| name | l3 |
| description | Terse CEO front-door — Level 3: full autonomy, CEO executes inline on this thread, escalate only at the DOA boundary. Use when: "/l3 <goal>", "L3 <goal>", you want the "全權處理 / get it done" behavior as one command without the CEO startup Q&A. Presets involvement=just-results, scope=Hold, project red lines plus -x additions (override --mode / --expand). Not for: offloading to a background foreman (→ /l4), hetero impl engine (→ /l5), participatory planning (→ dev-flow), research-only (→ survey).
|
/l3 — CEO autonomy, inline
Routing overlap? If this intent better matches a sibling skill, redirect per references/routing-tiebreaks.md (prefer explicit commands over conversational autonomy).
Terse front-door into autopilot:ceo-agent at Level 3: the CEO executes the
goal itself on this thread and escalates only at the DOA boundary.
Hard rules:
- Startup pre-filled, never re-asked on a clean goal: OKR from
<goal>;
involvement=3 just-results; scope=Hold (--expand → Expand); governance mode from the project
default (--mode changes only this run); project red lines plus -x <csv> additions.
- Before any TaskCreate, branch, worktree, runner, model, or inline implementation effect, run
node <plugin>/scripts/session-mode.js set --level l3 --repo-root <repo>. The command performs
canonical Mission policy/graph/source admission before writing its marker. Enforce requires
READY; shadow is disclosed without authority; off remains LEGACY.
- Posture: inline — no foreman dispatch.
/l3 is also the --solo degradation
target for /l4//l5//l6.
- The front-door changes startup ONLY — every
ceo-agent gate (size → project setup
→ admitted deliverables → finish-flow) still applies. Plan headings, modules, tests,
reviewers, and retries remain coverage/gates inside those deliverables.
P3.0 shadow translation
When the consuming project explicitly has .claude/owner-kernel-governance.json and an
explicit Autopilot source path, it may resolve the read-only mapping before work. In this example,
<autopilot-source> is a literal absolute source checkout or project-provided installed copy, not
an assumed environment variable:
node <autopilot-source>/scripts/owner-kernel.js translate-level \
--config .claude/owner-kernel-governance.json --level l3 [--expand] [-x <red-line-csv>] --check
Treat the result as a frozen topology preference and red-line disclosure, never as an approval,
action permit, or acceptance result. Do not fall back to Autopilot's own dogfood config when the
consuming project has no config or no explicit source path. A host that provides
ShadowTranslationRuntime may record the same mapping as a witnessed translation_used event; a
shell/skill invocation cannot mint that event and must not claim telemetry when no host bridge is
configured.
P3.0 does not replace the existing /l3 lifecycle, DOA, independent QC, or finish-flow rails.
/l4 through /l6 remain on their present strict worktree/dispatch paths until the supervised
engine bridge is complete.
Capability profile (shadow): /l3 fixes inline topology only. When the host supplies a current
verified envelope/grant/profile payload, pass it through unchanged; never infer guidance density
from the level or model name. A late mismatch requires a fresh-session handoff.
Because /l3 executes in the already-loaded inline session, it remains on the guided compatibility
host until an independently witnessed adapter launches it; appending autonomous prose is not a switch.
MUST-READ: ../ceo-agent/references/level-front-door.md
(front-door semantics) and ../ceo-agent/SKILL.md (DOA,
Prime Directives, quality gates).