deciding-who-decides
Decides who holds authority for a change — the agent at the edge or a human gate — by matching decision rights to reversibility, evidence, and consequence, and names the escalation trigger. Use when an agent could act on something irreversible, trust-bearing, or thinly evidenced. Do not use for trivial reversible edits, or to justify skipping a required human approval.
FlyFission · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add FlyFission/nuclear-grade-context-engineering --skill deciding-who-decides技能说明
name: deciding-who-decides description: Decides who holds authority for a change — the agent at the edge or a human gate — by matching decision rights to reversibility, evidence, and consequence, and names the escalation trigger. Use when an agent could act on something irreversible, trust-bearing, or thinly evidenced. Do not use for trivial reversible edits, or to justify skipping a required human approval.
Deciding Who Decides
Overview
Authority should sit where the policy, evidence, competence, and effective ability to intervene support it, not automatically at the top or with the actor holding the tool. This skill creates an evidence-conditioned authority record for one bounded decision episode. It separates prepare, recommend, verify, validate, verdict, accept, apply, reopen, and close rights; links evidentiary rights to exact Evidence IDs; names the authority that may assemble each right's basis; preserves unknown and disputed states; and names transfer or blocking triggers. It does not infer authorization from technical capability, repository activity, fluency, or nominal human presence.
Decision contract
- Claim checked: each decision right is assigned under an explicit local policy to a named actor; each right's declared basis and basis authority are visible; required Evidence IDs and their custody profile are visible; unresolved evidence cannot silently clear agent application; and transfer/reopen/close triggers are concrete.
- Artifact observed: the bounded action;
verification.mdEvidence IDs, coded source states, intended-use/V&V status, and coupling profile; declared policy identity, authority, custody, digest, and validity; effective intervention capability; and any existing human gate ->decision-authority.mdplus a declaration-relative structural result. - Decision affected: block -- does the declared record structurally permit agent application, require authority transfer or separate/dual control, or remain blocked/prohibited/indeterminate?
- Failure class: misplaced-authority or unsupported-basis (capability treated as authorization, unresolved/self-check evidence clearing agent action, prohibited role overlap, stale basis, or skipped gate).
- Next action: record raw observations, derive the declaration-relative result, and transfer or block before side effects when the basis or authority is unresolved.
When to Use
- An agent is about to act and it is unclear whether it may decide alone or must ask first.
- A change is irreversible, touches users, data, credentials, dependencies, agent authority, or a release.
- The evidence behind a decision is thin, contested, or rests only on the agent's own confidence.
- You are setting up an agent's standing authority and want explicit decision rights and escalation thresholds.
When Not to Use
- The edit is trivial and reversible with obvious proof and no new trust boundary.
- A required human approval already exists; this skill never exists to talk past it.
- An incident is live and you must stabilize first (use
responding-to-incidents).
Inputs
- The proposed action, its target, and whether it can be undone.
- The exact Evidence IDs in
verification.md, their coded source states, intended use, and custody/coupling profile. - The controlling policy ID/version, policy authority, policy custodian, policy digest, and validity or non-expiry basis.
- The consequence if it is wrong, and who is affected.
- The agent's granted authority, effective stop/reversal capability, and any standing rule or human gate that already applies.
- The actors proposed for prepare, recommend, verify, validate, verdict, accept, apply, reopen, and close.
Process
- Bound the decision episode: identify the exact candidate/action, controlling policy version, policy authority, policy custodian, policy digest, a non-expired UTC validity or concrete non-expiry basis, and whether the action is reversible. Validator-host time is not a trusted clock.
- Code source-linked evidence states from the record:
observed,bounded_absence,unknown, ordisputed. These are interpreted record states, not sensor-raw facts. Never convert missing telemetry into absence. - Link the intended-use/V&V status and evidence-custody/coupling profile from
verification.md. Do not infer substantive truth or independence from structure alone. - Allocate every decision right: prepare, recommend, verify, validate, verdict, accept, apply, reopen, and close. Name the proposed actor, evidence-basis authority, policy gate, required authority, and transfer trigger. Evidence-basis authority is mandatory and cannot be
not applicable. Recommend, verify, validate, verdict, accept, and close require decisive Evidence IDs. Prepare, apply, and reopen may usenot applicableonly for the Evidence IDs when the policy/trigger basis is explicit; a structurally clearable agent apply still requires decisive evidence. - Allocate
agent_permitted,human_required,separate_control_required,dual_authority_required,blocked_pending_evidence, orprohibited_for_agent. Derive one declaration-relative result:agent_apply_structurally_clearable, one of the five non-agent paths, orpolicy_result_indeterminate. Indeterminate never clears application. - Block agent application when decisive evidence is unknown, disputed, or classified as self-check. A nominal human approval does not cure absent evidence, absent authority, or absent intervention capability.
- Record reopen, supersession, interim-expiry, and closure controls. A materially changed action or evidence basis requires a new or successor episode; do not silently reuse prior acceptance.
- Run
ng validate <packet> --strict-authority; use--strict-custodyas well when the packet must require both activated records.
Outputs
- A completed
decision-authority.mdfor the bounded episode. - Explicit allocations for all nine decision rights, their basis authority,