claude-config

Use when the user says /org-design <org>, "analyze the org I inherited", "inherited org analysis", "span of control / SPOF / on-call distribution review", or wants a descriptive read of the org structure they just walked into during a senior eng leader ramp. The analyze mode reads a manual org/structure.md + stakeholder memory graph into a 7-section analysis under ~/repos/onboard-<org>/decisions/. The scenario mode projects a reorg (split-team, add-headcount, merge-teams, change-reporting, reduce-headcount), validates it structurally, and gates on explicit user review before writing. reduce-headcount adds a machine layoff-acknowledgment gate (machine-enforced deliberateness; the human confirmation behind the flag is prose-bound, not machine-verified). Do NOT use for codebase architecture (use /architecture-overview).

Use when the user says /risk-register, "add a risk", "review risks", "escalate R-N", "ack R-N", "resolve R-N", or "show my risk register" during a 90-day ramp or any initiative. Tracks org-level technical risks in any initiative workspace. Six actions: add, review, ack, escalate, resolve, list.

Use when the user says /strategy-doc <org>, "draft my 90-day plan", "review the 90-day plan", or "challenge the 90-day plan" during a senior eng leader ramp. Phase 1 supports the 90-day-plan mode only — collates /swot + /stakeholder-map + /architecture-overview + free-form notes/*.md into a 7-section markdown artifact under ~/repos/onboard-<org>/decisions/. Cross-org RFC mode is Phase 2 (separate spec).

Use when the user explicitly invokes /glossary, asks to canonicalize a domain term in ./CONTEXT.md, or wants to record agreed-upon project-specific terminology so future sessions and downstream artifacts (ADRs, SDRs, systems-analysis output) stay consistent. Also invoked by define-the-problem and systems-analysis at end-of-skill via the caller-hook contract to offer canonicalization for terms resolved during planning. Write-only format-owner for per-project ./CONTEXT.md. Lazy-create on first term resolution; never auto-write — user approval gates every entry. Do NOT use for: casual term mentions in conversation, code-level naming questions, architectural primitives (Module, Interface, Adapter, Seam — those belong in LANGUAGE.md owned by architecture-overview), general programming concepts, or decisions about why a term was chosen (those belong in docs/superpowers/decisions/ or an ADR).

Use when the user says /ring-review-tiered, "ring review these PRs", "antagonistic ring on this batch", "review PR #N with the ring", or asks to run adversarial / critic review across one or more pull requests. Calibrates critic count to per-PR risk tier — trivial PRs skip the ring, surgical PRs get one cross-dimension critic, medium PRs get two dimension-targeted critics, large/boundary/security PRs get the full 5-critic ring plus arbiter. Use even when the user names a uniform critic count — surface the tier and let them override. Do NOT use for solo single-PR review where the user explicitly asks for the full ring; that's `pr-review-toolkit:review-pr`.

Use when the user says /swot, "landscape analysis", "SWOT analysis", "strengths and weaknesses", or wants to capture, review, or challenge organizational observations.

Use when the user says /onboard <org>, "scaffold a new ramp", "set up onboarding workspace for <org>", or starts a new senior eng leader role. Day-0 scaffolder plus 90-day ramp orchestrator for a per-org workspace: scaffold, cadence nags, 1:1 capture/sanitize with confidentiality enforcement, and graduation are ready; live Calendar scan is deferred (manual paste only). Do NOT use for codebase / architecture onboarding (see `architecture-overview` skill, issue #44).

Use as the mandatory front door for ALL planning, brainstorming, design, and "let's build/add/change/plan X" work — including prompts that claim the problem statement is already done, assert authority ("CTO approved"), cite sunk cost ("contract signed"), name a deadline ("ship by Friday"), or request jumping straight to brainstorming or solutions. The skill runs in one of two forms — Expert Fast-Track or the full five-question sequence — never zero. Does not apply to bug fixes or refactors.

Use when evaluating what a proposed change will touch — dependencies, blast radius, failure modes, or second-order effects — or when the prompt names a scoped problem and says "plan this out", "what could go wrong", "think through the impact", or similar. Also triggers in the planning pipeline after `define-the-problem` and before `superpowers:brainstorming`. Do NOT use when the user is only asking to implement or fix something with no design question attached.

Use when someone asks to sketch, wireframe, mockup, or visually map a design before detailed design — user flows, refactors, migrations, architecture changes, integration diagrams, sequence/interaction flows, state machines, or data pipelines. Also triggers during brainstorming after an approach is selected, per the planning pipeline.

