prime

<SUBAGENT-STOP> If you were dispatched as a subagent to execute a specific task, skip this skill. Prime injects main-agent project context — subagents receive context from their dispatch prompt. </SUBAGENT-STOP> # Prime — Context Injection Rapidly load project context for the requested domain and output a scannable summary. This is NOT documentation — it's active context injection that makes your next actions more informed. ## Routing Based on `$ARGUMENTS`, pick ONE variant: | Argument | Domain | Key Entry Points | |----------|--------|-----------------| | _(empty)_ | Full project overview | boot.md + CLAUDE.md + src/ top-level | | `docker` | Docker services layer | docker-compose.yml, collectors, dashboard | | `channel` | Channel/messaging layer | src/channels/, telegram/, wechat/ | | `soul` | SOUL identity + skills + hooks | .claude/, SOUL/, hooks, skills | | `governance` | Governance pipeline | src/governance/, executor, approval, eval | ## Execution ### Step 1: Read Core Files **Always read first** (regardless of variant): 1. `.claude/boot.md` — identity and relationship context 2. `CLAUDE.md` — project rules and constraints **Then read variant-specific files** (see tables below). ### Step 2: Scan Structure Run `ls` on the relevant directories to understand current file layout. ### Step 3: Output Summary Format as a scannable brief (bullets, tables, NOT prose). Target: **<300 words**. ``` ## Prime: [variant] ### Structure [directory tree or table of key files] ### Entry Points [3-5 files that are the starting points for work in this domain] ### Key Patterns [2-3 patterns/conventions specific to this area] ### Current State [any notable recent changes from git log --oneline -5] ### Watch Out [1-2 gotchas or non-obvious things] ``` --- ## Variant: Full Overview (no argument) Read: - `src/` top-level directory listing - `docker-compose.yml` (if exists) - `package.json` or `pyproject.toml` (if exists) Focus: What is this project? What are the major modules? What's the tech stack? ## Variant: `docker` Read: - `docker-compose.yml` - `src/collectors/` directory listing - `dashboard/` directory listing (if exists) - `.env.example` or `.env` structure (DO NOT output values) Focus: What services run? What ports? What volumes? How do collectors work? ## Variant: `channel` Read: - `src/channels/__init__.py` - `src/channels/base.py` - `src/channels/telegram/channel.py` (first 80 lines) - `src/channels/config.py` - `src/channels/registry.py` Focus: How do messages flow in? What's the adapter interface? How is routing done? ## Variant: `soul` Read: - `.claude/boot.md` - `.claude/settings.json` - `SOUL/public/prompts/` directory listing - `.claude/skills/` directory listing - `.claude/hooks/` directory listing Focus: What skills exist? What hooks are active? What's the prompt architecture? ## Variant: `governance` Read: - `src/governance/executor.py` (first 80 lines) - `src/governance/approval.py` (first 50 lines) - `src/governance/eval/` directory listing - `src/governance/pipeline/` directory listing - `src/governance/supervisor.py` (first 50 lines) Focus: How does task execution work? What's the approval flow? What eval capabilities exist? ## Rules - Output is for YOUR context, not documentation. Be specific about file paths and line numbers. - If a file doesn't exist, skip it silently — don't report errors for missing optional files. - After priming, immediately proceed with the user's actual task. Don't wait for confirmation. - Prime is a warm-up, not a deliverable. Spend <60 seconds on it.