Skip to content
Skills Library
AI & agents · Testing

agentic-eval

/agentic-eval

Evaluate and improve AI workflow outputs with small goldens, rubrics, and repeatable checks; use when tuning skills, prompts, instructions, or agent behavior.

Author
ll7 ↗
Licence
GPL-3.0
Last repository change
23.10.2026

Assessment and security

90/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content25 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
1
Scripts
0
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 23.10.2026, without references to retired models or APIs.

Licence

GPL-3.0, otwarta.

What it does and when to use it

Evaluate and improve AI workflow outputs with small goldens, rubrics, and repeatable checks; use when tuning skills, prompts, instructions, or agent behavior.

Original author description

Evaluate and improve AI workflow outputs with small goldens, rubrics, and repeatable checks; use when tuning skills, prompts, instructions, or agent behavior.

How it works

  1. When to use
  2. Read First
  3. Workflow
  4. Proof and Guardrails
  5. Output
  6. Guardrails
Original SKILL.md headings · 60 lines, 265 words
SKILL.md preview
---
name: agentic-eval
description: Evaluate and improve AI workflow outputs with small goldens, rubrics, and repeatable checks;
  use when tuning skills, prompts, instructions, or agent behavior.
category: research-iteration
kind: atomic
phase: analysis
requires_write: true
requires_slurm: false
requires_benchmark_artifacts: false
delegates_to: []
output_schema: skill_run_summary.v1
---

# Agentic Eval

## When to use

Use this skill when the target is an AI workflow artifact (skills, prompts, rubrics, issue text,
review guidance, etc.) rather than core product code.

## Read First

- `AGENTS.md`
- `docs/dev_guide.md`
- `docs/ai/awesome_copilot_adaptation.md`
- `docs/code_review.md`

## Workflow

1. Define the artifact and objective
   - State the artifact, expected behavior, and what a pass/fail result looks like.
   - Set a small representative evaluation set.

2. Build a rubric
   - Use 3–6 concrete criteria.
   - Mark critical failures separately from quality preferences.

3. Create a baseline
   - Score the current artifact against the rubric before editing.

4. Improve one dimension at a time
   - Make a targeted change.
   - Re-run the same evaluation.
   - Keep the change only if improvement is clear and reproducible.

5. Capture proof
   - Compare baseline vs post-change with the same rubric and commands.
   - Keep the result compact: what improved, what regressed, what remains unknown.

## Proof and Guardrails

- Always preserve baseline and post-change results.
- Reuse the same corpus for all iterations.
- If changes drift from the target behavior, revert and document the false regression.
- Prefer `autoresearch` when the task has repeated measurable benchmarks.
- If benchmark evidence is required, treat failed/invalid runs as inconclusive, not proof.

## Output

Report:

- artifact inspected,
- rubric and dataset,
- baseline and final score,
- keep/discard decision with rationale,
- open questions.
## Guardrails

- Stay within the skill scope declared in `.agents/skills/skills.yaml`.
- Prefer repository scripts and canonical docs before ad-hoc commands.
- Record blockers and validation gaps instead of overstating completion.

Versions

RepositoryScoreLicence
This page · ll7/robot_sf_ll790GPL-3.0
forgivesam168/ai-dev-workflow100Apache-2.0
github/awesome-copilot98MIT
PracticalSwan/agent-skills98MIT
sandbaseai/sandbase-skills90Apache-2.0

Database snapshot: 13.11.2026 · package version f5bec30fe9a5 (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100skills-eval
    Evaluate Claude skill quality through auditing. Use when reviewing or auditing skills.
  • 100design-agent-eval-workflow
    Design evaluation workflows for agent, skill, prompt, and automation behavior before implementation. Use when choosing eval cases, graders, thresholds, regression runs, full-auto gates, or escalation points for Codex skills, repo-local agents, OpenAI Agents SDK services,…
  • 100edit-skills
    Conventions for agent instructions and skills in this repository. Use when the user asks to add, edit, or audit a skill or agent instructions.
  • 100eval
    Evaluates proposed skill, agent or workflow changes against recorded failures and unchanged behavior before adoption. Use when asked to evaluate agent behavior, collect failure causes or improve haechi from evidence; not for ordinary application tests or performance measurements.
  • 100latent-space-engineering
    Shapes agent behavior via instruction framing and style transfer. Use when composing dispatch prompts or writing skill instructions for parallel review agents.
  • 100skill-eval-improve
    Improves Agent Skills via validate → rule-based eval cases → plugin-eval → prompt evals → bounded edits with held-out gates. Use when tuning skill quality, routing, or adopting Chrome/Microsoft T-named quality gates—not for bulk validate-only or SkillOpt automation.

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.