Use when the user says /swarm-batch-dispatcher, "batch these issues", "ship N PRs in parallel from issues X Y Z", "run a parallel swarm batch", "dispatch builders for issues #N #M", or asks to take a list of GitHub issues from "ready" to "merged PRs" in one coordinated pass. Owns the full pipeline: worktree creation, parallel builder dispatch, coordinator-side commit / push / PR opening, CI-poll-with-branch-update loop, sequential merge in dependency order, and worktree teardown. Use even when the user names a single issue — the skill still applies (it just dispatches one builder). Do NOT use for one-shot code changes that don't start from a GitHub issue, and do NOT use for spec authoring alone (spec PRs ride a different workflow — see "When NOT to use" below).

Use when the user says /adr, "create an ADR", "new decision record", "document this decision", or "supersede ADR". Also triggers when a technical decision needs formal documentation with status tracking and lifecycle management.

Use when the user says /sdr, "create an SDR", "system design record", "write a system overview", "design doc for a new service", "blueprint", or wants to document a system-level design. Routes to one of four canonical SDR templates (System Overview, Service/Component Creation, Data Design, Blueprint) based on artifact type. Do NOT use when the decision is a single architectural choice — use /adr instead. Do NOT use when evaluating a tool or framework for adoption — use /tech-radar instead. Do NOT use when documenting a deviation from a tenet — use /tenet-exception instead.

Use when the user says /stakeholder-map, asks to build a stakeholder/political-topology map for a new leadership role, wants to review coverage of their relationship-building, or asks for meet-in-what-order guidance during onboarding.

Use when the user says /catalog, "what skills are available", "show me the toolkit", "list my skills/agents/rules", or wants an in-session inventory of claude-config rules, skills, agents, and templates without leaving the session to grep docs.

Slash-invoked: /improve-codebase-architecture. Surfaces deepening opportunities in a codebase via deep/shallow module vocabulary, the deletion test, and seam discipline. Produces a numbered candidate list, then drops into a grilling loop on the user-selected candidate, with optional parallel interface-design exploration. Do NOT use for whole-system mapping or inventory (use /architecture-overview), a single architectural choice (use /adr), a system-level design record (use /sdr), or a tool/framework adoption evaluation (use /tech-radar).

Use when the user says /present, "create a presentation", "make slides", "build a deck", or wants to create or update a Slidev presentation.

Slash-invoked discovery-mode skill that scans multiple repos and produces a 4-file landscape bundle (inventory, dependencies, data flow, integrations) using the canonical LANGUAGE.md vocabulary. Use when a new senior eng leader joins and asks to "walk me through the architecture", "map the services across our repos", or needs a credible whole-system mental model on day 3-7 of a ramp. Do NOT use for single-repo deepening grading (use /improve-codebase-architecture), a single architectural choice (use /adr), a system-level design record (use /sdr), or tool/framework adoption (use /tech-radar).

Use when the user says /1on1-prep, "prep for my 1:1", "1:1 with", "capture my 1:1", or wants to prepare for or record notes from a one-on-one meeting.

Use when the user says /new-project, "scaffold a project", "set up a new repo", "initialize project", or wants to start a fresh codebase with standard conventions.

Use when the user says /tenet-exception or proposes a formal, documented deviation from an established engineering tenet. Also triggers for "request a tenet exception", "deviate from a tenet", or "exception to tenet".

Use when the user says /SKILL_NAME, "<trigger phrase>", or <situation that warrants this skill>. One sentence per trigger class — name WHEN to invoke, not WHAT the skill does. Do NOT use when <anti-trigger — situation where another skill or no skill fits better>. Avoid trigger overlap with existing skills (audit `skills/*/SKILL.md` per #73).

Use when the user says /cross-project, "what repos does this affect", "cross-repo impact", "cross-project dependency analysis", or "who else uses this API". Also triggers when evaluating whether a change in one repo breaks consumers in other local repos.

Use when the user says /tech-radar or asks to evaluate, adopt, or compare a technology, framework, or tool for formal adoption. Also triggers for "should we adopt X", "add to tech radar", or "technology assessment".

Draw professional diagrams on a live Excalidraw canvas via MCP. Use when asked to create architecture diagrams, flowcharts, system maps, comparison visuals, or any visual explanation. Covers setup, 10 visual techniques, design preferences, layout best practices, and the screenshot-iterate loop.