reasons correctlyIf quality moves but lagging doesn't → agent got lucky or metric is mis-tiered
3. Flag low-confidence baselines

Any metric with denominator < 5 events is a low-confidence baseline. These metrics:

  • ARE included in the rubric (they measure real behaviors)
  • Are marked with **Confidence: low** (n=X) in the rubric
  • Must NOT drive priority decisions in Stage 5 — they inform direction only
  • Should note what denominator size would make them reliable (rule of thumb: n >= 10 for a rate metric to be meaningful, n >= 30 for statistical comparisons)
Show full SKILL.md (365 more words)Show less
4. Set direction

For each metric, indicate whether it should go up higher or down lower. Don't set arbitrary numerical targets — baseline + direction is enough.

Ceiling guard: If a metric's baseline is already 100%, its direction MUST be "↑ maintain" or "— already optimal", never "↑" as if it needs to go higher. A 100% metric is at ceiling — the goal is to sustain it, not improve it. Similarly, if a metric is at 0% and the desired direction is "↓", mark it "↓ maintain" or "— already at floor". Do not let any downstream stage (Stage 5 action plan, Stage 7 fixes) list a ceiling/floor metric as needing improvement.

5. Map insights to metrics (completeness check)

For every insight from eval/stage1_insights_summary.md, assign it to one of three categories:

  1. Mapped — directly linked to one or more metrics. List which ones.
  2. Indirectly mapped — supports a metric but isn't the primary driver. List the metric and explain the indirect relationship.
  3. Qualitative-only — no programmatic metric captures this insight. Explicitly mark it and state why (e.g., "requires LLM-as-judge," "measures explanation quality," "efficiency pattern with no clear denominator").

Every insight MUST appear in exactly one category. If you find an insight that should have a metric but doesn't, note it as a gap for future Stage 3 iterations — but do not invent metrics at this stage.

At the end, report:

  • X / N insights mapped to metrics
  • Y / N insights indirectly mapped
  • Z / N insights qualitative-only
6. Add invalidation notes

For each metric, write one sentence answering: "What would make this tier assignment wrong?"

Examples:

  • M1 (Leading): "Wrong if fixing batching also requires the agent to change its confirmation flow — that would make it Lagging."
  • M5 (Quality): "Wrong if cancellation compliance can be fixed by a single checklist instruction without requiring the agent to reason about policy — that would make it Leading."

These notes exist so Stage 5 can catch tier errors. If Stage 5 finds evidence that a tier is wrong (e.g., a single skill would move a "Quality" metric), it should flag the conflict rather than silently inheriting the error.

7. Write the rubric

Write to eval/baseline_metrics.md:

markdown
# Eval Rubric — Baseline Metrics

## Summary
| # | Metric | Tier | Baseline | Direction | Confidence |
|---|--------|------|----------|-----------|------------|
| M1 | First-call success rate | Leading | 37.6% | up | ok (n=29) |
| M2 | ... | ... | ... | ... | ... |

## Tier Definitions

- **Leading** — Single skill directly moves this. Should change first after deployment.
- **Lagging** — Multiple skills must coordinate. Improves as a consequence of adoption.
- **Quality** — Requires domain reasoning beyond instruction-following. Hardest to move.

## Metric Details

### M1: [name]
**Tier:** Leading
**Baseline:** 37.6% (685 / 1,821)
**Confidence:** ok (n=1821) | low (n=X) — needs n>=Y for reliable comparison
**Direction:** up higher is better
**What it measures:** [description]
**How it's computed:** [reference to function in compute_baselines.py]
**Skills that should move this:** [list insight/skill IDs from stage 1]
**Tier rationale:** [which flowchart question determined the tier]
**Invalidation note:** [what would make this tier wrong]

### M2: [name]
...

## Redundancy Analysis

| Pair | Denom overlap | Skill overlap | Subsumption? | Decision | Reasoning |
|------|---------------|---------------|--------------|----------|-----------|
| ... | ... | ... | ... | ... | ... |

## Insight Coverage

### Mapped (X / N)
- `insight_id` — [title] → M1, M3

### Indirectly mapped (Y / N)
- `insight_id` — [title] → supports M5 via [explanation]

### Qualitative-only (Z / N)
- `insight_id` — [title] — [why no metric: e.g., "requires LLM-as-judge"]

Outputs

  • eval/baseline_metrics.md — human-readable tiered rubric with redundancy analysis, confidence flags, insight coverage, and invalidation notes

© kayba-ai, Apache-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in .claude/skills/kayba-pipeline/stage-4-rubric of kayba-ai/agentic-context-engine.

