A personal marketplace of Claude Code plugins. Each one solves a problem I kept hitting while engineering with agents, from code getting written before the spec was settled to drafts that ship in someone else's voice.
The premise: agents didn't make the fundamentals optional. They made them enforceable.
Add the marketplace once:
/plugin marketplace add NCMcClure/agentic-engineering
Then install whichever plugins you want:
/plugin install but-first-planning@agentic-engineering
Skills arrive namespaced: /but-first-planning:spec-0-init,
/agentic-teamwork:teamwork-init, and so on.
Code written before the spec is settled is a loan against your future self.
A numbered planning pipeline (spec → grill → architect → plan → publish)
plus a build loop of coordinated TDD agents, all landing in one .plan/
directory, with an autopilot that runs the whole graph hands-off.
Read more →
Mixed-skill teams shouldn't rely on discipline to keep prototypes off main. Deterministic guardrails block commits to protected branches, force-pushes, secrets, junk files, and failing tests. The hooks run inside Claude Code and again in the terminal, with CI as the backstop for anything that slips through. GitHub and GitLab are both first-class. Read more →
A staged, autonomous optimizer for whatever language your repo speaks (Python, JS/TS, Go, and Rust are detected out of the box). It organizes the file tree into an AGENTS.md-hubbed progressive-disclosure layout (orientation hubs an agent reads before descending), decomposes god-files, and deepens shallow architecture, though every change sits behind CI-faithful validation with revert-on-red and nothing is ever pushed. Read more →
Claude Code dynamic workflows are programs; you should be able to see them.
An Unreal-Blueprint-style node canvas that compiles graphs to runnable
.claude/workflows/*.js and round-trips them back, plus a skill for
hand-authoring without the GUI. Read more →
Most ideas don't fail because they're wrong; they fail because they're presented in bullet points. Craft knowledge for the alternative: how narrative arc, visual design, and motion combine so the audience follows the idea instead of reading past it. Read more →
Everything an agent drafts for you ships in someone else's voice. A calibrated, tightly line-budgeted profile of how you write auto-loads into every session, skills audit text for AI tells by hunting clusters of co-occurring patterns rather than lone buzzwords, and reflective hooks fold your style feedback back into the profile. Prose only: code and commit messages stay out of scope. Read more →
Every line an agent writes gets paid for more than once: at generation, at
review, and in every future session that loads it as context. code-diet makes
the agent build at the lowest rung that works (a 7-rung ladder with hard
safety carve-outs, injected as a token-budgeted kernel), reviews diffs with
script-located candidates and cuts applied behind revert-on-red, and tracks
every deliberate shortcut as a debt: marker with an expiry trigger. Evolved
from the MIT-licensed ponytail plugin; borrowed stays labeled as borrowed.
Read more →
A plugin's true cost is invisible until it's installed, and authoring a good one means guessing at standards nobody wrote down. This is both halves of that problem on one rubric: create-plugin interviews you, scaffolds the layout, and authors against the six evaluation dimensions; evaluate-plugin scores any path or git URL into an HTML scorecard with an adopt/rework verdict; improve-plugin applies the fixes ranked by how many points each one buys back. Read more →
The contribution contract (layout invariants, the versioning rule, how to
add or update a plugin) lives in .claude/rules/dev-rules.md.
bash scripts/validate.sh checks the enforceable parts; CI runs it on every
push. Agents maintaining this repo read the same rules you do.
MIT; see LICENSE.