一键在 Manus 中运行任何 Skill

piensalo-verify

星标0

分支0

更新时间2026年7月12日 23:11

Independent verification of a candidate answer against domain verifier criteria: deterministic checks first, adversarial probes, evidence-level grading, disqualifier sweep, per-criterion verdict table. Use when asked to check, grade, review, or attack a finished answer or artifact - yours or another agent's - before it is trusted or shipped. Trigger phrases: 'verify this', 'is this answer right', 'review before we ship', 'grade this output'.

安装

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

在 Manus 中运行

来源

ralfyishere

ralfyishere/piensalo

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

piensalo-verify

The verification workflow of Piénsalo. Given a task and a candidate answer, grade the candidate - do not fix it. Verification is adversarial: your job is to find where it fails.

Primary workflow

Pick the domain verifier - software, mathematics, research, strategy, invention, writing, or generic if none fits. Criteria for every domain are in references/verification.md.
Disqualifier sweep first - check the domain's disqualifiers; any hit ends the review with verdict FAIL and the quoted evidence.
Run deterministic criteria - execute or mechanically check everything checkable (run tests, recompute numbers, diff scope). Quote actual output; a prediction is not a result.
Run adversarial criteria - construct probes the author visibly did not try (empty/boundary/hostile inputs, counterexamples, contradicting sources); predict the outcome before executing each probe.
Grade evidence criteria - for judgment-level criteria, classify the candidate's support as verified fact, inference, assumption, or guess; unsupported load-bearing claims fail the criterion.
Verdict table - one row per criterion: criterion | kind | evidence | verdict. Overall verdict fails if any disqualifying criterion fails. State what could not be checked here and why.

Conditional moves

When a deterministic criterion needs tools you lack, mark it NOT-RUNNABLE-HERE - never infer a pass.
When the candidate spans domains, apply generic criteria plus each touched domain's disqualifiers.

Output contract

Per-criterion verdict table, disqualifier findings quoted verbatim, list of unchecked criteria with reasons, and an overall PASS/FAIL/NOT-FULLY-CHECKABLE verdict.

Delivery notes (small-model packets)

If delegating steps to a smaller model: one bounded objective per packet, all inputs named explicitly, facts separated from instructions, uncertainty marked 'UNCERTAIN: ...' rather than invented away.

Before answering

Check the draft against references/failure-checks.md (the failure-mode catalog) and references/verification.md (domain verifier criteria). Repair procedures live in references/contracts.md.

同仓库更多 Skills

同仓库

audience-constraint-check

ralfyishere/piensalo

Check a deliverable against its actual consumer: who reads it, what they do next, what format and register they need, and what they will assert downstream on its basis. Activate when the deliverable has a named audience beyond the requester, or its text will be forwarded, announced, or acted on by someone absent from the conversation.

2026-07-120

boundary-case-check

ralfyishere/piensalo

For any rule-like deliverable — code, regex, query, formula, validator, policy — enumerate the boundary inputs (empty, zero, one, max, duplicate, malformed, hostile) and actually exercise the most breakable three before shipping. Activate when the deliverable is executable or rule-like and will meet inputs the author didn't type.

2026-07-120

calibrated-uncertainty

ralfyishere/piensalo

State every claim at its actual evidence level — verified, inferred, assumed, guessed — and give credences only where an observable could score them; ban certainty language on unverified claims. Activate when a response makes claims about unobserved, future, or unverified state, or when the user asks how confident or likely something is.

2026-07-120

causal-model-construction

ralfyishere/piensalo

Before intervening in a system, write the explicit causal chain from the proposed action to the intended outcome — every arrow marked verified or assumed — and check the weakest assumed arrow first. Activate when an intervention (fix, optimization, policy change, growth action) is proposed on a system whose mechanism has not been written down.

2026-07-120

cheapest-discriminating-test

ralfyishere/piensalo

With multiple live hypotheses and a limited test budget, design the cheapest single test whose possible outcomes separate the hypotheses — with a results-to-verdict table committed before running anything. Activate when two or more competing hypotheses are on the table and the next step is choosing what test, experiment, or probe to run.

2026-07-120

complete-the-delivery

ralfyishere/piensalo

Enumerate every deliverable the request actually asked for and map each one to the work produced, so nothing asked-for is silently dropped, stubbed, or rounded up to done. Activate when a request contains multiple parts — an enumerated list, several 'and'/'also' clauses, or a spec with more than one required artifact — or before delivering a draft against such a request.

2026-07-120

name	piensalo-verify
description	Independent verification of a candidate answer against domain verifier criteria: deterministic checks first, adversarial probes, evidence-level grading, disqualifier sweep, per-criterion verdict table. Use when asked to check, grade, review, or attack a finished answer or artifact - yours or another agent's - before it is trusted or shipped. Trigger phrases: 'verify this', 'is this answer right', 'review before we ship', 'grade this output'.
license	MIT

piensalo-verify

The verification workflow of Piénsalo. Given a task and a candidate answer, grade the candidate - do not fix it. Verification is adversarial: your job is to find where it fails.

Primary workflow

Pick the domain verifier - software, mathematics, research, strategy, invention, writing, or generic if none fits. Criteria for every domain are in references/verification.md.
Disqualifier sweep first - check the domain's disqualifiers; any hit ends the review with verdict FAIL and the quoted evidence.
Run deterministic criteria - execute or mechanically check everything checkable (run tests, recompute numbers, diff scope). Quote actual output; a prediction is not a result.
Run adversarial criteria - construct probes the author visibly did not try (empty/boundary/hostile inputs, counterexamples, contradicting sources); predict the outcome before executing each probe.
Grade evidence criteria - for judgment-level criteria, classify the candidate's support as verified fact, inference, assumption, or guess; unsupported load-bearing claims fail the criterion.
Verdict table - one row per criterion: criterion | kind | evidence | verdict. Overall verdict fails if any disqualifying criterion fails. State what could not be checked here and why.

Conditional moves

When a deterministic criterion needs tools you lack, mark it NOT-RUNNABLE-HERE - never infer a pass.
When the candidate spans domains, apply generic criteria plus each touched domain's disqualifiers.

Output contract

Per-criterion verdict table, disqualifier findings quoted verbatim, list of unchecked criteria with reasons, and an overall PASS/FAIL/NOT-FULLY-CHECKABLE verdict.

Delivery notes (small-model packets)

Before answering

Check the draft against references/failure-checks.md (the failure-mode catalog) and references/verification.md (domain verifier criteria). Repair procedures live in references/contracts.md.