菜单
Triggers: "AgentName, ...", "Hey AgentName", "what commands do I have", "what can you do with MCP", "command reference". Routes ambiguous intent to correct facade + op. Also handles vague phrases like "save this", "remember this", "search for patterns".
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Triggers: "learn from this", "harvest knowledge", "harvest this codebase", "extract patterns from". Bulk-extracts patterns from code/docs/PRs/articles into vault. Use vault-capture for single patterns, intake for ingesting external sources.
Triggers: "add a facade", "new tool", "extend vault", "add brain feature", "new skill", "extend agent". Extends agent internals with vault-first knowledge gathering.
Triggers: "I want to build something", "let's think about", "what if we", "creative exploration", "ideate". Open-ended creative exploration when requirements are not yet clear.
Triggers: "check against patterns", "pattern compliance", "convention check", "review against vault", "institutional review". Reviews code against vault patterns/conventions. Use deep-review for general quality/architecture.
Triggers: "curator status", "vault health", "how is the vault", "curator health", "check vault quality", "what needs grooming". Quick vault health + grooming recs. Use vault-curate for full maintenance.
Triggers: "deep review", "code smells", "architecture review", "is this well architected", "optimization review". Multi-pass structural health/smells/quality review. Use code-patrol for vault pattern compliance.
| name | soleri-agent-mode |
| tier | default |
| description | Triggers: "AgentName, ...", "Hey AgentName", "what commands do I have", "what can you do with MCP", "command reference". Routes ambiguous intent to correct facade + op. Also handles vague phrases like "save this", "remember this", "search for patterns". |
This skill governs how to interpret natural language directed at this agent and route it to the correct facade + op. Use it when a phrase is ambiguous, when the user addresses the agent directly, or when surfacing the command reference.
Before executing any agent op, apply this 3-step protocol:
Step 1 — Semantic parse Identify the verb + object in the user's message:
Step 2 — Confirm via route_intent (when ambiguous)
YOUR_AGENT_control op:route_intent
params: { prompt: "<user's phrase>" }
Use this when the phrase matches 2+ intents. Skip it for obvious, unambiguous commands.
Step 3 — Execute the mapped op
Use the NL → Op table below to select facade + op. Pass projectPath: "." wherever required.
| What the user says | Facade | Op | Notes |
|---|---|---|---|
| "search for X" / "find patterns on X" | YOUR_AGENT_vault | search_intelligent | mode: "scan" first |
| "what do we know about X" | YOUR_AGENT_vault | search_intelligent | mode: "full" if specific |
| "recall X" / "remember X from before" | YOUR_AGENT_memory | memory_search | crossProject if broad |
| "what patterns are strong" | YOUR_AGENT_brain | brain_strengths | days: 30 default |
| "recommend something" / "what should I do" | YOUR_AGENT_brain | brain_recommend | |
| "show me vault entries about X" | YOUR_AGENT_vault | load_entries | after scan |
| "what's in the vault" | YOUR_AGENT_vault | vault_stats |
| What the user says | Facade | Op | Notes |
|---|---|---|---|
| "save this" / "capture this" (detailed pattern) | YOUR_AGENT_vault | capture_knowledge | full structured entry |
| "quick save" / "jot this down" | YOUR_AGENT_vault | capture_quick | lightweight, fast |
| "remember this for the session" | YOUR_AGENT_memory | memory_capture | session-scoped |
| "save the session" / "log what we did" | YOUR_AGENT_memory | session_capture | end-of-session |
| "ingest this URL" / "add this article" | YOUR_AGENT_vault | ingest_url | external content |
| "ingest this text" | YOUR_AGENT_vault | ingest_text | raw text |
| What the user says | Facade | Op | Notes |
|---|---|---|---|
| "create a plan" / "plan this" | YOUR_AGENT_plan | create_plan | MANDATORY before code |
| "approve the plan" / "looks good" | YOUR_AGENT_plan | approve_plan | Gate 1 |
| "split the plan" / "break it into tasks" | YOUR_AGENT_plan | plan_split | Gate 2 |
