Safe Action Menu¶
Template — instantiates Agentic Control Loop Design
A fixed template of pre-approved, in-bounds actions for a high-risk setting, with an escalate path for anything the menu does not cover.
The Safe Action Menu is a fixed, pre-approved list of the moves an actor may make on their own in a high-risk setting — and, just as importantly, an explicit escalate path for everything not on the list. Its defining move is bounded choice by enumeration: rather than trusting the actor to judge, in the moment, what is safe, it settles that judgment in advance as a menu — these actions are yours to take without asking; anything else, you stop and escalate. It preserves real agency (the actor still chooses among the listed moves under live conditions) while making the boundary of legitimate action concrete and un-improvised. It is a snapshot of what is permitted now; it does not decide which listed move to take, nor does it schedule how the menu grows over time.
Example¶
On an offshore oil platform, a control-room operator watches separator pressure begin to climb. A Safe Action Menu for pressure anomalies sits at the console. In-bounds, act now: reduce inlet flow by up to 15%, open the designated relief path, increase cooling, notify the shift supervisor. Off-menu, escalate immediately: any action that would take a well offline, any manual override of a safety interlock, anything not listed.
The pressure keeps rising past what the menu's actions arrest. Here the menu earns its keep twice. First, the operator has already taken the four permitted actions without waiting for permission — bounded autonomy in a time-critical moment. Second, the next thing that might help (tripping an interlock) is explicitly off-menu, so instead of an operator improvising a dangerous novel action under stress, the escalate path fires and a supervisor with broader authority is looped in. The menu did not choose which action to try first or promise the anomaly would resolve; it drew the line between what the operator owns and what must go up.
How it works¶
- Enumerate the safe set in advance. Decide, calmly and ahead of time, which actions are low-enough risk to be taken without asking, so the high-stress moment is a lookup, not a judgment call.[n1]
- Make the boundary exhaustive by default. The menu is a whitelist: if an action is not listed, it is off-limits, which removes the grey zone where "I assumed I could" lives.
- Pair every menu with an escalate path. For anything the menu does not cover, name explicitly who to stop and go to; a menu without an escalate path just traps the actor when the situation exceeds it.
- Keep it short enough to use under pressure. A menu no one can scan in the moment it is needed is not a control; brevity is a safety property, not a convenience.
Tuning parameters¶
- Menu breadth — how many actions are pre-approved. A wider menu speeds response and signals trust but enlarges what can go wrong unsupervised; a narrower menu is safer but escalates more, slowing the actor and loading the escalation channel.
- Boundary strictness — strict whitelist versus listed-plus-reasonable-judgment. Strict removes ambiguity and is auditable but brittle when reality outruns the list; permissive adapts but reopens the grey zone the menu exists to close.
- Escalation threshold — how readily an off-menu situation triggers a stop. A low threshold is cautious but floods the escalation path and erodes local agency; a high one preserves autonomy but risks an actor pressing on past where they should have stopped.
- Refresh cadence — how often the menu is revised as conditions and tools change. Frequent revision keeps it matched to reality but costs re-approval; rare revision leaves an actor working from a menu that no longer fits the plant.
When it helps, and when it misleads¶
Its strength is preserving bounded agency exactly where unbounded discretion is unsafe: the actor keeps real, immediate choice among vetted options, while the actions that could cause serious harm are held behind an escalate path decided in calm rather than crisis. It is most valuable in high-error-cost, time-pressured settings — control rooms, clinical protocols, incident response — where "use your judgment" is both too slow and too dangerous.
Its failure mode is brittleness at the edges: a fixed menu cannot anticipate every situation, and a novel emergency can fall into the gap between "listed" and "escalate," leaving a capable actor frozen by a rule when improvisation was the safer move. Over-narrow menus also breed exactly the learned passivity the archetype warns against — an actor who escalates everything stops building judgment. And a stale menu confidently authorizes actions the plant has since made unsafe. The guarding discipline is to size the menu to genuine risk, keep the escalate path fast and blame-free so using it is cheap, and revise the list as the world it governs changes.
