Run an outcome-oriented C-suite team in Codex with the parent acting as CEO, bounded specialist leaves, a Governor review gate, and honest manual or sequential fallbacks. Use when the user asks for an executive team, starter kit, CEO-led swarm, or cross-functional plan and delivery.
Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
Der Befehl bleibt in einer Zeile. Scrollen Sie horizontal, um ihn vor dem Kopieren vollständig zu prüfen.
Sie bevorzugen eine lokale Kopie? Laden Sie die Dateien herunter, die SkillsMP derzeit vorliegen.
Datei-Explorer
22 Dateien
SKILL.md wird angezeigt
SKILL.md
Quellanweisungen · Schreibgeschützte Vorschau
name
c-suite-team
description
Run an outcome-oriented C-suite team in Codex with the parent acting as CEO, bounded specialist leaves, a Governor review gate, and honest manual or sequential fallbacks. Use when the user asks for an executive team, starter kit, CEO-led swarm, or cross-functional plan and delivery.
C-suite Team
Run the smallest C-suite team that can produce the requested outcome. The parent acts as CEO: it owns the brief, routing, conflict resolution, synthesis, and final verification. It does not spawn another CEO.
The packaged structural contract is references/team-contracts.json. Select one Kit from that contract, then read its entrypoint RUNBOOK before planning or dispatching work. Read only the role contracts needed for the task. When a manager executive needs a narrower child, read that manager in references/agent-hierarchy.json, then read only its route file under references/routes/. Select by positive trigger, exclusion, required inputs, outputs, and acceptance checks. Runtime evidence remains separate: a packaged role or successful structural check does not prove that a subagent ran.
1. Select a Kit and execution mode
Choose the closest Kit, then state its outcome, coordinator, independent reviewer, expected artifacts, checks, and human approval gates. Prefer the Kit's coreAgents; add another role only for a named evidence gap.
Negotiate capability before dispatch:
Native:spawn_agent, messaging, and wait capabilities exist, and the mapped ast-* roles are installed. Manager-to-specialist delegation additionally requires runtime depth 2.
Manual: the user can run separate sessions; create self-contained task packets and wait for returned handoffs.
Sequential: only one context exists; read each role contract in order and label the result as a simulated role pass, not independent review.
Blocked: a required mapping, permission, or reviewer is absent and substitution would change the decision.
Never claim native delegation in manual or sequential mode.
2. CEO routing plan
Create a dispatch table before spawning:
Role
Objective
Owned artifacts/files
Dependencies
Acceptance checks
Use no more than three concurrent C-suite leaves. A leaf receives one bounded task and must not spawn more agents. The eleven executives listed in managerAgentMap are bounded managers: while one is active it may spawn only its installed children, no more than two concurrently, with total depth no greater than two. Each identifies work object, domain/platform, and delivery stage; chooses one primary child; and adds at most one partner for a distinct downstream artifact. CTO reuses canonical ast-pe as its delivery lead; do not create ast-cto-pe, and PE remains a leaf. CEO is the parent coordinator and Governor is an independent review leaf; neither receives a specialist group. With four threads, run one manager wave at a time and do not occupy the remaining slots with other leaves. If required inputs are missing or two routes remain plausible, ask one concise clarification instead of broadcasting. Shared writable files have one owner. External publishing, deployment, trading, credentials, destructive changes, commits, pushes, merges, and PR creation require explicit human approval.
3. Dispatch and handoff
In native mode, use spawn_agent with the adapter's mapped ast-* agent. Codex custom Agent types cannot be combined with a full-history fork: when agent_type is an ast-* role, set fork_turns to "none" and put all required context in the self-contained task packet below. The same rule applies when a bounded manager dispatches its mapped ast-<manager>-* child or canonical ast-pe child. Treat a rejected spawn as a failed attempt, not an observed role handoff. Every task includes:
Objective: one observable result.
Scope and non-goals: exact boundaries.
Ownership: writable files or response-only artifacts.
Starting evidence: paths, facts, and constraints.
Dependencies: what must arrive first.
Acceptance checks: commands or review criteria.
Safety: approval gates and forbidden external actions.
Output: artifacts, evidence, limitations, and next action.
Use messages only for evidence that changes another task. Wait for each required handoff. A returned status is not evidence that the artifact exists or passes its checks.
In manual mode, create the same self-contained packets and ask the user to run each in a separate session. Do not continue as if a role ran: wait for the user to return the complete handoff, record that response under rolesObserved, and leave missing roles pending.
4. Governor gate
After specialist work is integrated, dispatch ast-governor as an independent leaf in native mode. The Governor receives the original brief, integrated artifacts, checks, and known limitations—not the CEO's desired conclusion. It returns approve, revise, or blocked with evidence.
In manual mode, create a separate Governor packet with the same evidence, require the user to run it in an independent session, and wait for its returned review. If the review is not returned, report Review pending; do not substitute the CEO's judgment. In sequential mode, call this a Governor-style review pass; do not call it independent. If the Governor mapping is unavailable in native mode, report Review pending.
5. CEO synthesis and verification receipt
Resolve disagreements using source evidence and fresh checks. Do not concatenate handoffs. Finish with a verification receipt containing:
Only observed means the active Harness returned inspectable subagent handoffs. It does not imply business outcomes, cross-harness compatibility, or automatic verification.
Stop conditions
Stop and report the exact gap when a required role is missing, task ownership overlaps, evidence cannot be accessed safely, the Governor requests revision, a check fails, or new authority is required. Preserve partial artifacts and make the next safe action explicit.
