agentsclimarketplace

Aoa verification

Skill 8Dionysus/aoa-skills/skills/core/engineering/aoa-verification

Bounded agent-facing skill bundles for coding agents and humans: reviewable execution workflows composed from reusable techniques.

Install
npx -y skills add 8Dionysus/aoa-skills --skill aoa-verification

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Turn one stable technical truth into bounded evidence through contract, coverage-audit, or property mode. Use to define a named consumer seam, audit what existing checks really constrain, or express a broad invariant after manual cases establish meaning. Compose with aoa-eval when an evaluation surface must first be found or applied. Do not use for undefined behavior, generic test plans, or green checks as whole-system proof.

SKILL.md

4.1 KB, as published. Nobody here has run it

aoa-verification

Intent

Start from owner meaning and observed behavior, then choose the smallest evidence form whose green result has an explicit claim limit.

Trigger boundary

Use this skill when:

  • a stable producer-consumer seam, an important invariant, or a property across many inputs/states needs evidence that can reveal a meaningful break

Do not use this skill when:

  • behavior or ownership is undefined, the change is private, the request is a generic test strategy, or automation would exist only to produce green status

Inputs

  • owner rule, system/boundary under review, named consumers where relevant
  • manual cases or failures, current checks, and available oracle

Outputs

  • exactly one typed mode result: evidence package, gaps, smallest authorized durable check or no-check decision, claim limit, and termination

Procedure

  1. Read references/contract.yaml and choose exactly one mode:

    ModeSelect whenRequired procedure
    contractA named consumer relies on a stable ABI, receipt, schema, tool result, or handoff.references/contract.md
    coverage-auditChecks exist and the question is what stable invariant they truly constrain.references/coverage-audit.md
    propertyCorrectness must hold across many inputs or states.references/property.md

  2. Read the selected reference completely. Do not load unrelated mode procedures.

  3. Read the authoritative owner rule before the subject implementation, checks, examples, or generated views. Use exact supplied paths directly; search only for a missing required input, not for ritual workspace orientation. Exercise expected, rejected, and motivating failure cases manually and state the oracle. Do not collect repository-wide inventories, hashes, or status unless the claim or effect boundary needs them.

  4. When the exact evaluation surface is unknown or must be selected/applied, use a task-local DAG: aoa-eval.select -> aoa-eval.apply -> the chosen verification mode. A named check may be run directly as evidence inside a verification task; cross-surface discovery and application remain aoa-eval responsibilities.

  5. Create durable automation only after manual evidence establishes a repeated or owner-declared long-lived rule and the active task authorizes the write. Remove session-only probes after learning.

Contracts

  • manual observation and owner meaning precede durable automation
  • evaluation selection/application and interpretation of what evidence proves remain distinct, composable effects
  • a check protects one claim and cannot certify the whole system
  • generated/exported views remain subordinate to owner sources
  • no-check is valid when recurrence, oracle, or stability is missing

Risks and anti-patterns

  • restating a fixture, snapshot, formatting, or field order as semantic law
  • property tests with a self-fulfilling oracle
  • adding validators after the fact to hide an unresolved manual failure
  • silently using verification as a replacement for eval inventory selection or treating a selected eval as proof of invariant coverage

Verification

  • confirm selected mode, owner rule, manual cases, and oracle
  • when an eval surface was discovered or applied, name the aoa-eval handoff and preserve its observation/proof limit
  • if automation is proposed, show the motivating failure it catches
  • make claim limit, evidence gap, consumer impact, and stop line explicit

Adaptation points

Owners supply local commands, generators, schemas, fixtures, consumers, and compatibility windows without duplicating this procedure.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.