ワンクリックでManusで任意のスキルを実行

agent-hard-constraint-author

スター2

フォーク0

更新日2026年6月15日 08:53

Turn validated agent-learning findings into human-approved executable constraints such as PHPStan rules, PHP-CS-Fixer rules, tests, or CI checks, with generation packages, manifests, fixtures, and validation evidence.

インストール

Codex または Claude でインストールこの Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。

Manusで実行

ソース

voku

voku/agent-learning

GitHub リポジトリを開く Creator のリポジトリを見る

ダウンロード

Manusで実行

Agent Hard Constraint Author

Use this skill when repeated learning should become an executable check. The point is to move stable behavior from prose into a rule that can fail validation.

Fast Path

Start from validated findings. Require several independent findings or one clearly documented critical incident.
Define the objective violation, scope, allowed boundaries, detectability, false-positive risk, validation commands, target rule path, registration files, and example rule paths.
Export a generation package before writing code:

vendor/bin/agent-learning constraint-export \
  --root infra/doc/agent-learning \
  --proposal proposal.2026-06-15.001

Generate or implement the rule in the consuming project, not in this package, unless the rule is package-generic.
Include valid, invalid, boundary, and false-positive fixtures where the engine supports fixtures.
Register the rule in the consuming project's analyzer configuration.
Validate both directions: the invalid fixture must fail with the intended rule identifier, and the valid fixture plus real target files must pass.
Record validation evidence and content hashes before marking the proposal applied.
Activate the manifest only after implementation and validation:

vendor/bin/agent-learning constraint-loop \
  --root infra/doc/agent-learning \
  proposal.2026-06-15.001 \
  --by agent \
  --commit working-tree \
  --validation infra/doc/agent-learning/validation-results/proposal.2026-06-15.001.json

Quality Bar

Prefer PHPStan or PHP-CS-Fixer when the violation is statically detectable.
Keep rule identifiers stable and engine-valid.
Make selection scope explicit; recall uses scope overlap, not semantic similarity.
Do not activate generated constraints silently.
Do not accept "file exists" or exit code alone as proof; inspect generated artifacts and fixture behavior.

このリポジトリの他の Skills

同じリポジトリ

agent-learning-ctx-evidence

voku/agent-learning

Use ctx-backed local agent-history search as bounded evidence references while creating, validating, or consolidating voku/agent-learning findings.

2026-07-062

agent-learning-consumer

voku/agent-learning

Set up and operate voku/agent-learning in a consuming repository, including learning-root config.json, findings, proposals, consolidation prompts, validation, and lifecycle commands. Use for end-user learning-loop workflows, not package source development.

2026-06-182

agent-learning-maintainer

voku/agent-learning

Maintain voku/agent-learning package source, tests, docs, CLI behavior, configurable paths, constraint lifecycle code, and local vendor syncs in consuming projects.

2026-06-182

project-agent-guidance-maintenance

voku/agent-learning

Maintain repository-owned agent skills, memory, docs, and installed copies after reviewed durable learnings.

2026-06-182

project-agent-learning

voku/agent-learning

Capture reviewable agent-learning findings for this repository after meaningful sessions.

2026-06-182

project-context-rule

voku/agent-learning

Turn stable repository guidance into local executable constraints such as static-analysis rules, tests, or CI checks.

2026-06-182

Agent Hard Constraint Author

Use this skill when repeated learning should become an executable check. The point is to move stable behavior from prose into a rule that can fail validation.

Fast Path

Start from validated findings. Require several independent findings or one clearly documented critical incident.

Define the objective violation, scope, allowed boundaries, detectability, false-positive risk, validation commands, target rule path, registration files, and example rule paths.

Export a generation package before writing code:

vendor/bin/agent-learning constraint-export \ --root infra/doc/agent-learning \ --proposal proposal.2026-06-15.001

Generate or implement the rule in the consuming project, not in this package, unless the rule is package-generic.

Include valid, invalid, boundary, and false-positive fixtures where the engine supports fixtures.

Validate both directions: the invalid fixture must fail with the intended rule identifier, and the valid fixture plus real target files must pass.

Record validation evidence and content hashes before marking the proposal applied.

Activate the manifest only after implementation and validation:

vendor/bin/agent-learning constraint-loop \ --root infra/doc/agent-learning \ proposal.2026-06-15.001 \ --by agent \ --commit working-tree \ --validation infra/doc/agent-learning/validation-results/proposal.2026-06-15.001.json

Quality Bar

Prefer PHPStan or PHP-CS-Fixer when the violation is statically detectable.

Keep rule identifiers stable and engine-valid.

Make selection scope explicit; recall uses scope overlap, not semantic similarity.

Do not activate generated constraints silently.

Do not accept "file exists" or exit code alone as proof; inspect generated artifacts and fixture behavior.

name	agent-hard-constraint-author
description	Turn validated agent-learning findings into human-approved executable constraints such as PHPStan rules, PHP-CS-Fixer rules, tests, or CI checks, with generation packages, manifests, fixtures, and validation evidence.