How it implements the components¶
The menu realizes the archetype's bounded-action slice — the permitted moves, their boundary, and the exit when the situation exceeds them:
action_repertoire— the menu is the actor's feasible action set, made concrete and vetted rather than left implicit.legitimate_action_boundary— as a whitelist, the menu draws the line between what may be done locally and what may not, closing the grey zone of assumed authority.override_and_escalation_path— every off-menu situation routes to a named stop-and-escalate, so the actor is never stranded when reality exceeds the list.
It fixes which actions are in bounds now but does not widen that boundary over time — the staged expansion of discretion, and the fading support_scaffold behind it, are Graduated Autonomy Ramp's; and it does not choose which listed action to take, which is Agency Loop Map's selection_policy arc. The menu is the fixed set of permitted moves; the ramp is the schedule by which that set grows.
Related¶
- Instantiates: Agentic Control Loop Design — it supplies the bounded action repertoire and its escalate path for high-risk settings.
- Sibling mechanisms: Agency Loop Map · Agency Health Dashboard · Model Assumption Register · After-Action Learning Cycle · Briefback or Intent Confirmation · Graduated Autonomy Ramp · Decision-Rights Matrix
Editorial Notes¶
Form Classification¶
Form family: Rule, Policy & Commitment
Rationale: Safe Action Menu operates by predeclares the exhaustive set of actions permitted without further approval under stress. That concrete deployed or enacted form is Rule, Policy & Commitment under the frozen taxonomy.
Nearest alternative: Representation, Specification & Plan — Although Representation, Specification & Plan can support this mechanism, the frozen evidence makes its operative form the act that predeclares the exhaustive set of actions permitted without further approval under stress; the alternative is therefore secondary rather than defining.
Review outcome: Adjudicated after independent review; high confidence.
Origin Attribution¶
Primary origin: Security Studies & Intelligence Analysis
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Multi-domain
Rationale: A preauthorized menu of in-bounds actions plus escalation for uncovered high-risk cases is an operational security and incident-response control. CISA and Ready.gov publish bounded choices such as assess, de-escalate, report, evacuate, shelter, or lock down, each conditioned on threat and authority; military command doctrine is a close contributing lineage.
Related originating lineages:
- Computer Science & Software Engineering — Computer science and software-engineering practice supplies a parallel or contributing lineage for the mechanism's defining operation: a fixed template of pre-approved, in-bounds actions for a high-risk setting, with an escalate path for anything the menu does not cover.
- Disaster Management & Risk Reduction — disaster_management contributes scenario exercises, protected action, and recovery preparedness to the mechanism's formative or independently convergent form; that contribution does not displace the primary security_intelligence lineage.
- Engineering & Design — Safety procedures independently constrain high-risk operation to certified actions.
- Law & Governance — law_governance contributes rights, exceptions, authority, review gates, and procedural accountability to the mechanism's formative or independently convergent form; that contribution does not displace the primary security_intelligence lineage.
- Medicine & Healthcare — medicine_healthcare contributes clinical safety, escalation, treatment delivery, and protected care to the mechanism's formative or independently convergent form; that contribution does not displace the primary security_intelligence lineage.
- Military & Strategic Studies — Military planning, readiness, and strategic operations supplies a parallel or contributing lineage for the mechanism's defining operation: a fixed template of pre-approved, in-bounds actions for a high-risk setting, with an escalate path for anything the menu does not cover.
- Organizational & Management Science — Decision-rights design materially generalizes menus to institutions.
Review resolution: The blind reviewers disagreed on primary lineage (military_strategic_studies versus security_intelligence); authoritative or primary research supports security_intelligence as the best historical origin. A preauthorized menu of in-bounds actions plus escalation for uncovered high-risk cases is an operational security and incident-response control. CISA and Ready.gov publish bounded choices such as assess, de-escalate, report, evacuate, shelter, or lock down, each conditioned on threat and authority; military command doctrine is a close contributing lineage. The cited Ready.gov, Emergency Plans; CISA, De-escalation Action Guide directly supports the defining operation used in that choice. All independently supported contributing domains are retained without an arbitrary cap, while domain_reach=multi_domain records later applicability separately from provenance.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Researched adjudication after independent review; high confidence.
Sources consulted:
Notes¶
[n1] The safe operating envelope in process-safety engineering defines the bounded region of conditions and actions within which a system may operate without escalation; the menu is that envelope expressed as an actionable whitelist plus an exit when conditions leave it. ↩