Open the folder on GitHubat commit 3a31983

Compare with similar skills

Kayba Stage 4 Rubric next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Kayba Stage 4 Rubric compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Kayba Stage 4 Rubric this skillkayba-ai/agentic-context-engine2.6k—~2.2kAutomated safety check: PassApache-2.0
DeepTutor CLIHKUDS/DeepTutor41k—~2.8kAutomated safety check: PassApache-2.0
AI Engineering Placement Quizrohitg00/ai-engineering-from-scratch67k—~2kAutomated safety check: PassMIT
Codebase to Coursezarazhangrui/codebase-to-course5.7k—~4.4kAutomated safety check: PassNone
AI Engineering Phase Quizrohitg00/ai-engineering-from-scratch67k—~2.1kAutomated safety check: PassMIT
Scholar EvaluationK-Dense-AI/claude-scientific-writer2.4k2 repos~2.9kAutomated safety check: NotesMIT

Similar skills

  • DeepTutor CLI

    HKUDS/DeepTutor

    Teaches the agent to set up and run DeepTutor from the command line: chat and capabilities, knowledge bases, partners, memory, sessions, notebooks and the server or Web app.

    41k GitHub stars~2.8k tokensUpdated 3 days ago
    EducationAuto-check passed

  • AI Engineering Placement Quiz

    rohitg00/ai-engineering-from-scratch

    Runs a 10-question quiz across five areas to place a learner in the AI Engineering from Scratch curriculum, so they skip what they already know.

    67k GitHub stars~2k tokensUpdated today
    EducationAuto-check passed

  • Codebase to Course

    zarazhangrui/codebase-to-course

    Turns a codebase into an interactive single-page HTML course for non-technical learners, with scroll modules, animated diagrams, quizzes and plain-English code translations.

    5.7k GitHub stars~4.4k tokensUpdated 6 mo ago
    EducationAuto-check passed

  • AI Engineering Phase Quiz

    rohitg00/ai-engineering-from-scratch

    Quizzes you on a completed phase of the AI Engineering from Scratch course, taking a phase number or name and mapping it to that phase's directory.

    67k GitHub stars~2.1k tokensUpdated today
    EducationAuto-check passed

  • Scholar Evaluation

    K-Dense-AI/claude-scientific-writer

    Provide qualitative-first, evidence-traceable developmental review of scholarly works and audit low-stakes research-assessment rubrics with optional local quality controls.

    2.4k GitHub starsUsed in 2 repos~2.9k tokens
    EducationAuto-check: notes

  • Evaluation

    guanyang/open-agent-hub

    This skill should be used when building agent evaluation systems: deterministic checks, regression suites, multi-dimensional rubrics, quality gates, production monitoring, baseline comparison, and…

    977 GitHub starsUsed in 2 repos~4.2k tokens
    EducationAuto-check passed

More from kayba-ai/agentic-context-engine

All 8 skills in this repo

  • Kayba Pipeline

    kayba-ai/agentic-context-engine

    End-to-end agent evaluation and improvement pipeline. An agent skill from kayba-ai/agentic-context-engine.

    2.6k GitHub stars~1.4k tokensUpdated 17 days ago
    Auto-check passed

  • Kayba Stage 1 API Analysis

    kayba-ai/agentic-context-engine

    Fetch pre-computed insights from the Kayba API and build a structured summary.

    2.6k GitHub stars~1.1k tokensUpdated 17 days ago
    Auto-check passed

  • Kayba Stage 2 Domain Context

    kayba-ai/agentic-context-engine

    Gather domain context about the repository and agent — system prompt, tool definitions, domain docs, and behavior patterns from traces.

