aoa-eval-select

Intent

Use this skill to find whether an existing eval, validator, test, script, report, or local-port artifact already covers the current pressure before designing a new eval.

Trigger boundary

Use this skill when:

• a task asks which eval should be used for a change, failure, or regression
• a repository has a local evals/ port and the agent needs to inspect it

before writing anything new

• central aoa-evals may already contain the relevant proof pattern or local

port contract

• existing validators, tests, or scripts may be the honest eval surface

Do not use this skill when:

• an already selected eval simply needs to be run; use aoa-eval-apply
• no eval exists and the next move is a local intake packet; use

aoa-eval-local-need

• the user asks for a new suite design after selection has already failed; use

aoa-eval-design

• source authority rather than eval coverage is the problem

Inputs

• target repo and touched paths
• local evals/PORT.yaml, evals/README.md, evals/intake/, and eval reports
• scripts, tests, validators, schemas, generated manifests, and route docs
• central aoa-evals docs and local-port standard
• MCP list/inspect packets when available

Outputs

• selected existing eval, validator, test, script, or report surface
• reason the surface fits or does not fit
• owner and proof-authority classification
• next route: apply, local need, design, or stop

Procedure

1. read the target repo route law and local evals/PORT.yaml when present
2. inventory local evals/, tests, validators, and scripts around the touched

paths

1. inspect central aoa-evals only for reusable proof doctrine, local-port

standard, or central bundle matches

1. prefer deterministic local scripts and tests for coding-agent behavior when

they already constrain the outcome

1. use MCP list/inspect only to narrow the search; verify important claims

against source files

1. classify the result as exact fit, partial fit, nearest wrong target, no fit,

or blocked by stale/missing owner evidence

1. hand off to aoa-eval-apply, aoa-eval-local-need, or aoa-eval-design

only after selection is clear

Contracts

• existing local proof surfaces beat new design pressure
• central proof doctrine may guide but not absorb repo-local intake
• selection reports confidence and nearest wrong target instead of guessing
• generated indexes and MCP packets are read models until source-verified

Risks and anti-patterns

• skipping local validators because a central eval sounds more official
• selecting a test that only checks incidental behavior
• turning a partial fit into a proof verdict
• reading all repos when touched paths already narrow the search

Verification

• confirm local port presence, absence, and status
• confirm selected surface and owner route
• confirm why nearest alternatives were rejected
• confirm next route is one of apply, local need, design, or stop
• confirm central aoa-evals was not rewritten

Technique traceability

Manifest-backed techniques:

• AOA-T-0003 from 8Dionysus/aoa-techniques at 1a7d146957108ecefc24219c7d56357c5a4a2c2c using path techniques/proof/evaluation-chain/contract-first-smoke-summary/TECHNIQUE.md and sections: Intent, Inputs, Outputs, Core procedure, Contracts, Risks, Validation
• AOA-T-0076 from 8Dionysus/aoa-techniques at 1a7d146957108ecefc24219c7d56357c5a4a2c2c using path techniques/governance/decision-routing/owner-layer-triage/TECHNIQUE.md and sections: Intent, Inputs, Outputs, Core procedure, Contracts, Risks, Validation
• AOA-T-0094 from 8Dionysus/aoa-techniques at 1a7d146957108ecefc24219c7d56357c5a4a2c2c using path techniques/proof/owner-truth-closeout/canonical-owner-with-validated-mirror/TECHNIQUE.md and sections: Intent, Inputs, Outputs, Core procedure, Contracts, Risks, Validation

Adaptation points

• local eval-port paths
• repo-local validation command names
• central proof bundle naming