eval-and-improve

A coding-agent workflow for running a repository’s evaluation suite and investigating failed test cases. It can address problems in the agent instructions or correct an evaluation case when its check is wrong.

In plain words
What is it for?
Use it to run `python -m evals`, check for regressions, inspect expected tool calls, and fix red evaluation cases.
Why use it?
It helps turn a failing evaluation run into a diagnosis of each failure and a passing suite, while checking both tool usage and results.

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/agno-agi/context/eval-and-improve
Any agent
npx skills add agno-agi/context --skill eval-and-improve
Clone the repo
git clone --depth 1 https://github.com/agno-agi/context

Made for: Claude Code, Codex.

Per session 115 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,634 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 $0.00115 $0.02634
Opus 5 $0.00057 $0.01317
Sonnet 5 $0.00023 $0.00527
Haiku 4.5 $0.00012 $0.00263

Measured yesterday against content hash f7e0954c9023, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

eval-and-improve 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 yesterday.

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.

.agents/skills/eval-and-improve/SKILL.md · 155 lines

How it starts

The opening of the file, as written. The whole thing — 155 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Eval and Improve

Coding-agent workflow — a /slash-command your coding agent (Claude Code, Codex, …) runs while developing this repo. Not a runtime skill the deployed @context agent runs; those live in skills/.

You're running @context's eval suite, diagnosing every failure, fixing what's in scope, and stopping when all cases pass. Surface area is two files: evals/cases.py (declares cases) and evals/__main__.py (runner). A case applies up to four checks, deterministic ones first (they're the spine; the judge corroborates):

  • structural — a zero-arg callable returning (passed, detail); when set, the agent is not run. Used by boundary_is_structural to assert the guest/owner toolset asymmetry with no model in the loop. Deterministic.
  • expected_tool_calls — agno's ReliabilityEval asserts which tools fired. Deterministic.
  • capture_only — for guest runs, asserts every tool that fired is on the capture-only allowlist (no read/act tool, checked at the trace level). Deterministic.
  • criteria — agno's AgentAsJudgeEval (LLM rubric, binary pass/fail), optionally narrowed with judge_guidelines. Keep it decisive so it doesn't flake.

No custom DSL beyond those fields on the Case dataclass.

0. Preconditions

  • Postgres reachable on 5432: nc -z localhost 5432 returns 0. If not, docker compose up -d context-db from the source repo. (docker compose ps is unreliable from worktrees or alternate clones.)
  • Venv active: source .venv/bin/activate. If .venv doesn't exist (fresh checkout or worktree), run ./scripts/venv_setup.sh first. evals/cases.py imports the agents directly from agents/, so no AgentOS server has to be running.
  • .env populated with OPENAI_API_KEY. evals/__main__.py calls evals.dotenv.load_dotenv() at startup, so you do not need to source .env first. Worktrees don't inherit .env (it's gitignored) — copy it from the source repo if missing.

Read the full file on GitHub · 155 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.

  1. yesterday First seen · 155 lines · 115 tokens per session scan A f7e0954c9023

Subscribe to this mod's changes

eval-and-improve is a skill published in the GitHub repository agno-agi/context (11 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 115 tokens to every session and 2,634 once invoked, about $0.0006 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.

Related

Other skills, from other repositories

verification

Prove that a coding task is actually complete. Use this after meaningful code changes, when tests/builds fail or are skipped, before marking a plan or goal complete, and whenever acceptance depends on runtime, security, recovery, performance, or cross-module evidence.

ageerle/ruoyi-ai · 54 tokens

repository-investigation

Investigate an unfamiliar repository before changing it. Use this whenever a coding task spans multiple modules, asks for architecture or root-cause analysis, names behavior whose implementation location is unknown, or risks editing before enough evidence is gathered.

ageerle/ruoyi-ai · 49 tokens

safe-refactoring

Execute behavior-preserving or intentionally scoped refactors safely. Use this for multi-file renames, component/service extraction, state-management changes, API migrations, concurrency refactors, or any request where unrelated user work and subtle contracts must be preserved.

ageerle/ruoyi-ai · 52 tokens

wegent-knowledge

Knowledge base management and search tools for Wegent. Provides capabilities to list, create, update, and search knowledge bases and documents using RAG retrieval. Use this skill when the user wants to manage knowledge bases, documents, or search for information programmatically.

wecode-ai/Wegent · 51 tokens

quantmind-dev

Contributor workflow for the QuantMind codebase. Covers contributor setup (environment + hooks), filing issues, commit format, pull request format, and component development across quantmind/ modules (etl, knowledge, configs, preprocess, rag, flows, mind, utils) with tests, examples, and verification. Use when setting…

LLMQuant/quant-mind · 90 tokens

notion

Notion workspace integration for searching pages, managing databases, creating postmortems, and exporting RCA findings.

Arvo-AI/aurora · 23 tokens