The Dev Toolchain is a portable library of standardized AI agent skills and automation workflows. Built under the SKILL.md open standard, it establishes a modular, language-agnostic platform to orchestrate consistent developer workflows, enforce commit compliance, and coordinate structured incident root cause analysis across diverse LLM clients.
This repository is built as a robust, modular library:
- Every tool has a guide: The manual (
SKILL.md) tells AI agents how to execute the skill. - Orchestration scripts automate tasks: Some tools include an optional shell script to handle terminal commands automatically.
The toolchain contains the following core engineering skills:
| Skill | Directory | Primary Purpose | Status |
|---|---|---|---|
| Commit Prepper | prepare-commit/ |
Enforces staging hygiene, repository state audits, and drafts semantic commit specs. | Active |
| ADR Generator | adr-gen/ |
Standardizes architectural pivots and trade-off matrices with immutable indexing. | Active |
| Tests Generator | tests-gen/ |
Scaffolds language-specific, table-driven unit test suites without external AST dependencies. | Active |
| Plan Generator | plan-gen/ |
Scaffolds a local plan.md template to sequence PRs and coordinate execution steps to prevent goal drift. | Active |
| Code Review Auditor | fresh-eye/ |
Performs pre-commit code quality, safety, and test coverage sanity audits on local diffs. | Active |
| RCA Generator | rca-gen/ |
Standardizes incident post-mortem documentation and timelines under docs/incidents/. |
Active |
Every tool inside this repository follows a simple pattern:
[tool-name]/
├── SKILL.md # [Required] The manual file telling the agent what to do
└── [tool-name].sh # [Optional] The script that runs the commands automatically
This is a declarative text file that contains:
- What the tool does.
- The rules and constraints to follow.
- A checklist to verify that everything works.
If present, the automation script runs the terminal audits and orchestration commands automatically. Diffs and log commands are set to output raw text directly to prevent terminal page hangs.
[ Code changes are made ]
│
▼
[ AI agent reads SKILL.md manual ]
│
▼
/───────────────────────────────\
< Does an automation script exist? >
\───────────────────────────────/
/ \
Yes / \ No
▼ ▼
[ Run script automatically ] [ Follow manual rules directly ]
│ │
▼ │
[ Telemetry logs printed ] │
│ │
\ ┌──────────────┘
\ /
▼ ▼
[ Task is completed cleanly! ]
Instead of copying files, you register the folder path so the AI always reads the original source files. This prevents duplicate files and version drift.
Modern AI tools scan your open folder automatically. If they find a SKILL.md file, they load it instantly. No setup command is needed.
Most platforms let you install skills by running a simple registration command pointing to the local folder or the remote GitHub repository URL:
-
From GitHub:
agent skill install <repo-url> --path <folder-name>
-
From Your Machine:
agent skill install <local-folder-path>
| Platform | Registration Method | How to Do It |
|---|---|---|
| Agy | User Directory / Project Path | Project-Specific: Copy the folder into your project agent folder: cp -r [tool-name] .agents/skills/Global: Copy into the global skills path: cp -r [tool-name] ~/.gemini/antigravity-cli/skills/[tool-name] |
| Codex | Chat Interface Installer | Type $skill-installer in the Codex chat window and provide the folder path to install. |
| GitHub Copilot | GitHub CLI Extension | Install the skill directly using the GitHub CLI: gh skill install ./[tool-name] |