スキル一覧に戻る
Mburdo

prime

by Mburdo

Opinionated AI agent dev stack with tools, guides, templates, and workflows to take you from good to great.

47🍴 6📅 2026年1月14日
GitHubで見るManusで実行

SKILL.md


name: prime description: New agent startup checklist for Agent Mail and Beads. Use when starting a new agent session, when the user says "prime" or "startup", or when beginning work on a multi-agent project.

Prime — Session Startup

Complete startup checklist for new agent sessions. Direct execution (no subagents).

Design rationale: This skill executes directly rather than spawning subagents because each step is a simple command sequence (~200 tokens each), not substantial analytical work. Per Lita research: "Simple agents achieve 97% of complex system performance with 15x less code." Subagent overhead would exceed actual work.

When This Applies

SignalAction
New session startingRun full checklist
User says "/prime"Run full checklist
User says "startup" or "initialize"Run full checklist
Multi-agent coordination neededRun full checklist

Tool Reference

Agent Mail (MCP)

ToolPurpose
macro_start_session(human_key, program, model)PREFERRED: All-in-one startup
ensure_project(human_key)Initialize project in Agent Mail
register_agent(project_key, program, model)Register and get agent name
set_contact_policy(agent_name, policy)Set to "open" for coordination
fetch_inbox(agent_name, limit)Check for messages
send_message(to, subject, body_md)Send greeting/announcements

Bash Commands

CommandPurpose
cm context "task description" --jsonGet patterns + anti-patterns before work
cass search "query" --robot --days 7Search session history
bd ready --jsonList available tasks
bv --robot-triageGet task recommendations
git statusCheck repo state
git log --oneline -5Recent commits

Execution Flow

Execute these steps directly. No subagents needed.

Step 1: Register Identity

PREFERRED — Use macro for simplicity:

result = macro_start_session(
    human_key="PROJECT_PATH",  # Absolute path to project
    program="claude-code",
    model="opus-4.5",
    inbox_limit=10
)
agent_name = result["agent"]["name"]  # e.g., "BlueLake"

ALTERNATIVE — Manual steps:

# 1. Ensure project exists
ensure_project(human_key="PROJECT_PATH")

# 2. Register agent (name is auto-assigned)
result = register_agent(
    project_key="PROJECT_PATH",
    program="claude-code",
    model="opus-4.5",
    task_description="Agent initializing..."
)
agent_name = result["name"]  # SAVE THIS

# 3. Set contact policy for coordination
set_contact_policy(
    project_key="PROJECT_PATH",
    agent_name=agent_name,
    policy="open"
)

Update terminal title:

echo -ne "\033]0;[{agent_name}] | PROJECT_NAME\007"

Step 2: Orient

Read project documentation:

Read("AGENTS.md")     # Agent workflow instructions
Read("CLAUDE.md")     # Project overview

Check recent session history:

cass search "PROJECT_NAME" --workspace "PROJECT_PATH" --robot --days 7 --limit 5

Get learned patterns:

cm context "starting work on PROJECT_NAME" --json

Check git state:

git status
git log --oneline -5
git branch

Discover active agents:

ReadMcpResourceTool(
    server="mcp-agent-mail",
    uri="resource://agents/PROJECT_PATH"
)

Step 3: Coordinate

Check inbox:

messages = fetch_inbox(
    project_key="PROJECT_PATH",
    agent_name=agent_name,
    include_bodies=True,
    limit=20
)

Categorize messages:

  • Urgent — Requires response before starting work
  • Informational — [CLAIMED], [CLOSED] announcements
  • Acknowledgement required — ack_required=True

Note file reservations from other agents (these are off-limits).

Send greeting (if other agents active):

send_message(
    project_key="PROJECT_PATH",
    sender_name=agent_name,
    to=[active_agents],
    subject="Agent Online",
    body_md="I'm online. Focusing on: [task]. What needs attention?",
    importance="normal"
)

Step 4: Discover Work

Get recommendations:

bv --robot-triage      # Priority analysis
bv --robot-next        # Single best task

Get ready tasks:

bd ready --json

Check already claimed:

bd list --status in_progress --json

Filter by file availability:

  • Check each ready task's required files
  • Exclude tasks where files are reserved by others
  • Recommend highest-priority available task

Step 5: Output Summary

Present to user:

## Agent Initialized

**Name:** {agent_name}
**Project:** PROJECT_NAME
**Branch:** {git_branch}

### Context
- Recent sessions: {summary from CASS}
- Active agents: {list or "none"}
- Inbox: {count} messages ({urgent} urgent)
- Blocked files: {list or "none"}

### Learned Patterns
{relevant patterns from cm}

### Recommended Task
**ID:** {task_id}
**Title:** {task_title}
**Priority:** {priority}
**Files:** {file_scope}

Ready to claim? Use `/advance` to claim and begin work.

Mandatory Gates

After priming, these gates apply to all work:

GateRequirementEvidence
TDD-firstWrite tests before implementation45.97% higher success rate
ubs --stagedSecurity scan before every commit~40% of LLM code has vulnerabilities
Max 3 iterationsStop after 3 repair attemptsSecurity degrades with more
Escalate on failureCreate spike, notify operatorDon't waste time debugging

Quick Reference

# Register (preferred)
macro_start_session(human_key=PROJECT_PATH, program="claude-code", model="opus-4.5")

# Orient
cass search "PROJECT" --robot --days 7
cm context "task" --json
git status

# Coordinate
fetch_inbox(agent_name=YOUR_NAME, limit=20)
send_message(to=[...], subject="Agent Online", ...)

# Discover
bv --robot-triage
bd ready --json

Anti-Patterns

Don'tWhyDo Instead
Skip registrationCan't coordinateAlways register first
Ignore inboxMiss urgent messagesCheck before claiming
Ignore reservationsFile conflictsRespect all reservations
Skip cm contextMiss learned patternsAlways check patterns
Claim without checkingDuplicate workVerify availability first

See Also

  • /advance — Claim and work on tasks
  • /recall — Get context from past sessions
  • beads-cli/ — bd command reference
  • beads-viewer/ — bv command reference

スコア

総合スコア

65/100

リポジトリの品質指標に基づく評価

SKILL.md

SKILL.mdファイルが含まれている

+20
LICENSE

ライセンスが設定されている

+10
説明文

100文字以上の説明がある

+10
人気

GitHub Stars 100以上

0/15
最近の活動

3ヶ月以内に更新がある

0/10
フォーク

10回以上フォークされている

0/5
Issue管理

オープンIssueが50未満

+5
言語

プログラミング言語が設定されている

0/5
タグ

1つ以上のタグが設定されている

0/5

レビュー

💬

レビュー機能は近日公開予定です