Forge技能：从会话记录自动提取AI知识，提升智能体学习效率

forge by boshu2/agentops

245 周安装量

222 GitHub Stars

GitHub

安装命令

npx skills add https://github.com/boshu2/agentops --skill forge

AI/机器学习自动化知识管理

🇨🇳中文介绍

Forge Skill

通常通过 SessionEnd 钩子自动运行。

从会话记录中提取知识。

工作原理

SessionEnd 钩子运行时执行：

ao forge transcript --last-session --queue --quiet

这将会话加入队列，等待知识提取。

标志

标志	默认值	描述
`--promote`	关闭	处理 `.agents/knowledge/pending/` 中的待处理提取项，并提升到 `.agents/learnings/`。整合了原有的提取技能功能。

提升模式

给定 /forge --promote：

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

联系我们

提升步骤 1：查找待处理文件

ls -lt .agents/knowledge/pending/*.md 2>/dev/null
ls -lt .agents/ao/pending.jsonl 2>/dev/null

如果未找到待处理文件，则报告“没有待处理的提取项”并退出。

提升步骤 2：处理每个待处理文件

对于 .agents/knowledge/pending/ 中的每个文件：

读取文件内容
验证其是否包含必填字段 (# Learning:、**Category**:、**Confidence**:)
复制到 .agents/learnings/（保留文件名）
从 .agents/knowledge/pending/ 中删除源文件

提升步骤 3：处理待处理队列

if [ -f .agents/ao/pending.jsonl ] && [ -s .agents/ao/pending.jsonl ]; then
  # 处理每个已排队的会话
  cat .agents/ao/pending.jsonl
  # 处理后清空队列
  > .agents/ao/pending.jsonl
fi

提升步骤 4：报告

Promoted N learnings from pending → .agents/learnings/
Queue cleared.

完成。 报告后立即返回。

给定 /forge [路径]：

步骤 1：识别记录

使用 ao CLI：

# 挖掘最近的会话
ao forge transcript --last-session

# 挖掘特定记录
ao forge transcript <路径>

没有 ao CLI： 查看最近的对话历史并手动提取学习内容。

步骤 2：提取知识类型

在记录中查找以下模式：

类型	信号词	权重
决策	"decided to"、"chose"、"went with"	0.8
学习	"learned that"、"discovered"、"realized"	0.9
失败	"failed because"、"broke when"、"didn't work"	1.0
模式	"always do X"、"the trick is"、"pattern:"	0.7

步骤 3：写入候选内容

写入到： .agents/forge/YYYY-MM-DD-forge.md

# Forged: YYYY-MM-DD

## Decisions
- [D1] <decision made>
  - Source: <where in conversation>
  - Confidence: <0.0-1.0>

## Learnings
- [L1] <what was learned>
  - Source: <where in conversation>
  - Confidence: <0.0-1.0>

## Failures
- [F1] <what failed and why>
  - Source: <where in conversation>
  - Confidence: <0.0-1.0>

## Patterns
- [P1] <reusable pattern>
  - Source: <where in conversation>
  - Confidence: <0.0-1.0>

步骤 4：建立搜索索引

if command -v ao &>/dev/null; then
  ao forge markdown .agents/forge/YYYY-MM-DD-forge.md 2>/dev/null
else
  # 没有 ao CLI：将高置信度的候选内容自动提升为学习内容
  mkdir -p .agents/learnings .agents/ao
  for f in .agents/forge/YYYY-MM-DD-*.md; do
    [ -f "$f" ] || continue
    # 提取置信度（数值型或分类）
    CONF=$(grep -i "confidence:" "$f" | head -1 | awk '{print $NF}')
    # 将分类转换为数值：high=0.9, medium=0.6, low=0.3
    case "$CONF" in
      high) CONF_NUM=0.9 ;; medium) CONF_NUM=0.6 ;; low) CONF_NUM=0.3 ;; *) CONF_NUM=$CONF ;;
    esac
    # 如果置信度 >= 0.7 则自动提升
    if (( $(echo "$CONF_NUM >= 0.7" | bc -l) )); then
      cp "$f" .agents/learnings/
      TITLE=$(head -1 "$f" | sed 's/^# //')
      echo "{\"file\": \".agents/learnings/$(basename $f)\", \"title\": \"$TITLE\", \"keywords\": [], \"timestamp\": \"$(date -Iseconds)\"}" >> .agents/ao/search-index.jsonl
      echo "Auto-promoted (confidence $CONF): $(basename $f)"
    fi
  done
  echo "Forge indexing complete (ao CLI not available — high-confidence candidates auto-promoted)"
fi

步骤 5：报告结果

按类型统计的提取项数量
forge 输出的位置
准备提升为学习内容的候选内容

锻造的候选内容从第 0 级开始：

Transcript → /forge → .agents/forge/ (Tier 0)
                           ↓
                   Human review or 2+ citations
                   OR auto-promote (confidence >= 0.7, ao-free fallback)
                           ↓
                   .agents/learnings/ (Tier 1)

