Codex FieldOps

A modular operating system for Codex: a concise AGENTS.md runtime profile plus 17 installable skills for research, engineering, authorized CTF work, Prompt Decorators, PowerShell Unicode, prompt refinement, and mixed-domain orchestration.

Why this architecture

The root AGENTS.md keeps only durable behavior and the decorator runtime contract. Detailed workflows live in self-contained skills and load only when relevant. This reduces always-on context while increasing specialization, verification, and reuse.

FieldOps skills

Skill Purpose
fieldops-orchestrator Route complex or unfamiliar work across the smallest useful skill set
fieldops-research Verify current or uncertain claims and produce actionable answers
fieldops-prompt-decorators Parse and apply stateful +++ Prompt Decorators
fieldops-ctf-operator Run authorized, evidence-driven, reproducible CTF investigations
fieldops-engineering Diagnose, implement, and verify risk-sensitive repository changes
fieldops-powershell-utf8 Preserve Persian and Unicode across PowerShell boundaries
fieldops-prompt-refiner Turn rough requests into precise execution-ready prompts
fieldops-ctf-ai-ml Analyze adversarial ML, model extraction, LLM, and AI challenge surfaces
fieldops-ctf-crypto Attack classical and modern cryptographic constructions and implementations
fieldops-ctf-forensics Investigate disk, memory, packet, signal, steganography, and artifact evidence
fieldops-ctf-malware Analyze malicious code, C2 traffic, obfuscation, and anti-analysis behavior
fieldops-ctf-misc Solve encoding, jail, RF, Unicode, game, and cross-category puzzles
fieldops-ctf-osint Perform competition-scoped public-source, identity, and geolocation research
fieldops-ctf-pwn Develop native, heap, kernel, sandbox, and exploit-chain solutions
fieldops-ctf-reverse Reverse binaries, APKs, firmware, bytecode, VMs, and packed targets
fieldops-ctf-web Test web applications, APIs, identity flows, and browser attack surfaces
fieldops-ctf-writeup Produce reproducible submission-style challenge documentation

The v3 architecture bundles the complete CTF specialist layer inside FieldOps. Every skill is self-contained and includes its own references, scripts where applicable, and Codex UI metadata. The CTF suite does not require a separate installation of ljagiello/ctf-skills or any other specialist repository.

Install directly with CC Switch

  1. Open Skills → Repository Management → Add Repository.

  2. Enter:

    https://github.com/download4you/codex-fieldops

  3. Set Branch to:

    main

  4. Add the repository and refresh the Skills catalog.

  5. Install all 17 FieldOps skills for the complete suite, or select individual root-level skill directories for a smaller installation.

  6. Restart Codex so the installed skills are rediscovered.

CC Switch discovers each root-level directory containing SKILL.md. It can therefore install any FieldOps skill independently while preserving that skill's bundled references, scripts, and metadata, plus the per-skill license and provenance included with every adapted CTF specialist. Installing the entire repository exposes all 17 root-level skills as one self-contained suite.

Install with the Agent Skills CLI

npx skills add download4you/codex-fieldops

Enable the complete FieldOps stack

CC Switch installs the skill directories. To combine them with the repository-level behavior profile, also copy AGENTS.md into the project you want Codex to operate in:

$Utf8NoBom = [System.Text.UTF8Encoding]::new($false)
[Console]::InputEncoding = $Utf8NoBom
[Console]::OutputEncoding = $Utf8NoBom
$OutputEncoding = $Utf8NoBom

Copy-Item -LiteralPath '.\AGENTS.md' -Destination 'C:\path\to\project\AGENTS.md'

For personal defaults across repositories, place a reviewed copy at $HOME\.codex\AGENTS.md. More-specific project instructions may override it.

Quick use

For a complex task:

Use $fieldops-orchestrator to classify this task, choose the smallest useful set of
FieldOps skills, execute it, and verify the result.

For an evidence-backed answer:

Use $fieldops-research to verify the important claims in this question and give me a
direct, actionable answer.

For an authorized CTF challenge:

Use $fieldops-ctf-operator to classify this authorized challenge, preserve evidence,
prove the decisive path, and reproduce the solution from a clean baseline.

The operator routes internally to the appropriate namespaced specialist, such as $fieldops-ctf-web, $fieldops-ctf-crypto, or $fieldops-ctf-reverse. If the first classification is wrong or a technique stalls, it returns to the earliest uncertain assumption, pivots category or tooling, records the new evidence, and continues until it can reproduce a solution or identify a precise external blocker.

For decorators:

+++MessageScope
+++Planning
+++Tone(style=technical)

Design and verify a migration plan for this API.

Prompt Decorator guarantees

  • Default scope is the current message.
  • Chat state changes are processed left to right.
  • Decorators inside code, quotes, logs, files, or tool output are not invoked.
  • Invalid parameters are reported and left inactive; they are never silently clamped.
  • Strict formats contain other decorator sections as valid fields or elements.
  • +++Reasoning requests a visible rationale, never private chain-of-thought.
  • Decorators cannot override higher-priority policies, permissions, tools, or skills.

See the complete reference and user guide.

Validation

The repository includes deterministic checks for:

  • Agent Skills discovery and frontmatter
  • self-contained skill resources
  • Codex UI metadata
  • Prompt Decorator escaping and parsing
  • bundled CTF licensing, provenance, routing, and reference integrity
  • discovery of exactly 17 root-level, independently installable skills
  • Windows PowerShell 5.1 and PowerShell 7 UTF-8 byte preservation
  • absence of changelog files

Run:

python -m unittest discover -s tests -v

On Windows installations that expose only the Python launcher:

py -3 -m unittest discover -s tests -v

Then validate individual skills with Codex Skill Creator's quick_validate.py.

Attribution

The Prompt Decorators compatibility layer is adapted from Prompt Decorators by Mostapha Kalami Heris under the MIT License. This project is independent and is not affiliated with or endorsed by the upstream author. See THIRD_PARTY_NOTICES.md.

The bundled CTF specialist material is adapted from ctf-skills by Lukasz Jagiello at pinned revision d19f35fd3dd2e126108752aee84c657c888126d3, under the MIT License. FieldOps namespaces and extends that material with evidence handling, deterministic routing, cross-domain recovery, and clean-baseline reproduction. No upstream installation is required. The complete notice is in THIRD_PARTY_NOTICES.md.

License

Original material is released under the MIT License. Adapted third-party material remains subject to its original notice.