evaluate

A separate-context evaluator that checks completed work against the acceptance criteria in a plan. TDD, or test-driven development, is not required; the evaluator can use different review levels based on risk.

In plain words
What is it for?
Use it during evaluation to run a lightweight review for ordinary changes or a deeper review for architecture, security, data-loss risk, or public API changes.
Why use it?
It reduces the chance that the person or agent who built the work will approve its own mistakes or overlook unmet requirements.

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/harness-mini/harness-mini/evaluate
Any agent
npx skills add harness-mini/harness-mini --skill evaluate
Clone the repo
git clone --depth 1 https://github.com/harness-mini/harness-mini

Made for: Claude Code, Codex.

Per session 76 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 943 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 $0.00076 $0.00943
Opus 5 $0.00038 $0.00472
Sonnet 5 $0.00015 $0.00189
Haiku 4.5 $0.00008 $0.00094

Measured 2d ago against content hash e8220eae51c2, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

evaluate 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.

.claude/skills/evaluate/SKILL.md · 72 lines

How it starts

The opening of the file, as written. The whole thing — 72 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are grading work against its plan's acceptance criteria — never against vague satisfaction. Agents confidently praise their own work; evaluation exists to break that.

Pick the tier (record it in the plan's eval: field)

"Always spawn an Opus evaluator" is too expensive for every change — if the default is too heavy, people bypass it, and then there is no firewall. So scale rigor with risk. Default is L1.

Tier Who grades Use for Cost
L0 builder attaches evidence; main agent runs a compact checklist tiny, low-risk changes whose criteria are 100% machine-checkable (docs, glue, a one-line fix) ~free
L1 (default) an independent lightweight reviewer — a fresh context (haiku sub-agent, or a separate prompt/thread on a CLI without sub-agents) normal slices low
L2 the full evaluator agent (Opus / strongest available) cross-slice work, architecture, security, data-loss risk, public-API changes, release gates high

When in doubt, go up a tier. L0 is only honest when no criterion needs judgment — the gate is the passing tests, not the builder's opinion. If a criterion needs a human/agent call, it is at least L1.

The firewall is the separate context, not the tooling. No sub-agents available? Run L1/L2 as a fresh prompt/thread with only the criteria + the diff (see docs/codex-getting-started.md / docs/cursor-getting-started.md).

Procedure (every tier)

  1. Read the plan's acceptance criteria (from to-prd). They are the contract.
  2. Verify by interaction, not by reading. Run the tests. Run the app/endpoint where possible. A criterion passes only with evidence you produced — never the builder's claim. (At L0 the "evidence" is the attached test output the main agent re-runs and confirms.)
  3. For each criterion emit PASS/FAIL + the evidence + (on FAIL) the smallest concrete gap.
  4. L1/L2: apply clean-code as a secondary lens — cite specific violations.

Read the full file on GitHub · 72 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. 2d ago First seen · 72 lines · 76 tokens per session scan A e8220eae51c2

Subscribe to this mod's changes

evaluate is a skill published in the GitHub repository harness-mini/harness-mini (2 stars, last pushed 21d ago), licensed MIT. It adds 76 tokens to every session and 943 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.