自动运行 - 通常通过钩子触发
提取，而非解释 - 捕获所说的内容
按置信度评分 - 并非所有提取项都同等重要
排队等待审核 - 候选内容需要验证

SessionEnd 钩子调用

钩子触发： 会话结束时运行 session-end.sh

发生的情况：

钩子调用 ao forge transcript --last-session --queue --quiet
CLI 分析会话记录，寻找决策、学习、失败、模式
CLI 将会话 ID 写入 .agents/ao/pending.jsonl 队列
下一次会话开始时触发 /forge --promote 来处理队列

结果： 会话记录自动加入队列等待知识提取，无需用户操作。

用户说： /forge <路径> 或 "mine this transcript for knowledge"

发生的情况：

Agent 识别记录路径或使用 ao forge transcript --last-session
Agent 扫描记录以寻找知识模式（决策、学习、失败、模式）
Agent 按置信度（0.0-1.0）为每个提取项评分
Agent 将候选内容写入 .agents/forge/YYYY-MM-DD-forge.md
Agent 使用 ao forge markdown 为 forge 输出建立索引
Agent 报告提取数量和候选内容位置

结果： 记录被挖掘以获取可重用知识，候选内容已准备好供人工审核或通过 2+ 次引用进行提升。

问题	原因	解决方案
未找到提取项	记录缺少知识信号或 ao CLI 不可用	检查记录是否包含决策/学习内容；验证 ao CLI 是否已安装
置信度评分低	信号弱或对话模糊	将会话重点放在具体决策和明确的学习内容上
forge --queue 失败	CLI 不可用或权限错误	手动将会话元数据追加到 `.agents/ao/pending.jsonl`
重复的 forge 输出	同一会话被多次锻造	写入前检查 forge 文件名；ao CLI 会自动处理去重

🇺🇸English

Forge Skill

Typically runs automatically via SessionEnd hook.

Extract knowledge from session transcripts.

How It Works

The SessionEnd hook runs:

ao forge transcript --last-session --queue --quiet

This queues the session for knowledge extraction.

Flags

Flag	Default	Description
`--promote`	off	Process pending extractions from `.agents/knowledge/pending/` and promote to `.agents/learnings/`. Absorbs the former extract skill.

Promote Mode

Given /forge --promote:

Promote Step 1: Find Pending Files

ls -lt .agents/knowledge/pending/*.md 2>/dev/null
ls -lt .agents/ao/pending.jsonl 2>/dev/null

If no pending files found, report "No pending extractions" and exit.

Promote Step 2: Process Each Pending File

For each file in .agents/knowledge/pending/:

Read the file content
Validate it has required fields (# Learning:, **Category**:, **Confidence**:)
Copy to .agents/learnings/ (preserving filename)
Remove the source file from .agents/knowledge/pending/

Promote Step 3: Process Pending Queue

if [ -f .agents/ao/pending.jsonl ] && [ -s .agents/ao/pending.jsonl ]; then
  # Process each queued session
  cat .agents/ao/pending.jsonl
  # After processing, clear the queue
  > .agents/ao/pending.jsonl
fi

Promote Step 4: Report

Promoted N learnings from pending → .agents/learnings/
Queue cleared.

Done. Return immediately after reporting.

Manual Execution

Given /forge [path]:

Step 1: Identify Transcript

With ao CLI:

# Mine recent sessions
ao forge transcript --last-session

# Mine specific transcript
ao forge transcript <path>

Without ao CLI: Look at recent conversation history and extract learnings manually.

Step 2: Extract Knowledge Types

Look for these patterns in the transcript:

Type	Signals	Weight
Decision	"decided to", "chose", "went with"	0.8
Learning	"learned that", "discovered", "realized"	0.9
Failure	"failed because", "broke when", "didn't work"	1.0
Pattern	"always do X", "the trick is", "pattern:"	0.7

