agentic-engineering

Plugin agentic-engineering in mjenkinsx9/agentic-engineering. Commands read from the repo's manifest — 12 days ago.

Skills (32)

• authoring-skills

  Route a request to create, write, build, author, or improve an Agent Skill to the right workflow. Use when the user wants to make a new skill, turn a repeated task or prompt into a skill, write a SKILL.md, or sharpen an existing skill that mis-triggers or underperforms—even if they do not name a com

• babysitting-pr

  Keep one open PR merge-ready until it lands — subscribe to its activity and, on each CI failure, review comment, or merge conflict, decide whether to fix it, escalate it, or skip. Invoke the `babysitting-pr` skill with `<owner/repo#N or PR URL>` when you want a PR watched hands-off through to merge.

• bugfix

  Lightweight lane for small, well-understood fixes: reproduce and prove the root cause, then write a compact single-phase plan and execute it through the normal review pipeline. Invoke the `bugfix` skill with `<what-is-broken>`. Explicit-trigger only; routes anything bigger to the full `spec` → `plan

• build-skill

  Author or sharpen a portable Agent Skill by doing the work first, then codifying it. Invoke build-skill explicitly with a skill name. Scope one task, drive three or more live practice runs, codify SKILL.md and tests.md from those runs, review against the predictability lens, and verify before promot

• cleanup-worktrees

  Remove git worktrees that are clean and fully merged into main, and only those. Use when asked to clean up, sweep, or prune worktrees. Checks each worktree's status is empty, refreshes `origin/main`, and confirms the branch is merged via both `git branch --merged` and a merge-base ancestry check aga

• cmux

  Drive cmux (the terminal-multiplexer / agent-fleet CLI) from natural language — open, inspect, prompt, read, and tear down cmux workspaces and panes, and fan out fleets of Claude Code / Codex / pi agents across isolated git worktrees. Use whenever asked to deploy, spawn, check on, or tear down cmux

• code-audit

  Correctness-and-intent review of the current worktree's changes. Invoke the `code-audit` skill with `[low|medium|high|max]`. Fans out by-concern reviewer subagents, adversarially verifies every finding, and writes a read-only report to code-reviews/. Not refactor, security, performance, or testing —

• codebase-design

  Shared vocabulary and principles for designing deep modules — depth, seams, adapters, leverage, locality, the deletion test. Use when designing or reviewing a module's interface, deciding where a seam goes, deepening a cluster of shallow modules, making code more testable, exploring alternative inte

• commit

  Commit the current work only after the repo's own checks pass. Use when asked to commit, check in, or save work to git, when a finished task needs committing, or when asked to add pre-commit hooks (husky, pre-commit) to a repo. Discovers the repo's gate (hooks, package scripts, CONTRIBUTING, CI), ru

• dashboard

  Start, inspect, or stop the worktree pipeline dashboard through the installed toolbelt entry point. Invoke the `dashboard` skill with `start` or `stop`.

• diagnosing-bugs

  Diagnose a bug by proving its cause before changing code — tighten the feedback loop, reproduce and minimise, hypothesise, instrument, then fix behind a regression test that was red first. Use when chasing a bug, a crash, a failing or flaky test, a stack trace, an unexpected null/undefined, or a reg

• dogfood

  Live-test a skill in this repo to confirm it works as intended — the Definition-of-Done check applied to a skill. Validates the SKILL.md statically, checks registration + invocation policy, dogfoods it on a real input across a happy and an edge path, verifies the skill's own contracts hold (read-onl

• domain-modeling

  Build and sharpen a project's domain model. Use when the user wants to pin down domain terminology or a ubiquitous language, record an architectural decision, or when another skill needs to maintain the domain model.

• execute

  Launches or resumes the executor on an approved or in-progress plan. Invoke the `execute` skill with `<plan-file>`. Pre-flights the plan, spawns the executor subagent to implement it autonomously, then relays the result and offers the review stack — `review-plan` plus `code-audit`. Explicit-trigger

• grilling

  Stress-test a plan, design, or decision by interviewing the user one question at a time until shared understanding is reached. Use when the user says 'grill me on this', 'stress-test this plan/design', 'interview me about this', 'poke holes in this', or asks to be challenged on an approach before bu

• handoff

  Compact the current conversation into a handoff document so a fresh session can resume the work without re-deriving context. Captures the live working state — current goal, what's half-done, the exact next action, open questions — not a permanent record. Writes to the repo-local handoffs/ directory

• improve-codebase-architecture

  Scan a codebase for architectural deepening opportunities — shallow modules, leaky seams, low-locality code — present them as a visual report of before/after candidates, then grill through whichever one the user picks. Explicit-trigger only; a heavyweight periodic architecture review, invoked as the

• new-app

  Bootstrap a new or existing application repository into the Agentic Engineering spec-to-operations workflow. Use when starting an application, preparing an empty repository for agent-assisted development, or adding the workflow artifacts and local instructions to an existing app before product work

• plan

  Interactive planning skill — grills you to alignment, then writes a structured Markdown implementation plan that an executor agent can follow and a reviewer agent can check against. Explicit-trigger only; invoke the `plan` skill and never auto-start.

