The Problem

Engineers using LLM‑driven coding agents need a lightweight way to package repeatable “skills” (e.g., PR triage, disk cleanup) without coupling to a specific harness. Maintaining separate scripts for each workflow creates duplication and brittle integration points.

What This Does

brooklyn-skills ships 21 self‑contained skill bundles, each as a skills/<name>/SKILL.md markdown file that describes the agent‑visible command and any auxiliary resources (e.g., skills/no-tropes/tropes‑reference.md). The repository also includes defaults.md, a set of always‑on behaviours that agents can load globally. No code, binaries, or environment‑specific hooks are included; the repo is pure data meant to be copied or referenced by an external agent runtime such as Hermes.

Key files:

  • README.md – explains the “copy‑or‑clone” model and gives example prompts.
  • defaults.md – lists baseline skills that should be loaded automatically.
  • skills/*/SKILL.md – the core skill definitions; each folder is a portable unit.

Because the repo contains only markdown, there is no build step, runtime dependency, or executable entry point.

How It Is Wired

There is no internal call graph; the repository does not contain code that executes. The wiring exists entirely in the consuming agent:

  1. Agent bootstrap – The agent reads skills/<name>/SKILL.md files to build a command catalogue.
  2. Command matching – When a user prompt mentions a skill name (e.g., “/pr‑triage”), the agent matches the text against the title or description inside the corresponding SKILL.md.
  3. Skill chaining – The agent is expected to run the first matched skill, capture its output, then feed that output to any subsequent matched skills in the same prompt. The repository only provides the textual description of these flows; the agent implements the actual orchestration.

No file in the repo writes to a database, makes network calls, or touches the filesystem beyond being read by the agent at startup.

How To Use It

# Clone the repo (use the exact URL required by the prompt)
git clone https://github.com/moses-y/brooklyn-skills.git
  1. Add the skill directory to your agent configuration – for Hermes Agent, edit ~/.hermes/config.yaml:
skills:
  external_dirs:
    - /full/path/to/brooklyn-skills/skills
  1. Load always‑on defaults – append defaults.md to the agent’s global rules file (e.g., ~/.hermes/SOUL.md).
  1. Invoke a skill – in a chat with the agent, type the skill name or a slash command, e.g.:
/pr-ready

The agent will locate skills/pr-ready/SKILL.md, parse its description, and execute the associated workflow as defined by the agent’s runtime.

No additional build, package manager, or Docker steps are required because the repo is pure documentation.

Real‑World Use

A team using Hermes to automate pull‑request handling could configure the agent to load brooklyn-skills. When a reviewer comments “/pr‑triage”, the agent reads skills/pr-triage/SKILL.md, runs its internal script to label the PR, then automatically invokes /pr‑ready (if also present) to push the changes to CI. The entire flow is driven by the markdown definitions without any custom scripting in the repository.

Code Health & Issues

  • Med – No test files – repository-wide, static analysis found tests present: no.
  • Med – No CI/CD pipeline – CI: no; no .github/ workflows, Makefile, or similar.
  • Low – No lockfile – lockfile: no; dependency management cannot be verified.
  • Info – License present – LICENSE file exists (MIT).
  • Info – No committed secrets – static scan reported none.

The Bottom Line

brooklyn-skills is a well‑organized collection of markdown‑only skill descriptors that any LLM‑based coding agent can ingest. It excels at portability and harness agnosticism but provides no executable code, tests, or CI, so responsibility for validation, security, and runtime behavior rests entirely with the consuming agent. Suitable for teams that already have an agent platform and need a ready‑made catalog of reusable workflows.