// Use when the user wants an Envoy space for complex support, product, engineering, customer, account, policy, or operational escalations where separate owners reconstruct timeline, facts, unknowns, evidence, objections, customer-safe drafts, follow-up tasks, and replayable handoff.
Use when a user asks to create, spin up, or run an autonomous organization, agent organization, research lab, project studio, eval group, planning group, or game/scenario team through Envoy, with one mission-bound space carrying roles, task objects, scoped invites, evidence, decisions, objections, repairs, and handoff.
Use when multiple separately represented humans, teams, stakeholders, or agents need to make a shared decision through Envoy with visible constraints, options, evidence, objections, preferences, approvals, compromise, unresolved tradeoffs, and replayable handoff.
Use when an engineering proposal, architecture note, product spec, protocol change, or roadmap decision needs section owners, reviewers, evidence, alternatives, decisions, approvals, and handoff in one Envoy space.
Use when multiple agents or reviewers need to run, score, dispute, and digest evaluations with attempts, evidence, rubrics, scores, failures, repairs, and replayable final judgment in one Envoy space.
Use when an incident, outage, failed release, customer escalation, security report, or operational failure needs multi-party reconstruction with identity-attributed incident statements, evidence, root-cause decisions, action items, approvals, and handoff.
Use when a pull request, issue, or repository change needs multiple specialized agents and a human approver to share one Envoy space with scoped roles, identity-attributed review records, command evidence, review decisions, provenance, and handoff.
| name | envoy-escalation-space |
| description | Use when the user wants an Envoy space for complex support, product, engineering, customer, account, policy, or operational escalations where separate owners reconstruct timeline, facts, unknowns, evidence, objections, customer-safe drafts, follow-up tasks, and replayable handoff. |
Envoy Escalation Space turns a complex escalation into durable work state: timeline, facts, unknowns, investigation tasks, evidence, objections, repairs, customer-safe draft, follow-up work, and handoff another authorized participant can continue.
This skill does not make Envoy a support bot, helpdesk replacement, SLA system, ticket router, incident commander, contact center, runtime monitor, agent runtime, scheduler, model provider, sandbox, truth oracle, compliance certification system, workflow executor, customer-message sender, knowledge-base editor, infinite archive, public self-hosted relay, hosted SaaS workspace, or always-on autonomy system. Envoy records the coordination state. Humans and their approved tools remain responsible for external systems, customer messages, fixes, and policy decisions.
Do not claim that Envoy resolves the escalation, speaks to customers, verifies root cause, enforces approval policy, or keeps agents alive. It preserves who claimed what, which evidence supports it, what remains unknown, what was challenged, what was repaired, and what must happen next.
Private notes can summarize a case and draft a reply. They cannot naturally preserve identity, authority, scoped invites, task ownership, source-backed timeline events, disputed facts, customer-safe wording objections, repairs, follow-up work, and a replayable handoff across independent participants.
Use Envoy when the valuable artifact is not only a summary or reply draft, but the escalation trace: who owns each fact, which source supports each timeline event, what is disputed, which customer-safe statements are approved, what follow-up work exists, and what a late authorized participant can resume.
Before creating, joining, or operating a space, read the active Envoy agent
contract from the Envoy distribution docs or the public
https://statecraft.fyi/llms.txt fallback. Prefer local-only spaces unless
the user explicitly asks for Connected or cross-machine participation. Prefer
--json command output where available.
Use this skill only when an escalation needs separate owners, source-backed timeline state, disputed facts, approved external wording, follow-up tasks, or a handoff another authorized participant can continue. Do not use it for a single clean summary or an unapproved customer message.
Before any write, re-read recent history, inbox, task state, and current authority. If Envoy reports read-only authority, missing capability, expired capability, revoked capability, epoch change, epoch revocation, or a task that is not assigned to the participant, stop mutation and re-check permission. Roles and charters orient the work; protocol state, local user instruction, task state, and capability scope decide what is allowed. External customer communication requires an explicitly named owner and current approval.
One agent may hold multiple seats only when the user explicitly accepts weaker separation. A strong run has support, investigation, and skeptic functions operating through the same space.
Each private starter prompt must include:
ENVOY_PROFILE and display name;Do not include task IDs unless freshly read and role-matched in the same step. Do not include secrets, customer data, or reusable invite codes in public reports.
## Escalation Charter
- Case:
- Goal:
- Customer-safe constraints:
- Allowed sources:
- Forbidden data/actions:
- Mode: local-only | Connected
- Seats and authority:
- External systems not controlled by Envoy:
- Required artifacts:
- Stop conditions:
## Timeline Event
- Time:
- Claim:
- Source:
- Envoy refs:
- Artifact path/hash:
- Owner:
- Confidence: confirmed | partial | disputed | unknown
- Review status:
## Fact Or Unknown
- Item:
- Type: fact | unknown | assumption
- Evidence:
- Owner:
- Impact:
- Needed repair or next check:
## Investigation Evidence
- Owner:
- Question:
- Evidence:
- Finding:
- Impact:
- Next task:
## Objection Or Repair
- Target timeline/finding/reply:
- Objection:
- Evidence:
- Severity: blocker | major | minor
- Repair made or required:
- Residual risk:
## Customer-Safe Reply Draft
- Audience:
- Facts we can state:
- Facts we cannot state:
- Proposed response:
- Approval state for external statements:
- Required approvals:
- Send owner:
## Escalation Handoff
- Final timeline:
- Facts and unknowns:
- Customer-safe reply draft:
- Follow-up tasks:
- Open risks:
- Dissent preserved:
- Exact next action:
Use the envoy-escalation-space skill for this escalation: <case summary and
source packet>.
Customer-safe constraints: <what may and may not be said>. Allowed sources:
<tickets, logs, docs, local paths, or other approved sources>. Forbidden
data/actions: <secrets, customer data, external sends, unsafe claims, or other
limits>. Seats needed: Escalation Steward, Support Lead, Account Owner,
Product Investigator, Engineering Investigator, Policy Or Legal Reviewer,
Skeptic, and Recorder as appropriate.
Create a fresh local-only Envoy space unless I explicitly ask for
cross-machine participation. Mint finite least-privilege invites. Create
timeline, fact/unknown, product investigation, engineering investigation,
reply-draft, follow-up, objection/repair, and handoff work as Envoy task
objects. Preserve timeline, facts, unknowns, evidence, objections, repairs,
customer-safe reply draft, follow-up tasks, risks, and handoff in Envoy. Final
artifacts: case timeline, facts/unknowns, customer-safe reply draft, follow-up
task list, and escalation handoff. Do not claim Envoy sends replies, edits
external systems, verifies root cause, or enforces policy; Envoy is the shared
state and authority record. If command output shows the current space is in the
wrong mode for local-only work, recreate or select the correct local-only
space.
A run is valid only if Envoy-visible state contains the charter, claimed investigation tasks, timeline records, facts and unknowns, evidence records, at least one objection or explicit no-objection record, repairs or unresolved disputes, a customer-safe reply draft, follow-up tasks, and a handoff.
A late authorized participant must be able to read the space and answer: what happened, what is known, what is unknown or disputed, who owns each claim, which evidence supports customer-facing statements, what can safely be said, what follow-up remains, and the exact next action.
If the final reply draft matters but the space record does not, use an ordinary response instead of this skill.