Skip to content

rchalamala/engineering

v1.0.0MIT

Native agent execution with focused design, debugging and verification skills.

Engineering

A curated skill plugin for Claude Code, Codex, Devin, dsh and Cursor. Focused design, debugging and verification skills load in the native agent loop when needed. Independent work and review can use subagents without making every implementation step a separate agent.

Install

Claude Code:

/plugin marketplace add rchalamala/agent-skills
/plugin install engineering@rchalamala

Codex:

codex plugin marketplace add rchalamala/agent-skills
codex plugin add engineering@rchalamala

For Devin, install the plugin in your personal scope and start a new session.

devin plugins install rchalamala/agent-skills

Invoke /engineering:engineering in the new session. Devin reads the included Claude plugin manifest. See Devin's plugin reference.

For dsh on the Nix-managed Linux devbox, update dotfiles and switch to the new generation. New dsh sessions discover the skills in the managed Claude plugin's skills/ directory.

Cursor: import rchalamala/agent-skills as a GitHub marketplace in Customize, then install engineering.

Load engineering for software work, or add this to your agent instructions:

For engineering work, load the engineering skill.

The plugin has no startup hooks, credentials, personal memory or model defaults. It does not activate another pack's orchestration. The engineering skill owns workflow and testing policy; its focused skills supply reusable practices. Project instructions and explicit user requests take precedence.

Included skills

  • Native-loop workflow: engineering.
  • Design and testing: codebase-design, domain-modeling, prototype, tdd.
  • Explicit interviews/surveys: grill-me, grill-with-docs, improve-codebase-architecture, with their grilling dependency.
  • Transfer: handoff.
  • Diagnosis and quality: systematic-debugging, verification-before-completion, receiving-code-review.

Matt Pocock's skills and Superpowers are MIT-licensed. Their licenses are in licenses/; upstream.lock.json records source revisions, selected directories, namespace transformations and hashes. Copies retain supporting files. Manual skills also include Codex's explicit-invocation policy. Upstream references to superpowers: are translated to this plugin's engineering: namespace; its TDD dependency resolves to Matt's tdd. Recorded per-file patches make generic project paths explicit and remove repeated test-seam approval prompts.

Maintenance

The source lives in dotfiles/agent-plugin/ in the monorepo. The public repository is a generated mirror, not an editing destination. Machine configuration loads this same source. Exported history replaces monorepo messages and identities; only this directory's contents are published.

To update an upstream, inspect the diff from its locked revision, recopy only selected skill directories and their references, apply the recorded namespace transformation, preserve explicit-invocation policies, and update the file hashes and revision. Review dependencies before accepting new skill calls. Bump the version in all plugin manifests together.

python3 -m venv .venv
.venv/bin/python -m pip install --require-hashes --requirement tests/requirements.txt
.venv/bin/python scripts/check.py
.venv/bin/python -m unittest discover -s tests

This checks distribution integrity, not agent effectiveness. Compare the native agent with and without the plugin on identical tasks in isolated workspaces. Test activation and non-activation, behavior, integration, interruption recovery, and delivery. Run each task repeatedly. Keep correctness separate from time and cost; an unavailable check is untested, never a pass.