How Field-Journal Stores and Retrieves Previous Task Solutions to Avoid Redundant Work

Field-journal eliminates redundant reverse-engineering and penetration-testing work by storing every completed task as a self-describing Markdown file and forcing AI agents to read relevant precedents before acting.

The field-journal system in the zhaoxuya520/reverse-skill repository acts as a version-controlled knowledge base that transforms individual task completion into reusable organizational memory. By combining structured storage, mandatory precedent loading, and automation helpers, it ensures that AI-driven skills never solve the same problem twice.

Storage Layout: How Field-Journal Archives Task Solutions

All journal entries reside under skills/field-journal/ and follow a strict naming convention: YYYY-MM-DD_<scenario>.md. For example, 2026-08-06_cortex-m-msc-firmware-self-keyed-rotate-xor.md captures a specific firmware reverse-engineering session.

Core Structure

The directory contains three mandatory components that enforce consistency:

• _template.md — Defines required sections (Scope, Evidence, Finding, etc.) and includes a de-identification checklist that must be completed before committing.
• _index.md — An auto-generated chronological index of all entries, maintained by skills/scripts/extract-summaries.ps1.
• Precedent Libraries — Three high-level knowledge repositories that summarize reusable patterns:

Precedent File	Purpose
precedent-auth.md	Authorization context (must be read first for any operation)
precedent-reverse.md	Reverse-engineering operation patterns
precedent-pentest.md	Penetration-testing operation patterns

These precedent files are ordinary Markdown documents stored alongside dated entries, making them immediately readable without specialized parsing.

Retrieval Mechanism: How Field-Journal Surfaces Previous Solutions

The system enforces precedent retrieval through two mechanisms: a central rule engine and explicit skill directives.

RULES.md Hot-Path Enforcement

In RULES.md (line 36), a critical rule states:

"Hesitating about whether an operation is allowed → read precedent-reverse.md or precedent-pentest.md"

This forces any AI interpreter to load stored experience when uncertainty arises. The rule acts as a guardrail that prevents autonomous action without consulting relevant historical context.

NOW Directives in Skill Files

Individual skill definitions embed explicit retrieval commands. For example, in [skills/radare2/SKILL.md](https://github.com/zhaoxuya520/reverse-skill/blob/main/skills/radare2/SKILL.md) line 13:

### NOW

读取 `../field-journal/precedent-reverse.md`

When executed, the AI interpreter:

1. Parses the directive
2. Opens the referenced file via standard file-read operations
3. Injects the precedent contents into the LLM context window

Because field-journal uses plain Markdown, no external database or query language is required—simple filesystem operations suffice.

Eliminating Redundant Work

When a new task matches a previously recorded scenario, the AI surfaces the existing journal entry or summarized precedent instead of reconstructing the solution. This design treats each completed task as a training example for future operations.

Automation Helpers for Field-Journal Maintenance

Three scripts ensure the knowledge base remains current and compliant:

• skills/scripts/extract-summaries.ps1 — Rebuilds _index.md by scanning all journal entries and prepending new records.
• skills/scripts/scan-leaks.ps1 — Validates that entries respect the de-identification checklist from _template.md before PR submission.
• CONTRIBUTE-BACK.md — Restricts pull requests to files within skills/field-journal/, ensuring the journal remains the sole mutable artifact for experience sharing.

Practical Examples: Adding and Retrieving Field-Journal Entries

Creating a New Journal Entry

#!/usr/bin/env bash

# Create a new entry for a pentest exploit discovered on 2026-08-15

ENTRY="skills/field-journal/2026-08-15_exploit-xor-cipher.md"
cp skills/field-journal/_template.md "$ENTRY"

# Fill in the placeholders manually or with an editor

vim "$ENTRY"

# Update the index and commit

bash skills/scripts/extract-summaries.ps1
git add "$ENTRY" skills/field-journal/_index.md
git commit -m "[field-journal] pentest: XOR cipher exploit"

Retrieving a Precedent Within a Skill

### NOW

读取 `../field-journal/precedent-pentest.md`   <!-- pulls prior pentest workflow -->

The interpreter loads precedent-pentest.md and presents its documented workflow to the LLM, enabling direct reuse.

Validating De-identification

# Scan the new entry for unreplaced placeholders before commit

$file = "skills/field-journal/2026-08-15_exploit-xor-cipher.md"
.\skills\scripts\scan-leaks.ps1 -Path $file

Summary

• Storage: Field-journal stores task solutions as dated Markdown files in skills/field-journal/ with standardized templates and auto-generated indexes.
• Retrieval: RULES.md mandates precedent loading when AI agents hesitate, while NOW directives in skill files trigger explicit reads.
• Reuse: Plain Markdown format eliminates parsing complexity—standard file operations enable knowledge injection into LLM contexts.
• Governance: PowerShell scripts automate index maintenance and compliance checks, while contribution rules isolate journal changes.

Frequently Asked Questions

What makes field-journal different from a conventional wiki?

Field-journal enforces programmatic retrieval through RULES.md and NOW directives rather than relying on human browsing. The hot-path rule at line 36 of RULES.md makes precedent consultation mandatory, not optional, ensuring AI agents cannot bypass historical knowledge. This embedded retrieval mechanism transforms passive documentation into active system memory.

How does field-journal prevent sensitive data leaks?

The _template.md file includes a de-identification checklist that must be completed before any commit. The scan-leaks.ps1 script validates compliance automatically, checking for unreplaced placeholders or residual sensitive content. Only the skills/field-journal/ directory is mutable in contribution workflows, creating a controlled surface for experience sharing.

Can field-journal work without AI agents?

Yes—the underlying structure functions as a conventional documentation system. Human practitioners can browse dated entries in _index.md or consult precedent files directly. However, the retrieval mechanism's full value emerges when paired with an AI interpreter that parses NOW directives and RULES.md constraints.

Why use Markdown instead of a database?

Markdown's human readability and universal parseability eliminate dependencies on query languages or specialized drivers. As implemented in zhaoxuya520/reverse-skill, a simple file-read suffices to inject precedent knowledge into any LLM context. This design prioritizes transparency and version-control compatibility over structured storage performance.

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