    2.6k GitHub stars~1.9k tokensUpdated 17 days ago
    Auto-check passed

  • Kayba Stage 3 Metrics

    kayba-ai/agentic-context-engine

    Define metrics from Kayba insights, implement them as Python measurement code, run against traces, and iterate until the metrics are clean and meaningful.

    2.6k GitHub stars~3.4k tokensUpdated 17 days ago
    Auto-check passed

  • Kayba Stage 5 Action Plan

    kayba-ai/agentic-context-engine

    Triage each insight into discard/code-fix/prompt-fix and produce a prioritized action plan with specific recommendations.

    2.6k GitHub stars~2.9k tokensUpdated 17 days ago
    Auto-check passed

  • Kayba Stage 6 Hitl

    kayba-ai/agentic-context-engine

    Human-In-The-Loop gate that presents the action plan with full context, collects an informed approval/modification/rejection decision, and records the outcome.

    2.6k GitHub stars~2.6k tokensUpdated 17 days ago
    Auto-check passed

Categories

Questions about Kayba Stage 4 Rubric

What does Kayba Stage 4 Rubric do?

Organize computed metrics into a tiered evaluation rubric with leading, lagging, and quality indicators. Kayba Stage 4 Rubric is an agent skill from kayba-ai/agentic-context-engine. Organize computed metrics into a tiered evaluation rubric with leading, lagging, and quality indicators.

When should I use Kayba Stage 4 Rubric?

Kayba Stage 4 Rubric fits situations like: the user says run stage 4; invoked by the kayba-pipeline orchestrator.

How do I install Kayba Stage 4 Rubric in Claude Code?

Run `npx skills add kayba-ai/agentic-context-engine --skill kayba-stage-4-rubric -a claude-code`. Or copy the skill folder (.claude/skills/kayba-pipeline/stage-4-rubric in kayba-ai/agentic-context-engine) into .claude/skills/kayba-stage-4-rubric in your project. Claude Code loads it when a task matches its description.

How do I install Kayba Stage 4 Rubric in Codex?

Run `npx skills add kayba-ai/agentic-context-engine --skill kayba-stage-4-rubric -a codex`. Or copy the skill folder (.claude/skills/kayba-pipeline/stage-4-rubric in kayba-ai/agentic-context-engine) into .agents/skills/kayba-stage-4-rubric in your project. Codex loads it when a task matches its description.

Can I use Kayba Stage 4 Rubric in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add kayba-ai/agentic-context-engine --skill kayba-stage-4-rubric -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/kayba-stage-4-rubric, .gemini/skills/kayba-stage-4-rubric, .github/skills/kayba-stage-4-rubric and .opencode/skills/kayba-stage-4-rubric in your project.

What does Kayba Stage 4 Rubric need to run?

SKILL.md names no scripts, command-line tools or credentials: Kayba Stage 4 Rubric is instructions for the agent only.

Does Kayba Stage 4 Rubric access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Kayba Stage 4 Rubric safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Kayba Stage 4 Rubric use?

Kayba Stage 4 Rubric is published under the Apache-2.0 licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Kayba Stage 4 Rubric use?

About 2.2k tokens (SKILL.md is roughly 8.7k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Kayba Stage 4 Rubric?

Skills that share tags, products or a category with Kayba Stage 4 Rubric: DeepTutor CLI (HKUDS/DeepTutor, 41k stars), AI Engineering Placement Quiz (rohitg00/ai-engineering-from-scratch, 67k stars), Codebase to Course (zarazhangrui/codebase-to-course, 5.7k stars) and AI Engineering Phase Quiz (rohitg00/ai-engineering-from-scratch, 67k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Kayba Stage 4 Rubric?

kayba-ai (a GitHub organization) maintains it in kayba-ai/agentic-context-engine, which has 2,590 GitHub stars. The repository holds 8 skills in this directory. The repository was last updated on September 24, 2026.

Source: kayba-ai/agentic-context-engine on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.