harnessloop-delegationlisted

# Harnessloop Delegation Check whether Harnessloop can safely rely on subagents, swarm, or another delegated execution mechanism. This skill is a control-plane check; it does not perform business execution or accept a round. ## Input Contract Accept an explicit skill invocation such as `$harnessloop-delegation`, `$harnessloop-delegation check`, or `$harnessloop-delegation check handoff <path>`. Treat `harnessloop:delegation` as a natural-language alias only; `$harnessloop:delegation` is not a valid skill invocation. Useful input includes: - `target-project`: defaults to the current working directory. - `task-type`: read-only discovery, evidence collection, low-context execution, adversarial review, acceptance testing, or unknown. - `delegation-mechanism`: Codex subagent, Claude Code swarm, Claude Code subagent, external worker, manual handoff, or unknown. - `expected-model` and `expected-effort`: from `setup/cost-context-policy.md` or user instruction. - `observed-model` and `observed-effort`: only when directly reported by the delegated mechanism or reliable metadata. - `handoff-path`: delegated handoff file, if checking an existing or completed handoff. - `scope`: read-only, write-bounded, review-only, or unknown. - `verification-method`: how model, effort, scope boundaries, output path, and evidence citations can be checked. If any required fact is missing, ask a focused question. Do not infer model, effort, tool identity, delegation mechanism, or success from surrou