npx skills add https://github.com/colonyops/hive --skill hcHoneycomb 是 hive 内置的任务协调系统。指挥者智能体创建史诗和任务;工作者智能体认领叶子项,记录进度并标记工作完成。
会话 ID 和仓库密钥会自动从工作目录中检测。
hive hc create <title> 创建单个史诗或任务(任务需要 --parent)
hive hc create 从标准输入 JSON 批量创建
hive hc list [epic-id] 列出项目(可选史诗过滤器)
hive hc show <id> 显示项目 + 评论
hive hc update <id> 更新状态或分配
hive hc next <epic-id> 获取史诗的下一个可执行任务
hive hc comment <id> <message> 向项目添加评论
hive hc context <epic-id> 显示史诗上下文块
hive hc prune 移除旧的已完成项目
# 获取史诗上下文(Markdown 格式,适合粘贴到提示中)
hive hc context <epic-id>
# 以 JSON 格式获取上下文,用于编程
hive hc context <epic-id> --json
# 一步获取并分配
hive hc next <epic-id> --assign
# 或者先获取,再分配
hive hc next <epic-id>
hive hc update <id> --assign --status in_progress
评论会累积在上下文中并消耗令牌——只有在有值得为下一个智能体保留的内容时才添加一条。相比于几条增量记录,更推荐一条实质性的记录。
适合评论的情况:
广告位招租
在这里展示您的产品或服务
触达数万 AI 开发者,精准高效
做出了一个不明显的决定(“因为 X 选择了轮询而非 webhook”)
在任务中途停止,下一个智能体需要了解情况
发现了阻塞问题或外部依赖
hive hc comment <id> "切换到乐观锁——悲观锁在高负载下导致死锁"
停止时的一条评论,前缀为 CHECKPOINT:,就足够了。包含已完成和待完成的内容。
hive hc comment <id> "CHECKPOINT: 数据库模式已完成,接下来需要连接 API 处理器"
hive hc update <id> --status done
echo '{
"title": "实现身份验证",
"type": "epic",
"children": [
{"title": "添加 JWT 库", "type": "task"},
{"title": "实现登录端点", "type": "task"},
{"title": "实现登出端点", "type": "task"},
{"title": "添加会话中间件", "type": "task"}
]
}' | hive hc create
所有项目以 JSON 行输出(每行一个):
{"id":"hc-abc1","type":"epic","title":"实现身份验证",...}
{"id":"hc-abc2","type":"task","title":"添加 JWT 库","epic_id":"hc-abc1",...}
...
hive hc create "修复登录重定向" --type task --parent <epic-id>
hive hc create "新功能史诗" --type epic
# 列出史诗的所有开放任务
hive hc list <epic-id> --status open
# 列出分配给特定会话的项目
hive hc list --session <session-id>
# 显示项目 + 完整的评论历史
hive hc show <id>
hive hc create [title]单项目模式(标题作为位置参数):
--type epic|task — 项目类型(默认:task;任务需要 --parent)--desc <text> — 项目描述--parent <id> — 父项目 ID(任务必需)批量模式(无位置参数,从标准输入或 --file 读取 JSON):
--file <path> — 从文件而非标准输入读取 JSONhive hc list [epic-id][epic-id] — 可选位置参数,按史诗过滤(仅返回子项,不包括史诗本身)--status open|in_progress|done|cancelled — 按状态过滤--session <id> — 按会话 ID 过滤输出:使用 --json 时为 JSON 行;否则为彩色树状图(编程解析时使用 --json)。
hive hc show <id>输出:JSON 行——首先是项目,然后是按时间顺序排列的评论。
hive hc update <id>--status open|in_progress|done|cancelled — 新状态--assign — 分配给当前会话--unassign — 移除会话分配hive hc next <epic-id>返回下一个可执行的叶子任务(开放/进行中,且没有开放/进行中的子任务)。
<epic-id> — 必需的位置参数,用于限定到某个史诗--assign — 分配给当前会话并将状态设置为 in_progress如果未找到可执行任务,则退出并报错。
hive hc comment <id> <message>向项目添加评论。ID 之后的所有位置参数将连接起来作为消息。
谨慎使用——评论会累积在上下文块中,并且对后续读取此史诗的每个智能体都可见。相比于几条增量记录,更推荐一条有意义的记录。对于交接,请使用 CHECKPOINT: 前缀。
hive hc context <epic-id>组装包含以下内容的上下文块:
--json — 以单个 JSON 对象输出(默认:markdown)hive hc prune--older-than <duration> — 移除早于此时间的项目(默认:168h / 7 天)--status <status> — 要清理的状态,可重复(默认:done, cancelled)--dry-run — 显示数量但不移除输出:{"action":"pruned","count":N}
{
"id": "hc-abc123",
"repo_key": "owner/repo",
"epic_id": "hc-epic1",
"parent_id": "hc-epic1",
"session_id": "mysession",
"title": "实现登录端点",
"desc": "可选描述",
"type": "task",
"status": "in_progress",
"depth": 1,
"created_at": "2026-01-01T00:00:00Z",
"updated_at": "2026-01-01T01:00:00Z"
}
史诗具有 type: "epic" 和空的 epic_id(它们是自己的根)。
{
"id": "hcc-def456",
"item_id": "hc-abc123",
"message": "CHECKPOINT: 身份验证中间件完成,需要测试",
"created_at": "2026-01-01T02:00:00Z"
}
| 状态 | 含义 |
|---|---|
open | 未开始,可供认领 |
in_progress | 正在积极处理中 |
done | 已完成 |
cancelled | 已放弃,将不会完成 |
每周安装量
1
仓库
GitHub 星标数
19
首次出现
今天
安全审计
安装于
zencoder1
amp1
cline1
openclaw1
opencode1
cursor1
Honeycomb is hive's built-in task coordination system. A conductor agent creates epics and tasks; worker agents claim leaf items, record progress, and mark work done.