• prototype-spike

  Run a bounded, throwaway experiment that answers one design question before planning — \"does this approach even work?\", \"which of these two actually holds up?\", \"what does this really cost?\". Invoke the `prototype-spike` skill with `<the design question>` when a genuine design uncertainty bloc

• research

  Investigate a question against high-trust primary sources and capture the findings as a cited Markdown file in the repo. Use when the user wants a topic researched, docs or API facts gathered, or reading legwork delegated to a background agent. Not for codebase exploration (read the code or use a re

• resolving-merge-conflicts

  Resolve git merge, rebase, or cherry-pick conflicts safely — reconstruct the result that honors both sides' intent, or abort when that isn't possible. Use when a merge/rebase/cherry-pick stops with conflicts, when `git status` shows unmerged paths, or when a pull leaves conflict markers in files. Wo

• review-plan

  Plan-conformance review — checks executed work against its approved plan's success criteria and AGENTS.md's Definition of Done. Invoke the `review-plan` skill with `<plan-file>`. Re-runs every phase's validation itself, maps each changed line to a plan task, fans out per-criterion checkers, and writ

• sandbox

  Run risky or destructive dev work inside a disposable Docker container instead of the live machine. Use when a task involves destructive git operations (rebase/reset/filter experiments), installing hooks or packages into a repo to see what breaks, running an unfamiliar installer or untrusted code, o

• security-audit

  Exploitability review of the current worktree's changes — hunts for vulnerabilities the diff introduced or newly exposed, not correctness or style. Invoke the `security-audit` skill with `[low|medium|high|max] [--full] [--comment]`. Maps trust boundaries, fans out attack-class hunters, adversarially

• ship

  Open an evidence-bearing PR for the current branch — re-runs the repo's gate and self-review at the exact HEAD being shipped, refuses on red/stale/self-caught findings, creates the PR with per-command proof in the body, and requests the repository's configured automated review, if any. Invoke the `s

• skill-safety-scan

  Safety scan of an agent skill before it lands in `skills/` — reads the skill's SKILL.md plus its references/assets and flags prompt-injection/instruction-override, exfiltration, and unsafe tool use, so a malicious or careless skill cannot be promoted unreviewed. Invoke the `skill-safety-scan` skill

• spec

  Interactive product-specification skill that turns a rough idea, current conversation, or wayfinder ticket into an approved product behavior contract saved under specs/. Invoke the `spec` skill with `<idea>`, `--from-conversation`, or `--from-wayfinder <issue>`. Explicit-trigger only; not technical

• tdd

  Run a strict test-driven development loop for implementation work. Use when the user asks to build or fix something test-first, mentions TDD or red-green-refactor, wants integration-style behavior tests, or an approved plan marks a phase as strict TDD. Not a planning or review skill; it is the execu

• todo-cleanup

  Reviews and clears a repository TODO/backlog file by removing completed entries and converting every remaining untracked work item into GitHub issues before deleting it. Use when the user asks to clean TODO.md, migrate backlog items to GitHub issues, remove done TODOs, or ensure TODO entries are iss

• updating-npm-package

  Safely move an existing npm dependency to a newer version — check the registry for the latest, read the changelog across the span, auto-apply semver-safe minor/patch bumps and prove them against the repo's gate, and for majors produce a migration validation report for a human instead of applying. Us

• wayfinder

  Plan a huge chunk of work — more than one agent session or one `plan` document can hold — as a shared map of decision tickets on GitHub issues, and resolve them one at a time until the way to the destination is clear. Use for multi-session, foggy efforts where the open decisions aren't all visible y

Subagents (1)

• executor

  Autonomous executor for an approved or resumable in-progress plan. Given the path to a plan file, implements it phase by phase on its own branch with per-phase commits, surfacing only on a real blocker or unsettled decision. Launched by the /execute skill; not for direct user invocation.

Similar plugins