chore: rename main-agent skill to trellis-orchestrator

.agents/skills/trellis-orchestrator/SKILL.md

@@ -1,12 +1,12 @@
 ---
-name: main-agent
-description: "Coordinate multi-agent implementation as the primary agent with Fast, Default, and Detailed execution modes; own intent, scope, evidence, convergence, delegated execution, Git integration, and the task finish lifecycle."
+name: trellis-orchestrator
+description: "Orchestrate Trellis coding tasks across research, implementation, review, convergence, and final integration."
 ---
 
-# Main Agent
+# Trellis Orchestrator
 
-Use this skill when acting as the primary coordinator of a multi-agent coding
-task. The main agent owns the user's intent, scope, decisions, integration,
+Use this skill to orchestrate a multi-agent Trellis coding task. The
+orchestrator owns the user's intent, scope, decisions, integration,
 convergence, and final evidence. It delegates application-code edits and
 independent review, but it may inspect source and diffs, run non-writing
 diagnostics, and reproduce worker claims when needed to make coordination
@@ -71,7 +71,7 @@ Fast → Default → Detailed
 
 Do not automatically downgrade during an active task. Once `Detailed` has
 created persistent state, continue maintaining it so its status remains
-trustworthy. A user may explicitly redefine the mode, but the main agent must
+trustworthy. A user may explicitly redefine the mode, but the orchestrator must
 explain what existing state will remain authoritative.
 
 Escalate `Fast → Default` when evidence reveals multiple independently
@@ -82,7 +82,7 @@ straightforward local repair.
 Escalate `Default → Detailed` when evidence reveals a non-trivial dependency
 graph, cross-scope findings, multiple ownership or architecture decisions,
 long-running or replacement-prone workers, repeated convergence failure, or
-state that cannot be kept reliably in the main agent's compact runtime context.
+state that cannot be kept reliably in the orchestrator's compact runtime context.
 
 When escalating, preserve completed evidence, accepted scopes, open findings,
 worker ownership, and rework-round counts. Do not restart the task or reset a
@@ -95,7 +95,7 @@ coding work. A task merely listed under `.trellis/tasks/` is not enough.
 
 - Resolve the task with `python3 ./.trellis/scripts/task.py current --source`
   and read its `task.json` status. If no current task exists, refuse the request
-  and tell the user exactly: `main-agent skill 需要配合任务使用`. Do not create a
+  and tell the user exactly: `trellis-orchestrator skill 需要配合任务使用`. Do not create a
   task, invent a path, or handle the work inline.
 - During `planning`, dispatch only bounded `trellis-research` work. Do not
   dispatch `trellis-implement` or `trellis-check` until the task status is
@@ -129,7 +129,7 @@ evidence-dependent rather than mandatory for every task.
   for revalidation of its findings.
 - Only the main session may stage, commit, push, or merge. Workers must not
   perform those operations.
-- The main agent may inspect source, diffs, task artifacts, worker reports,
+- The orchestrator may inspect source, diffs, task artifacts, worker reports,
   lifecycle state, and changed-file names; coordinate workers; update task,
   planning, or project-spec documentation; and run bounded no-write checks. It
   must not implement, edit, or format application code.
@@ -138,7 +138,7 @@ evidence-dependent rather than mandatory for every task.
   self-review presented as independent acceptance.
 - Require the implementer to format and statically check the files it owns.
   Route formatting or production-code validation failures back to that
-  implementer; the main agent may reproduce them in no-write mode but must not
+  implementer; the orchestrator may reproduce them in no-write mode but must not
   auto-fix them.
 
 Before delegating, form a short runtime plan covering the goal, affected areas,
@@ -206,7 +206,7 @@ integration boundary.
 ```text
 Research 0..N by unknown question or semantic boundary
         ↓
-Main-agent synthesis: dependencies + scopes + acceptance criteria
+Orchestrator synthesis: dependencies + scopes + acceptance criteria
         ↓
 Choose one ready scope
         ↓
@@ -235,7 +235,7 @@ Done / reopen scope / escalate to Detailed
 ### Rolling Scope Acceptance
 
 - Choose the next `ready` scope by semantic dependency, not directory order.
-- Keep only compact runtime state in the main-agent context, for example:
+- Keep only compact runtime state in the orchestrator context, for example:
 
   ```text
   Current scope: extension
@@ -428,7 +428,7 @@ Every check report includes:
 - passed, fixed, and open findings;
 - every validation command marked `passed`, `failed`, `skipped`, or `blocked`.
 
-No report means `review_pending`, never `review_passed`. The main agent's own
+No report means `review_pending`, never `review_passed`. The orchestrator's own
 inspection or a passing command cannot manufacture independent acceptance.
 
 ## Final Validation And Git Integration

.codex/config.toml

@@ -37,5 +37,5 @@ project_doc_fallback_filenames = ["AGENTS.md"]
 # (this file) takes precedence over user-level `~/.codex/config.toml`.
 [agents]
 max_depth = 1
-# Combined native-subagent runtime ceiling; role-specific caps live in main-agent/SKILL.md.
+# Combined native-subagent runtime ceiling; role-specific caps live in trellis-orchestrator/SKILL.md.
 max_concurrent_threads_per_session = 11