| "check plan drift" / "what changed" | YOUR_AGENT_plan | plan_reconcile | post-execution |
| "complete the plan" / "close it out" | YOUR_AGENT_plan | plan_complete_lifecycle | final step |
| "list active plans" | YOUR_AGENT_plan | chain_list |
| What the user says | Facade | Op | Notes |
|---|---|---|---|
| "orchestrate this" / "AgentName, orchestrate X" | YOUR_AGENT_orchestrate | orchestrate_plan | vault+brain+plan in one call |
| "execute the plan" / "start executing" | YOUR_AGENT_orchestrate | orchestrate_execute | tracks progress |
| "finish up" / "wrap the session" | YOUR_AGENT_orchestrate | orchestrate_complete | epilogue: vault + session |
| What the user says | Facade | Op | Notes |
|---|---|---|---|
| "what has the brain learned" / "brain stats" | YOUR_AGENT_brain | brain_stats | |
| "give feedback on that" | YOUR_AGENT_brain | brain_feedback | reinforcement signal |
| "what patterns are failing" | YOUR_AGENT_brain | brain_decay_report | |
| "radar analysis" / "spot patterns" | YOUR_AGENT_brain | radar_analyze | |
| "show radar candidates" | YOUR_AGENT_brain | radar_candidates |
| What the user says | Facade | Op | Notes |
|---|---|---|---|
| "health check" / "is everything working" | YOUR_AGENT_admin | admin_health | |
| "list all tools" / "what tools do you have" | YOUR_AGENT_admin | admin_tool_list | |
| "session briefing" / "catch me up" | YOUR_AGENT_admin | session_briefing | |
| "diagnose X" | YOUR_AGENT_admin | admin_diagnostic | |
| "run routing accuracy" | YOUR_AGENT_control | routing_accuracy |
| What the user says | Facade | Op | Notes |
|---|---|---|---|
| "clean the vault" / "groom" | YOUR_AGENT_curator | curator_groom | |
| "find duplicates" | YOUR_AGENT_curator | curator_detect_duplicates | |
| "vault health audit" | YOUR_AGENT_curator | curator_health_audit | |
| "archive old entries" | YOUR_AGENT_vault | vault_archive | |
| "push vault to git" / "sync vault" | YOUR_AGENT_vault | vault_git_push |
| What the user says | Facade | Op | Notes |
|---|---|---|---|
| "Hello, AgentName!" / activate | YOUR_AGENT_core | activate | params: projectPath |
| "Goodbye, AgentName!" / deactivate | YOUR_AGENT_core | activate | params: deactivate: true |
| "who are you" / "what mode is this" | YOUR_AGENT_core | identity | |
| "register this project" | YOUR_AGENT_core | register | params: projectPath |
| "route this intent" | YOUR_AGENT_control | route_intent | |
| "switch mode" / "morph to X" | YOUR_AGENT_control | morph | |
| "what are the behavior rules" | YOUR_AGENT_control | get_behavior_rules |
These phrases are ambiguous — resolve using context signals:
| Phrase | Context signal | Route to |
|---|---|---|
| "save this" | Detailed explanation, pattern discussion | vault.capture_knowledge |
| "save this" | Quick note, side comment | vault.capture_quick |
| "save this" | "For this session only" | memory.memory_capture |
| "remember this" | Repeatable rule or pattern | vault.capture_knowledge |
| "remember this" | Session-scoped fact | memory.memory_capture |
| "search for X" | Broad/exploratory | vault.search_intelligent mode:scan |
| "search for X" | Specific known entry | vault.search_intelligent mode:full |
| "search for X" | Across past sessions | memory.memory_search crossProject:true |
| "check this" | Code/PR in context | → deep-review skill |
| "check this" | System health | admin.admin_health |
| "plan this" | Work task | orchestrate.orchestrate_plan |
| "plan this" | User wants to see plan first | plan.create_plan |
Resolution rule: When still ambiguous after checking context signals, ask one clarifying question before routing. Never guess and execute a write op.
When the user says "AgentName, orchestrate [task description]":
YOUR_AGENT_orchestrate op:orchestrate_plan — vault lookup + brain recommendations + structured plan in one callYOUR_AGENT_plan op:plan_split — decompose into tasksYOUR_AGENT_orchestrate op:orchestrate_execute — track execution per taskYOUR_AGENT_orchestrate op:orchestrate_complete — epilogue: vault capture + session recordNever skip Gate 1 or Gate 2. Never proceed past a gate without explicit user approval.