empire

Trigger when user says: "team review", "have specialists review", "review my changes", "re-review", "review again", "another pass", "ask the team", "specialist review", "/empire-dev:team-review", "have the team look at this", "get specialists to review", "run a team review", "do a specialist review". Spawns parallel specialist subagents to review diffs and consolidates findings. Never posts to GitHub.

Render the canonical PR description body. Use when drafting, writing, updating, or regenerating a "PR description", "PR body", "pull request description", "pull request body", "PR summary", "PR template", "GitHub PR body", or any text bound for `gh pr create --body`, `gh pr create --body-file`, `gh pr edit --body`, `gh pr edit --body-file`, or a GitHub PR template. Triggers on "draft a PR", "write the PR", "summarize this branch for review", "regenerate PR body". MUST be invoked before any `gh pr create --body*` or `gh pr edit --body*` per empire-git rules. Outputs markdown for stdin.

Gate then merge a single pull request safely. Use when the user says "merge this PR", "merge the PR", "land this", "ship this PR", "merge when green". Verifies CI is green (investigates and attempts a fix when red), rebases and resolves conflicts when the branch is behind, triages unresolved review threads intelligently (resolve, fix, or ask — never blindly block), then merges and deletes the branch. Fast-forwards the local base checkout (e.g. updates `master`) when it is checked out.

Trigger when user says: "address review", "address the review", "address review comments", "address PR comments", "handle review comments", "fix review comments", "respond to the review", "reply to review comments", "resolve review comments", "address the feedback", "work through the review", "/empire-dev:address-review". Verifies every unresolved review comment adversarially, evaluates whether the reviewer's proposed change is the best fix, rechecks any alternative it prefers, implements accepted fixes in parallel, and replies to every comment in a direct tone. One compact TLDR confirmation gates all GitHub writes (push + replies).

Trigger when user says: "socratic review", "socratic pr review", "socratic code review", "review this PR socratically", "review the PR with questions", "ask questions on the PR", "question-style review", "leave socratic comments", "/empire-dev:socratic-pr-review". First explains in plain words what the PR does and sanity-checks its direction, then runs team-review, turns findings into short question-style inline comments, walks each past the user with a quick note on the code it touches, and posts ONE review (approve / request changes / comment) at the end. Posts to GitHub only after the user confirms the comments and the final verdict.

Use for `/empire-dev:handoff` and when one coding task should run unattended through the full chain — implement, review, address feedback, open the PR, pass CI, label — delivering a finished pull request, not a check-in per step. Often points at a ticket or spec. Invoking the skill IS the ship-intent signal: it authorizes the full push-and-PR chain. NOT for a single stage in isolation (review, PR body, implement, plan each have their own skills). Flags judgment calls for human review instead of guessing silently. Trigger phrases: "handoff this", "drive this to a PR", "take this and run with it", "implement, review, and open a PR", "build it, have the team review, then push a PR", "do the whole thing", "run this autonomously", "I'm stepping away — carry this to a PR", "ship this for me".

Diagnose system design problems and guide architecture decisions — catches over- and under-engineering, surfaces missing integration points, drives toward a walking skeleton before full build-out. Use when designing a system, when architecture feels too complex or too vague, or when integration points are unclear. Findings stay local. Trigger phrases: "system design", "how should I structure this", "too much abstraction", "under-engineered", "where do I start building", "design this system", "walking skeleton", "/empire-dev:shape".

Break overwhelming development tasks into right-sized, independently deliverable units — diagnoses why tasks stall and applies decomposition patterns. Use when a task feels too big to estimate, when unsure where to start, when blocked by dependencies, or when scope keeps growing. Trigger phrases: "task too big", "can't estimate", "overwhelmed by scope", "where do I start", "break this down", "epic needs breakdown", "slice this up", "/empire-dev:slice".

