aiworker-validate

name	aiworker-validate
description	Route AIWorker validation to the correct mode: remote fleet, local source worker, local published CLI, or remote Coder Claude Code. Use for AIWorker validation planning or execution requests.
disable-model-invocation	true
argument-hint	fleet-remote | worker-source-local | cli-release-local | coder-claude-code
arguments	["mode","target"]

AIWorker Validate

Use this as the canonical AIWorker validation entrypoint. This skill is manual only because every mode can touch real services, local credentials, long-running processes, or real executor sessions.

If $mode is provided, use it. Otherwise infer exactly one mode from the user request:

Mode	Load this reference	Scope
fleet-remote	references/fleet-remote.md	Shared remote fleet, gateway, Fleet Web UI, enrollment, or a temporary local worker attached to the fleet.
worker-source-local	references/worker-source-local.md	Local worker validation where the product-under-test is the current repo checkout, workspace command, or built source bundle.
cli-release-local	references/cli-release-local.md	Local black-box worker validation where the product-under-test is a published @zonease/aiworker-cli package.
coder-claude-code	references/coder-claude-code.md	Remote Coder workspace validation with a published CLI and Claude Code executor.

Routing Rules

• Do not combine modes unless the user explicitly asks for a combined campaign.
• Keep fleet/gateway/enrollment work in fleet-remote.
• Keep source-tree worker testing in worker-source-local.
• Keep published-package black-box testing in cli-release-local.
• Keep remote Coder workspace testing in coder-claude-code.
• Use $target as the version, server hint, debug root, or project path when it is supplied. If it is missing, discover it from the repo or ask only when a safe default does not exist.

Legacy aiworker-test*, aiworker-release-debug, and aiworker-coder-claude-engine skills were removed. Use this skill directly.

Always-On Safety

• Preserve the real user HOME for Codex / Claude Code executor tests so existing auth and sandbox config remain available.
• Isolate AIWorker state with AIWORKER_HOME, worker DB paths, data roots, logs, pidfiles, debug roots, and project directories.
• Never write bearer tokens, bootstrap tokens, master keys, cookies, auth headers, private URLs, generated credentials, or secret-like prompt content into repo docs, skill files, changelogs, final answers, screenshots, or persisted reports.
• Prefer tmux for long-running servers. If unavailable, use setsid or nohup with explicit pidfile/logfile cleanup.
• Never use broad cleanup such as kill $(lsof -ti:PORT). Match only confirmed listeners with lsof -tiTCP:PORT -sTCP:LISTEN.
• Verify stateful claims with DB, filesystem, event-stream, or service evidence; do not trust LLM self-report for admission, memory, artifact, cron, or persisted state.
• Record confirmed product issues through PMA as sanitized BUG/TODO/QA tasks. Do not fix code unless the user expands scope.

Completion Checklist

Before reporting done:

• the selected mode is explicit;
• the relevant reference file was followed;
• commands, pass/fail status, evidence paths, cleanup, and skips are recorded;
• sensitive values are redacted or omitted;
• out-of-scope surfaces are named instead of silently tested;
• skill/frontmatter validation and git diff --check pass when files changed.