Sanitized Maintainer System Profile

This document records the real development setup observed during dry-run and doctor checks. Personal paths, usernames, emails, credentials, local libraries, and secrets are intentionally omitted or replaced with placeholders.

This page is not required for normal installation. It explains the environment that motivated the repository design, so readers can understand why the installer supports Linux, mounted Windows profiles, WSL-backed SageMath, and multiple agent homes. Treat it as an example deployment, not as a requirement for your own machine.

Roots

Substrate

Placeholder

Notes

Linux home

<LINUX_HOME>

Primary development root used for local tests.

Mounted Windows home

<WINDOWS_HOME>

Windows profile inspected from Linux/WSL-style mount.

Execution Topology

The observed setup is best understood as shared research logic plus agent-local installation targets:

  • Codex, Claude, DeepSeek, Copilot, OpenCode, and Antigravity each load skills from their own supported local skill/config locations; OpenClaw is represented as a fake-root-only target for normal installer flows, with reviewed v2 real-system skill-file writes only through openclaw-target-*.

  • This repository holds the reusable skill bodies and dependency metadata.

  • The installer detects which agent homes exist, installs only those targets, and skips absent agents without requiring their tools.

  • Runtime-backed workflows use logical dependencies, not personal paths. For example, a skill asks for python-runtime, tex-runtime, or sage-runtime; precheck decides whether that capability is local, WSL-backed, missing, or degraded.

For a research task, the agent instruction layer chooses the workflow, while the software layer supplies concrete capabilities such as library lookup, document parsing, database access, figure compilation, and math verification.

Detected Agents

Substrate

Codex

Claude

DeepSeek

Linux

present at <LINUX_HOME>/.codex

present at <LINUX_HOME>/.claude

present at <LINUX_HOME>/.deepseek

Windows profile

present at <WINDOWS_HOME>/.codex

present at <WINDOWS_HOME>/.claude

not detected

If an agent home is absent, the installer skips that agent and does not require its dependencies.

Existing Skill Layouts

Agent

Existing layout observed

Installer behavior

Codex

Existing skills under <HOME>/.codex/skills

Primary Codex target. Existing unmanaged files are skipped by default; canonical installs and migrations write here.

Codex optional workspace

Optional <HOME>/.agents/skills when present

Compatibility or workspace-local target, not the default global target.

Claude

Existing skills under <HOME>/.claude/skills; some legacy aliases such as deep-research

Canonical names are used for new installs; aliases are detected and skipped unless migrated.

DeepSeek

Existing skills under <HOME>/.deepseek/skills

Existing unmanaged skills are skipped by default.

The dry-run state had no managed ai-agents-skills instruction blocks yet.

Tool Detection Summary

Tool

Linux observation

Windows-profile observation

Related skills

python-runtime

system Python 3.10 with ssl, venv, and pip

native Windows Python can be detected from C:\Python3*, per-user Python installs, Program Files installs, or PATH-style candidates; mounted checks can verify package markers without running python.exe

deep-research-workflow, zotero, docling, digest skills, graph-verifier, tikz-draw, session-logs

tex-runtime

pdflatex from TeX Live detected

TeX Live under C:\texlive\*\bin\windows and common MiKTeX roots can be detected as present-unverified from a mounted Windows filesystem

tikz-draw

sage-runtime

not detected on Linux PATH

WSL-backed Sage is checked via wsl.exe when runnable, current local WSL paths when precheck runs from WSL/Linux, mounted WSL rootfs paths when present, and WSL ext4.vhdx presence as a degraded inspection gap

sagemath, optional tikz-draw graph mode

Skill-To-Software Relationship

Skill area

Main software or capability

Research planning and synthesis

agent instructions plus optional Python helper runtime

Paper/library workflows

Zotero credentials and local library access are external configuration, not repo content

External paper retrieval

getscipapers-style helper/runtime is treated as an external or runtime-backed dependency

Document parsing

Local-only Docling runtime wrappers, Python, optional OCR tools, and optional local model artifacts

Database lookup

public HTTP APIs; API keys, when needed, are supplied externally

Digest workflows

Python runtime and user-managed topic/feed files outside the repo

TikZ figures

TeX engine; optional SageMath and graph helpers

Math verification

SageMath when available; Python/NetworkX for lightweight graph checks

Multi-agent workflows

agent orchestration instructions; no extra binary required by default

Privacy Boundary

The repo should contain reusable skill logic, docs, and installers. It should not contain personal paths, auth files, credentials, session logs, downloaded papers/books, Zotero databases, Calibre libraries, or local runtime state.

Related pages: Dependencies, Windows, Linux, Audit And Migration.