Evaluate architecture decisions, document trade-offs, select patterns for context. Use when making technology choices, evaluating architectural patterns, writing Architecture Decision Records (ADRs), assessing technical debt, or comparing design alternatives. Findings stay local. Trigger phrases: "architecture decision", "ADR", "which pattern should I use", "evaluate trade-offs", "technology choice", "design pattern selection", "weigh the options", "/empire-dev:weigh".

Batch cleanup across MULTIPLE stale worktrees and orphaned branches. Use when tidying up accumulated worktrees, asking about old branches, cleaning up the workspace, or saying "stale worktrees", "orphan branches", "prune worktrees", "clean up old branches", "purge stale worktrees", "housekeeping". For finishing ONE specific worktree, use worktree-close instead. Also triggers for `/empire-git:worktree-cleanup [--dry-run] [--days N]`.

Finish work in a SINGLE worktree: push, remove the worktree, optionally delete the branch. Use when done with a specific worktree, wrapping one up, pushing and moving on, or no longer needing a parallel environment. Triggers on "close this worktree", "I'm done with this worktree", "wrap up this branch", "push and remove", "tear down this worktree". For batch cleanup of MULTIPLE stale worktrees, use worktree-cleanup instead. Also triggers for `/empire-git:worktree-close [branch] [--push] [--discard] [--force]`.

Answer questions about the worktree toolkit: VSCode integration, gitignore, port offsets, env files, dependency installs, workflow tips. Use when user asks how worktrees work, "how do I open this in VSCode", "why is .env copied", "what about port collisions", "how do worktrees affect main repo", "worktree workflow", "show me the worktree FAQ", or any question about the setup. Also triggers for `/empire-git:worktree-help [question]`.

Read-only inventory of active worktrees: branch, status, sync state, staleness. Use when user asks what worktrees exist, "list worktrees", "show my worktrees", "what's in flight", "which branches have worktrees", "stale worktrees", "any forgotten worktrees", or wants a parallel-work overview. Reports state only, never deletes or modifies. For removal use worktree-close (single) or worktree-cleanup (batch). Also triggers for `/empire-git:worktree-list [--stale]`.

Fold ONE worktree's branch INTO another branch using git merge. Use when combining a worktree branch into a parent, batching small fixes into one branch before opening a single PR, or folding sub-feature branches back together. Triggers on "merge this worktree into X", "fold sub-branches back", "combine worktree branches", "merge feat/X into main locally". Always a real `git merge`, defaults to `--no-ff`. For pushing to remote, use `worktree-close --push` instead. For batch teardown of stale worktrees, use worktree-cleanup. Also triggers for `/empire-git:worktree-merge [branch] --into <target> [--no-close] [--ff]`.

Create or reopen a git worktree for parallel development. Use when user wants to "open a worktree", "spin up a branch", "work on X separately", "work on this in parallel", "start a parallel task", "isolated environment for an agent", "side branch without switching", or mentions worktrees, parallel branches, or stacked work. Also triggers for `/empire-git:worktree-open [branch | task description] [--base <branch>]`.

Diagnose requirements problems; discover real needs vs. stated wants. Use before implementation to separate problem from solution, surface hidden constraints, and bound scope to a viable V1. Trigger phrases: "requirements analysis", "what should I build", "clarify requirements", "is this the right problem", "define scope", "what does the user need", "distill requirements", "/empire-product:distill". Findings stay local — never post to GitHub.

Diagnose why names don't work and guide creation of names that do — brand, product, skill, character, place names, and titles. Use when a name feels wrong but no one can say why, a name family feels disjointed, a name is forgettable, or systematic naming is needed. Trigger phrases: "name this", "this name doesn't feel right", "brand naming", "product naming", "names don't match", "forgettable name", "wrong associations", "mint a name", "/empire-product:mint". Findings stay local — never post to GitHub.

