菜单
Use when onboarding to the Valor AI codebase, understanding the system architecture, or when the user asks 'how does this work'. Comprehensive codebase orientation guide.
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
| name | prime |
| description | Use when onboarding to the Valor AI codebase, understanding the system architecture, or when the user asks 'how does this work'. Comprehensive codebase orientation guide. |
| disable-model-invocation | true |
Get oriented in the Valor AI system well enough to add features effectively.
This is Valor - an AI coworker that runs on its own Mac. Not an assistant, not a tool - a colleague with agency. The supervisor assigns work, Valor executes autonomously.
You ARE this codebase. When users say "you" or "your features," they mean this code.
Telegram → Bridge (Telethon) → Redis AgentSession queue (bridge is I/O only)
Standalone Worker (python -m worker) → sole session execution engine → headless session runner (`agent/session_runner/`, one `claude -p` subprocess per turn)
Reflection scheduler (python -m reflections) → own launchd subprocess; enqueues recurring work the worker executes
Key components:
bridge/telegram_bridge.py): Telegram user account via Telethon; enqueues sessions and routes output (nudge loop) — no SDLC awarenessworker/__main__.py): executes AgentSessions via the headless session runner (agent/session_runner/, harness in agent/sdk_client.py).mcp.json): modular capabilities (memory, BYOB, Sentry, Notion); GitHub via gh CLIconfig/identity.json + config/personas/segments/): structured identity data and composable persona segmentsai/ # This repo
├── .claude/skills/ # Project-only skills (/prime, /setup, /sdlc, /update, /do-deploy)
├── .claude/skills-global/ # Global skills — hardlinked to ~/.claude/skills/ on every machine by /update
├── agent/ # Session queue, SDK client, output routing
├── bridge/ # Telegram bridge
├── worker/ # Standalone worker service (python -m worker)
├── reflections/ # Out-of-process reflection scheduler (python -m reflections)
├── tools/ # Local Python tools (valor-* CLIs)
├── config/ # identity.json, personas/segments/, reflections.yaml
├── scripts/valor-service.sh # Service management
├── docs/features/ # Feature index — how things actually work
└── CLAUDE.md # Development guide (READ THIS)
In order:
CLAUDE.md - development principles, commands, architectureconfig/personas/segments/ - Valor's identity, work patterns, and toolsdocs/features/README.md - feature index, when you need how-something-works detailCreate .claude/skills-global/<name>/SKILL.md for skills every machine should know (global bodies stay repo-agnostic; repo specifics go in the .claude/skill-context/ seam), or .claude/skills/<name>/SKILL.md for project-only skills. See "Global vs. Project-Only Skills" in CLAUDE.md.
New Python tools are invisible to the agent until wired into a CLI entry point (pyproject.toml [project.scripts]) or imported by the bridge directly.
./scripts/valor-service.sh status # Check if running
./scripts/valor-service.sh restart # Restart bridge, watchdog, and worker after changes
./scripts/valor-service.sh logs # View logs
Check system status:
./scripts/valor-service.sh status
tail -20 logs/bridge.error.log
curl -s localhost:8500/dashboard.json # full system state as JSON
After making changes:
git add . && git commit -m "Description" && git push
./scripts/valor-service.sh restart
Run /prime at the start of any session to get oriented.
Pre-publish editorial gate for external-audience content: removes AI-writing tells and blocks hollow drafts. Use on documents, presentations, emails, posts, or web copy once a first draft exists. Triggered by 'de-slop this', 'editorial pass', 'humanize this', 'remove the AI tells', or invoked as a mandatory gate by email, do-presentation, linkedin, and x-com.
Stage, launch, grade, and schedule a Claude Managed Agent (CMA) from a build-sheet. Triggered by 'build the agent', 'launch this agent', 'deploy a managed agent', or as /imagine-agent's handoff.
Create/review/maintain Claude Code Routines (cloud scheduled agents, aka 'Cowork'). Triggered by 'set up a routine', 'schedule a cloud agent', 'cowork task', 'migrate to a routine'.
Create a polished Marp slide deck about a feature, concept, or system. Triggered by 'make a presentation', 'create slides', 'do-presentation', or 'explain this as a deck'.
Use when reading, searching, drafting, or sending email. Triggered by 'read my email', 'check my inbox', 'send an email', 'reply to that email', 'search my mail', or any request to read/send mail.
Use when reading posts or engaging on LinkedIn. Triggered by requests to comment on LinkedIn, browse the feed, write or interact with posts, check DMs, or read/reply to LinkedIn messages.