AI Engineering

Last updated: 2026-09-10.

AI Engineering is a library of Markdown skill packages, guide material, support templates, and protocol distributions for LLM coding assistants. It is built around a simple split:

• docs/ explains the repository for human readers.
• knowledge/ stores the source and derived material consumed by skills and packages.
• protocols/ stores protocol distributions such as CCDP.

Start Here

• Repository overview
• Skill library
• Collaboration framework
• Knowledge library anatomy
• Building and installing
• Protocols
• Contributing
• Origins

What Is Here

The repository currently includes:

• domain/tooling skill packages for Rust, Fennel, Go, Erlang/OTP, C++, JavaScript/Deno, Cobalt, Tailwind CSS, Visual Design, Biome, and Deno lint;
• the methodology-stewardship skill for closing retrospectives, recurring collaboration difficulties, and evidence-backed process recommendations;
• the scientific-methods method skill for practical inquiry, controlled comparison, experiment planning, evaluation rubrics, evidence capture, and regression analysis;
• the laboratory-recordkeeping method skill for research collections, laboratory notebooks, experiment/run records, provenance, preservation, and reconstruction;
• the document-extraction method skill for preparing PDF, EPUB, HTML, and converted-source inputs with structure, media, locator, manifest, readiness, and caveat records;
• the concept-cards method skill for provenance-bearing concept-card extraction, validation, verification, reconciliation, relationships, competency questions, and memory-admission decisions;
• the collaboration-framework composite framework/operational skill, with its framework material under knowledge/;
• reusable support material such as the templates/GUIDE.md support template;
• the Composite Cognition Dispatch Protocol under .

For implementation handoffs, engineering-methods provides From Slice to Implementation Prompt and a CC prompt template, available standalone and through collaboration-framework.

Autonomous Mode requires explicit Operator selection of role/model/settings before execution. It defaults to an implementer and independent reviewer, with stable workspace ownership, complete ledgers and retained useful records. Supporting artifacts remain optional and need-driven; no budget response is required. Its execution and record routes retain existing review, scope and integration gates.

Quick Commands

make help              # list package, validation, and install targets
make all               # build all installable skill zips into target/skills/
make rust              # build target/skills/rust-guidelines.zip
make fennel            # build target/skills/fennel-guidelines.zip
make go                # build target/skills/go-guidelines.zip
make cpp               # build target/skills/cpp-guidelines.zip
make js                # build target/skills/javascript-deno-guidelines.zip
make erlang            # build target/skills/erlang-guidelines.zip
make cobalt            # build target/skills/cobalt-guidelines.zip
make design            # build target/skills/visual-design-system.zip (Markdown methods and text cases; runnable specimens live in examples/design/)
make tailwindcss       # build target/skills/tailwindcss.zip
make deno              # build target/skills/deno-js-linter.zip
make biome             # build target/skills/biome-js-linter.zip and biome-linter.zip
make collab-framework  # build target/skills/collaboration-framework.zip
make document-extraction  # build target/skills/document-extraction.zip
make concept-cards      # build target/skills/concept-cards.zip
make methodology-stewardship # build target/skills/methodology-stewardship.zip
make laboratory-recordkeeping # build target/skills/laboratory-recordkeeping.zip
make print-skill-zips  # list release-uploadable skill zip paths
make check-skills      # validate SKILL.md descriptions
make check-skill-versions # validate source and generated skill versions
make audit-skills      # measure source descriptions and aggregate pressure
make audit-live-skills # measure descriptions actually rendered by Codex
make test-skill-tools  # verify parsing and per-file approval behavior
make check-package-paths
make install           # install built skills into ~/.agents/skills
make test-skill-install # verify destination prompts and install behavior

If the default ~/.agents/skills directory is missing or is a dangling symlink, make install prompts for a destination. When CODEX_HOME is set, the prompt suggests that profile's skills directory. You can enter another path or type cancel. Without an interactive terminal, installation stops with instructions for setting INSTALL_DIR.

INSTALL_DIR accepts either an environment variable or a Make command-line override, including paths containing spaces. For a profile session with CODEX_HOME set:

INSTALL_DIR="$CODEX_HOME/skills" make install
# Or supply a directory directly:
make install INSTALL_DIR="/path/to/profile/skills"

An explicit destination is created if needed and skips the prompt. Use the same INSTALL_DIR override with make uninstall to remove that profile's installed ai-engineering skills.

Skill checks require Python and PyYAML. See Skill Description Maintenance for setup, installed-catalog evaluation, and LLM shortening proposals with per-file approval.

CCDP is packaged separately from installable skills:

make ccdp
make ccdp-package
make check-ccdp-package

Repository Layout

ai-engineering/
├── README.md          # repository orientation
├── Makefile           # package, validation, install, and CCDP targets
├── docs/              # end-user repository documentation
├── knowledge/         # skill source and derived knowledge substrate
├── protocols/         # protocol distributions, including CCDP
├── templates/         # cross-cutting templates, including GUIDE.md
├── scripts/           # package and validation helpers
└── assets/            # README images and other public assets

Current Boundaries

Billo Research holds our labs and their research records. This repository maintains the reusable scientific-methods and laboratory-recordkeeping guidance those collections follow.

The docs/ guides are explanatory wrappers. They should help readers choose, install, build, and contribute without duplicating the full material under knowledge/.

Skill kind and topology are separate. Kind says what a skill is about, such as domain/tooling, framework/operational, or method work. Topology says how a skill composes: an atomic skill is loaded for one clear, self-contained purpose, while a composite skill selects, sequences, routes, governs, composes, etc., multiple loadable components.

The rust skill is an example of an atomic domain/tooling skill. Whereas collaboration-framework is a composite framework/operational skill and remains the daily-driver composer for most users of the ai-engineering repo. To clear up an additional conceptual point, protocols (such as CCDP) are RFC-style documents, not installable skill packages.

License

MIT - see LICENSE.

Individual knowledge bases synthesize material from sources under various licenses, documented inside each skill's own source materials. When in doubt, defer to the original sources.