Generate a personal elevator pitch or a repository/project pitch. Trigger phrases: "elevator pitch", "pitch this project", "introduce myself", "personal pitch", "how do I pitch myself", "pitch for this repo", "/empire-product:pitch", "tell me about yourself", "what's a good way to introduce this", "explain this repo in one line", "describe this project", "intro paragraph", "tagline", "one-liner", "GitHub repo description", "package.json description", "what does this do".

Diagnose thinking failures; audit whether reasoning serves inquiry or defense. Use when reasoning feels stuck or circular, a conclusion feels defended not discovered, confidence is high but evidence thin, analysis grows elaborate without growing accurate, or the same approach keeps failing. Trigger phrases: "check my thinking", "am I reasoning well", "why am I stuck", "reasoning feels circular", "probe my logic", "/empire-product:probe". Two modes: self-monitoring (audit own process) and user coaching (diagnose user's thinking with questions, not declarations).

Trigger when user says: "competitor analysis", "compare competitors", "/empire-product:recon", "competitor matrix", "competitor research", "feature gap", "scout competitors", "size up competition", "pricing comparison vs competitors", "positioning analysis", "competitive landscape". Maps competitor pricing, features, positioning, and gaps into a side-by-side matrix with confidence-tagged data and a positioning angle. Different from `/empire-research:compare`, which evaluates tools, libraries, vendors, or architectural choices — NOT competitors. Findings stay local — never posted externally.

Trigger when user says: "vet this idea", "vet idea", "validate idea", "go no go", "pressure test", "is this idea good", "kill the idea", "should I build this", "fatal flaw check", "what do you think of this product", "stress test the idea", "brutal honesty on this idea". Pressure-tests a product idea: web research for competitors and demand, fatal-flaw hypothesis, anti-sycophancy mode, structured go/no-go output with pivots. Findings stay local — never posted externally.

Trigger when user says: "compare libs", "compare frameworks", "/empire-research:compare", "evaluate options", "side by side", "head to head", "X vs Y", "which is better", "tooling comparison", "weigh these options", "decide between these". Closed comparison of tools, libraries, frameworks, vendors, or architectural choices — NOT competitors (use `/empire-product:recon`). User has a known option set; produces a side-by-side matrix on user-defined dimensions and recommends a winner. Findings stay local — never posted externally.

Investigate complex claims: decompose into atomic verifiable components, resolve vague entities, verify each independently, separate confirmed facts from narrative interpretation. Use to fact-check complex claims, investigate narratives mixing fact with interpretation, or when a claim bundles multiple assertions. Trigger phrases: "investigate this claim", "fact-check this", "is this true", "decompose this narrative", "viral content check", "trace this claim", "/empire-research:dissect". Findings stay local — never post to GitHub.

Trigger when user says: "explore options", "what could we do for X", "research approaches", "/empire-research:explore", "investigate approaches", "spawn research team", "what are the options", "options analysis", "explore solutions", "have the team explore". Open-ended exploration: shallow scan enumerates 3–5 candidate approaches, user picks subset to deep-dive, parallel research per approach, consolidated comparison with recommended direction. Findings stay local — never posted externally.

Reconcile empire-* plugin skill-routing snippets into a rules file (project AGENTS.md or user-global ~/.claude/CLAUDE.md). Idempotent. Use when the user asks to "sync empire rules", "update AGENTS.md from empire", "apply empire routing rules", "install empire skill rules", "refresh empire CLAUDE.md", "rewrite empire rules", or after installing/updating an empire-* plugin to write its routing block.

Turn a concept, codebase, flow, or comparison into a diagram instead of a wall of text — terminal-native ASCII (architecture, flowcharts, state machines, sequences, trees, comparison tables); mermaid only when output targets GitHub or a markdown viewer. Use when the user wants something drawn, mapped, or visualized rather than described in prose. Trigger phrases: "draw this", "diagram this", "visualize", "show me a diagram", "map this out", "flowchart", "architecture diagram", "sequence diagram", "state machine", "ascii diagram", "make this visual", "/empire-visual:visualize".