Skip to content

Guide · Agent skill

How to use the Pathfinder Codebase Architecture Mapper agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

Pathfinder is an agent skill that makes your coding agent act as an architecture reviewer instead of a coder. It maps a codebase into feature groups with Mermaid flowcharts, finds the same concern implemented in several places, and proposes one simpler unified design. This guide explains what the skill does, how to install it, how to trigger it and what it needs, so you can run an architecture audit before a large refactor.

Pathfinder Codebase Architecture Mapperthedotmack/claude-mem · Apache-2.0

Quick start

  1. 1Download the ZIP. Download the Pathfinder skill ZIP from this page.
  2. 2Unzip into a skills folder. Extract it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder.
  3. 3Or upload in the Claude apps. Go to Settings, then Capabilities, then Skills, and upload the ZIP.
  4. 4Check the requirements. Confirm the agent can read your repository and has subagent support, and that /make-plan and /do are available.
  5. 5Ask for an architecture audit. Use a natural request such as finding the ideal path or auditing architecture before a refactor.
  6. 6Review the feature boundaries. Check 00-features.md before the flowcharts and duplication reports are used.
  7. 7Hand off to /make-plan. Copy the prompts from 04-handoff-prompts.md into /make-plan for each unified system.

What the Pathfinder skill makes the agent do

Pathfinder turns the agent into an orchestrator that splits a codebase into feature groups. For each feature it draws a Mermaid flowchart whose nodes carry file and line references, so every box in the diagram points back to source code.

It then searches across features for duplicated concerns, such as multiple capture paths, parallel queue implementations, duplicated storage or migration code, or parallel parsing layers. Finally it proposes a unified architecture and writes copy-paste prompts for the /make-plan command.

  • Feature inventory with boundaries (00-features.md)
  • One flowchart per feature (01-flowcharts/<feature>.md)
  • Duplication report with file:line evidence (02-duplication-report.md)
  • Proposed unified architecture with a combined flowchart (03-unified-proposal.md)
  • Per-system /make-plan prompts (04-handoff-prompts.md)

Install the Pathfinder skill

Download the ZIP for the skill from this page. Unzip it into your personal skills folder at ~/.claude/skills/ to make it available in all your projects, or into a project's .claude/skills/ folder to limit it to that repository.

You can also upload the skill in the Claude apps. Go to Settings, then Capabilities, then Skills, and upload the ZIP there.

The package contains a single file, SKILL.md, which holds the instructions the agent follows.

  • Personal install: ~/.claude/skills/
  • Project install: .claude/skills/ inside the repository
  • Claude apps: Settings → Capabilities → Skills, then upload the ZIP

How to trigger the skill with natural requests

The skill's description tells the agent when to use it. Ask for it directly in plain language. The source names these trigger phrases: "find the ideal path," asking to unify duplicated systems, or auditing architecture before a refactor.

Be specific about the goal, for example the repository you want mapped and whether you want a duplication audit before a refactor. The skill only produces diagrams, reports and prompts, so say so if you want a plan rather than changes.

  • "Find the ideal path for how our job queues are organised in this repo."
  • "Audit the architecture of this codebase before we split the refactor into tasks."
  • "Unify the duplicated storage code across these features."

What you need before running it

Pathfinder depends on the agent reading your source. It uses grep and file-level evidence, and its own instructions reject diagrams drawn from memory. Make sure the agent can access the repository.

The workflow fans out parallel discovery tasks, so the host agent needs subagent support. The handoff step also relies on the /make-plan and /do commands from the same plugin environment.

  • A code repository the agent can read
  • Subagent support in the host agent
  • The /make-plan and /do commands from the same plugin environment, for the handoff step

Run the workflow and read the output

The skill runs in phases. First a feature discovery subagent proposes feature boundaries, and the orchestrator reviews them before any flowcharts are drawn. Then one flowchart subagent runs per feature, followed by two duplication subagents that work within and across features.

The orchestrator writes the proposal itself. Every duplication claim must cite at least two file:line locations. Review the output folder, named PATHFINDER-<date>/, starting with 00-features.md, because bad feature boundaries make everything after them less useful.

  • Phase 0: feature discovery and boundary review
  • Phase 1: per-feature flowcharts in parallel
  • Phase 2: within-feature and cross-feature duplication hunt
  • Phase 3: unified proposal written by the orchestrator
  • Phase 4: handoff prompts for /make-plan

Safety notes: scripts and network

According to the listing, Pathfinder does not run scripts and does not use the network. It works by reading your source files and writing Markdown and Mermaid files into the PATHFINDER output folder at the repository root.

The skill is explicitly told not to write implementation code. Review any handoff prompts before you pass them to /make-plan, since those prompts are what the next stage will act on.

  • Runs scripts: no
  • Uses network: no
  • Writes report files to PATHFINDER-<YYYY-MM-DD>/ at the repository root

Tips for better results

Give the skill a bounded scope. The workflow is multi-phase and built for large codebases, so pointing it at a small project or a single file adds overhead without much benefit.

Check the subagent citations. The skill rejects subagent reports that lack source citations, but you should still spot-check a few file:line references against the code before acting on the proposal.

  • Use it for large refactors, not single-file changes
  • Spot-check file:line references in the flowcharts
  • Treat the unified proposal as a draft to review, not a decision

What people use it for

  • Mapping an unfamiliar codebase into feature-level flowcharts with cited file and line locations
  • Finding parallel job queues or duplicated storage code before a refactor
  • Drafting a single unified design for systems that grew in parallel
  • Splitting an approved architecture change into per-system planning prompts

Frequently asked questions

How do I install the Pathfinder agent skill?

Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings, then Capabilities, then Skills.

What does Pathfinder need to run?

It needs a code repository the agent can read, subagent support in the host agent because the workflow runs parallel discovery tasks, and the /make-plan and /do commands from the same plugin environment for the handoff step.

Is Pathfinder free to use?

The source lists the skill under the Apache-2.0 licence, which is part of the thedotmack/claude-mem plugin. The source does not state pricing, so check the repository's licence file for the terms that apply.

Does Pathfinder change my code?

No. The skill is told not to write implementation code. It produces diagrams, a duplication report, a proposed unified flowchart and handoff prompts, and the actual changes are left to a separate planning and execution step.

Does Pathfinder run scripts or use the internet?

According to the listing, it runs no scripts and uses no network access. It works from the files it can read in your repository.

Why does Pathfinder say the diagrams are invalid if they are drawn from memory?

The skill depends on grep and file-level evidence. Its instructions say diagrams drawn from memory are not valid, so if the agent cannot read the source, the skill is not suitable.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Pathfinder suits engineers preparing a large refactor, teams learning an unfamiliar codebase, and tech leads who want cited evidence of duplicated logic before splitting work. Skip it for small projects or single-file edits, and if you want the agent to make the changes itself, since it stops before implementation.

Download the skill