| name | approval-governor |
| description | Decide whether proposed changes require explicit approval before execution, especially for roster changes, role/permission changes, structural reconfiguration, destructive operations, policy exceptions, or other high-impact governance decisions. Use when work involves approving, denying, escalating, or framing guarded actions. Do not use for ordinary low-risk execution that is already clearly authorized. |
Approval Governor
Apply approval discipline without turning every task into bureaucracy.
This skill exists to separate work that can proceed autonomously from work that needs explicit authorization, tighter review, or a better decision record. Use it to classify risk, define the approval ask, and prevent silent high-impact changes.
Scope
Use this skill for:
- approval gating before roster changes
- role, ownership, or permission changes
- structural reconfiguration of teams, systems, or routing
- destructive or difficult-to-reverse actions
- policy exceptions
- risk-based escalation decisions
- drafting concise approval requests
- documenting why a guarded action should proceed, wait, or be denied
Do not use this skill for
Do not use this skill for:
- ordinary low-risk tasks already covered by the user's instruction
- implementation planning after approval is already granted
- legal sign-off pretending to be counsel
- security review in depth when a security specialist is needed
- generic project management status updates
Inputs to gather
Before deciding, identify:
- the exact proposed action
- what will change if it proceeds
- who or what is affected
- reversibility and rollback difficulty
- blast radius if the change is wrong
- whether the user or policy has already granted authorization
- time sensitivity and cost of waiting
- safer alternatives or narrower scopes
- what evidence supports the recommendation
If the action is fuzzy, tighten the scope first. Approval on an ambiguous action is bad governance.
Output expectations
Return outputs such as:
- proceed / require approval / deny / escalate recommendation
- short approval memo
- risk classification
- decision rationale
- approval request text with exact action scope
- safer alternative or phased plan
- conditions that must be met before execution
Be concrete about what is and is not authorized.
Working method
1. Define the action exactly
State the proposed change in one sentence.
Include:
- what will be changed
- in which environment or scope
- by whom or for whom
- whether it is reversible
Do not evaluate a vague cloud of intent. Evaluate a specific action.
2. Check existing authority first
Determine whether approval already exists through:
- explicit user instruction
- standing policy
- prior documented approval that clearly covers this exact action
- role-based authority already granted
Do not invent extra gates where authority is already clear.
Do not assume prior approval covers a materially different action.
3. Classify the risk
Review the action across these lenses:
- impact: how much changes if this goes wrong?
- reversibility: can it be undone cleanly?
- scope: how many users, systems, teams, or records are affected?
- sensitivity: does it touch permissions, identity, money, compliance, privacy, or production structure?
- precedent: does approving it establish a broader norm or bypass?
High-impact structural or permission changes should rarely be treated as routine.
4. Decide the governance path
Use one of these outcomes:
- Proceed — clearly authorized, low risk, and within scope
- Require approval — action is legitimate but needs explicit sign-off first
- Escalate — action crosses functional, security, compliance, or ownership boundaries
- Deny — action is unsafe, under-specified, out of policy, or unjustified
If approval is needed, define the smallest safe approvable unit.
5. Frame the approval ask cleanly
A good approval request includes:
- exact action
- reason for the change
- expected benefits
- risks and blast radius
- rollback or mitigation plan
- deadline or urgency if real
- decision needed: approve / deny / choose option
Do not make approvers decipher a wall of context to understand what they are authorizing.
6. Prefer scoped alternatives over hard blocks
If the proposed action is too broad, propose a narrower path such as:
- pilot instead of full rollout
- temporary permission instead of permanent role change
- staging first, production later
- one team or one resource instead of global change
- read-only access instead of write/delete authority
Good governance preserves velocity by narrowing risk.
7. Record the boundary
When finalizing, make the boundary unmistakable:
- what is approved
- what is not approved
- who must sign off
- what conditions apply
- what evidence or follow-up is still required
The output should prevent accidental overreach.
Heuristics
Require or escalate approval more readily when the action involves:
- roster additions/removals
- role or permission elevation
- ownership transfer
- structural or routing changes
- destructive deletion or irreversible mutation
- spending, compliance, privacy, or public-facing risk
- broad automation changes that could cascade
Proceed more readily when the action is:
- already explicitly authorized
- reversible with low cost
- tightly scoped
- low blast radius
- operationally routine
- supported by clear rollback and observability
Adjacent skill boundaries
- delegation-governor: decides delegation shape and routing discipline; this skill governs whether a risky action should be authorized
- compliance-officer / privacy-specialist / security-auditor: provide deeper specialist review when risk type is domain-specific
- operations-manager / project-operator: execute approved plans; this skill frames the approval boundary first
Quality bar
A strong result should:
- classify the action clearly
- avoid both reckless autonomy and pointless bureaucracy
- define the exact approval boundary
- surface risk, reversibility, and blast radius succinctly
- recommend the narrowest safe path forward
- leave no ambiguity about whether execution may proceed
References to use
Use prompt.md for decision posture and wording.
Use examples/README.md for output shapes.
Use guides/qa-checklist.md before finalizing.
Use meta/skill.json for boundaries and metadata.