Step 3: Write Candidates

Write to: .agents/forge/YYYY-MM-DD-forge.md

# Forged: YYYY-MM-DD

## Decisions
- [D1] <decision made>
  - Source: <where in conversation>
  - Confidence: <0.0-1.0>

## Learnings
- [L1] <what was learned>
  - Source: <where in conversation>
  - Confidence: <0.0-1.0>

## Failures
- [F1] <what failed and why>
  - Source: <where in conversation>
  - Confidence: <0.0-1.0>

## Patterns
- [P1] <reusable pattern>
  - Source: <where in conversation>
  - Confidence: <0.0-1.0>

Step 4: Index for Search

if command -v ao &>/dev/null; then
  ao forge markdown .agents/forge/YYYY-MM-DD-forge.md 2>/dev/null
else
  # Without ao CLI: auto-promote high-confidence candidates to learnings
  mkdir -p .agents/learnings .agents/ao
  for f in .agents/forge/YYYY-MM-DD-*.md; do
    [ -f "$f" ] || continue
    # Extract confidence (numeric or categorical)
    CONF=$(grep -i "confidence:" "$f" | head -1 | awk '{print $NF}')
    # Normalize categorical to numeric: high=0.9, medium=0.6, low=0.3
    case "$CONF" in
      high) CONF_NUM=0.9 ;; medium) CONF_NUM=0.6 ;; low) CONF_NUM=0.3 ;; *) CONF_NUM=$CONF ;;
    esac
    # Auto-promote if confidence >= 0.7
    if (( $(echo "$CONF_NUM >= 0.7" | bc -l) )); then
      cp "$f" .agents/learnings/
      TITLE=$(head -1 "$f" | sed 's/^# //')
      echo "{\"file\": \".agents/learnings/$(basename $f)\", \"title\": \"$TITLE\", \"keywords\": [], \"timestamp\": \"$(date -Iseconds)\"}" >> .agents/ao/search-index.jsonl
      echo "Auto-promoted (confidence $CONF): $(basename $f)"
    fi
  done
  echo "Forge indexing complete (ao CLI not available — high-confidence candidates auto-promoted)"
fi

Step 5: Report Results

Tell the user:

Number of items extracted by type
Location of forge output
Candidates ready for promotion to learnings

The Quality Pool

Forged candidates enter at Tier 0:

Transcript → /forge → .agents/forge/ (Tier 0)
                           ↓
                   Human review or 2+ citations
                   OR auto-promote (confidence >= 0.7, ao-free fallback)
                           ↓
                   .agents/learnings/ (Tier 1)

Key Rules

Runs automatically - usually via hook
Extract, don't interpret - capture what was said
Score by confidence - not all extractions are equal
Queue for review - candidates need validation

Examples

SessionEnd Hook Invocation

Hook triggers: session-end.sh runs when session ends

What happens:

Hook calls ao forge transcript --last-session --queue --quiet
CLI analyzes session transcript for decisions, learnings, failures, patterns
CLI writes session ID to .agents/ao/pending.jsonl queue
Next session start triggers /forge --promote to process the queue

Result: Session transcript automatically queued for knowledge extraction without user action.

Manual Transcript Mining

User says: /forge <path> or "mine this transcript for knowledge"

What happens:

Agent identifies transcript path or uses ao forge transcript --last-session
Agent scans transcript for knowledge patterns (decisions, learnings, failures, patterns)
Agent scores each extraction by confidence (0.0-1.0)
Agent writes candidates to .agents/forge/YYYY-MM-DD-forge.md
Agent indexes forge output with ao forge markdown
Agent reports extraction counts and candidate locations

Result: Transcript mined for reusable knowledge, candidates ready for human review or 2+ citations promotion.

Troubleshooting

Problem	Cause	Solution
No extractions found	Transcript lacks knowledge signals or ao CLI unavailable	Check transcript contains decisions/learnings; verify ao CLI installed
Low confidence scores	Weak signals or vague conversation	Focus sessions on concrete decisions and explicit learnings
forge --queue fails	CLI not available or permission error	Manually append to `.agents/ao/pending.jsonl` with session metadata
Duplicate forge outputs	Same session forged multiple times	Check forge filenames before writing; ao CLI handles dedup automatically

Weekly Installs

195

Repository

boshu2/agentops

GitHub Stars

198

First Seen

Feb 2, 2026

Security Audits

Gen Agent Trust HubPass SocketPass SnykPass

Installed on

codex193

claude-code192

opencode191

github-copilot191

kimi-cli191

gemini-cli191

AI Elements：基于shadcn/ui的AI原生应用组件库，快速构建对话界面

56,200 周安装

Forge技能：从会话记录自动提取AI知识，提升智能体学习效率

🇨🇳中文介绍

Forge Skill

工作原理

标志

提升模式

相关 Skills

提升步骤 1：查找待处理文件

提升步骤 2：处理每个待处理文件

提升步骤 3：处理待处理队列

提升步骤 4：报告

手动执行

步骤 1：识别记录

步骤 2：提取知识类型

步骤 3：写入候选内容

步骤 4：建立搜索索引

步骤 5：报告结果

质量池

关键规则

示例

SessionEnd 钩子调用

手动记录挖掘

故障排除

🇺🇸English

Forge Skill

How It Works

Flags

Promote Mode

Promote Step 1: Find Pending Files

Promote Step 2: Process Each Pending File

Promote Step 3: Process Pending Queue

Promote Step 4: Report

Manual Execution

Step 1: Identify Transcript

Step 2: Extract Knowledge Types

Step 3: Write Candidates

Step 4: Index for Search

Step 5: Report Results

The Quality Pool

Key Rules

Examples

SessionEnd Hook Invocation

Manual Transcript Mining

Troubleshooting

最新 Skills