Session ID and repo key are auto-detected from the working directory.
hive hc create <title> Create a single epic or task (tasks require --parent)
hive hc create Bulk create from stdin JSON
hive hc list [epic-id] List items (optional epic filter)
hive hc show <id> Show item + comments
hive hc update <id> Update status or assignment
hive hc next <epic-id> Get next actionable task for an epic
hive hc comment <id> <message> Add a comment to an item
hive hc context <epic-id> Show epic context block
hive hc prune Remove old completed items
# Get epic context (markdown, suitable for pasting into prompt)
hive hc context <epic-id>
# Get context as JSON for programmatic use
hive hc context <epic-id> --json
# Get and assign in one step
hive hc next <epic-id> --assign
# Or get first, then assign
hive hc next <epic-id>
hive hc update <id> --assign --status in_progress
Comments accumulate in context and consume tokens — only add one when there is something worth preserving for the next agent. Prefer a single substantive note over several incremental ones.
Good reasons to comment:
A non-obvious decision was made ("chose polling over webhooks because X")
Stopping mid-task and the next agent needs orientation
A blocker or external dependency was discovered
hive hc comment <id> "Switched to optimistic locking — pessimistic caused deadlocks under load"
One comment at stop time, prefixed with CHECKPOINT:, is enough. Include what is done and what remains.
hive hc comment <id> "CHECKPOINT: DB schema done, need to wire API handlers next"
hive hc update <id> --status done
echo '{
"title": "Implement Authentication",
"type": "epic",
"children": [
{"title": "Add JWT library", "type": "task"},
{"title": "Implement login endpoint", "type": "task"},
{"title": "Implement logout endpoint", "type": "task"},
{"title": "Add session middleware", "type": "task"}
]
}' | hive hc create
All items are output as JSON lines (one per line):
{"id":"hc-abc1","type":"epic","title":"Implement Authentication",...}
{"id":"hc-abc2","type":"task","title":"Add JWT library","epic_id":"hc-abc1",...}
...
hive hc create "Fix login redirect" --type task --parent <epic-id>
hive hc create "New feature epic" --type epic
# List all open tasks for an epic
hive hc list <epic-id> --status open
# List items assigned to a specific session
hive hc list --session <session-id>
# Show item + full comment history
hive hc show <id>
hive hc create [title]Single-item mode (title as positional arg):
--type epic|task — item type (default: task; tasks require --parent)--desc <text> — item description--parent <id> — parent item ID (required for tasks)Bulk mode (no positional arg, reads JSON from stdin or --file):
--file <path> — read JSON from file instead of stdinhive hc list [epic-id][epic-id] — optional positional arg to filter by epic (returns children only, not the epic itself)--status open|in_progress|done|cancelled — filter by status--session <id> — filter by session IDOutput: JSON lines with --json; colored tree without (use --json when parsing programmatically).
hive hc show <id>Output: JSON lines — item first, then comments in chronological order.
hive hc update <id>--status open|in_progress|done|cancelled — new status--assign — assign to current session--unassign — remove session assignmenthive hc next <epic-id>Returns the next actionable leaf task (open/in_progress, no open/in_progress children).
<epic-id> — required positional arg to scope to an epic--assign — assign to current session and set status to in_progressExits with error if no actionable tasks found.
hive hc comment <id> <message>Adds a comment to an item. All positional args after the ID are joined as the message.
Use sparingly — comments accumulate in the context block and are visible to every subsequent agent reading this epic. Prefer one meaningful note over several incremental ones. For handoffs, prefix with CHECKPOINT:.
hive hc context <epic-id>Assembles context block containing:
Epic title and description
Task counts by status
Tasks assigned to current session (with latest comment) — My Tasks
All open/in-progress tasks not assigned to the current session — Other Open Tasks
--json — output as single JSON object (default: markdown)
hive hc prune--older-than <duration> — remove items older than this (default: 168h / 7 days)--status <status> — statuses to prune, repeatable (default: done, cancelled)--dry-run — show count without removingOutput: {"action":"pruned","count":N}
{
"id": "hc-abc123",
"repo_key": "owner/repo",
"epic_id": "hc-epic1",
"parent_id": "hc-epic1",
"session_id": "mysession",
"title": "Implement login endpoint",
"desc": "Optional description",
"type": "task",
"status": "in_progress",
"depth": 1,
"created_at": "2026-01-01T00:00:00Z",
"updated_at": "2026-01-01T01:00:00Z"
}
Epics have type: "epic" and empty epic_id (they are their own root).
{
"id": "hcc-def456",
"item_id": "hc-abc123",
"message": "CHECKPOINT: auth middleware done, need tests",
"created_at": "2026-01-01T02:00:00Z"
}
| Status | Meaning |
|---|---|
open | Not started, available to claim |
in_progress | Actively being worked on |
done | Completed |
cancelled | Abandoned, will not be completed |
Weekly Installs
1
Repository
GitHub Stars
19
First Seen
Today
Security Audits
Gen Agent Trust HubPassSocketPassSnykPass
Installed on
zencoder1
amp1
cline1
openclaw1
opencode1
cursor1
开源项目教练指南 - 诊断问题、制定行动计划、优化开源项目运营
31,600 周安装