claude-kit

Collaborative design conversation for any new feature, project, or non-trivial change. Use when I want to think through a problem before building. Phrases like 'let's think through', 'help me design', 'spec this out', 'how should we approach', or any substantial new effort without an existing spec.

Use to clean up local branches and worktrees left over after Branch-and-PR efforts, or when the SessionStart nudge flags reapable OR stranded branches. Triggers: branch cleanup, reap or prune merged branches, recover stranded post-merge commits, leftover or stale local branches, worktree cleanup, too many local branches sitting around.

Neutral, evidence-first evaluation of a decision or judgment where my own preference, ownership, enthusiasm, or a desired conclusion is baked into the framing. Use for non-code go/no-go calls - especially high-stakes, hard-to-reverse, or emotionally-loaded personal and business decisions - and for moments like 'is this a good idea?', 'should I do X?', 'am I being rational about Y?', or 'are you sure?' asked with no new evidence. NOT for code, diffs, specs, or architecture (use the adversarial-reviewer agent), executing already-agreed work, or neutral lookups.

Convene one fresh-context judge (the consultant agent) to rule on a question this session cannot settle. Use mid-execution at the trigger floor - a second failed attempt at the same problem, any BLOCKED that turns on a decision, a debugging dead end, a weighty decision the spec does not cover - and when the operator asks to 'get a consult on X' or wants a 'second opinion on this problem'. The session that needs this rarely feels stuck; it feels almost done, so check the floor, not the feeling. Not a diff review (the adversarial and blind reviewers) and not design-time divergence (design-council).

Use when taking or resuming the machine-coordinator seat, running a coordination loop over several live sessions, or coordinating the machine's sessions across repos. Triggers: claiming the coordinator role, a reconciliation pass over the live roster and the machine's session registry, cross-repo sequencing or handoff brokering, machine-resource contention between sessions, seat handoff to a successor session.

My C# house style. Use whenever writing or modifying ANY C# code: new services, handlers, helpers, MediatR notifications, models, DI registration, or refactoring existing C#. Signature traits: #region organization, section comments ending in periods, grouped fields with label comments. Trigger on any C# work even when style is not named.

Use during finishing-work when a plan reaches Complete, when a new plan is written that should be indexed and cross-referenced, or when asked to tidy, retrofit, or reorganize a project's docs/ folder. Symptoms: completed plans piling up in docs/plans/, a backlog doc that only grows, plans that never reference each other, a docs/ tree with no index.

Convene a read-only, multi-lens design council to pressure-test competing approaches at a genuine architecture fork. Offered by the brainstorming skill when a decision is material and hard to reverse - architecture, schema/data-model, build-vs-buy, a migration direction - and directly invocable when I ask to 'convene the council', 'pressure-test this approach', or 'get multiple angles on this design' before building. Not for reviewing written code (use adversarial-reviewer), non-code judgment calls (use cold), or a session stuck mid-execution (use the consult: one seat with a ruling mandate, convened by the session itself, where the council is multi-lens, sits at design time, and never runs without me present to adjudicate).

Autonomous execution of an approved spec or plan from docs/plans/. Use when I say to proceed, implement, build, or continue an agreed plan, or when resuming a session that has an In Progress plan doc.

Completion pass for a finished effort. Use when all sections of a plan in docs/plans/ are implemented, or I say wrap up, finish, close out, or hand off.

Use when running a kaizen pass on the kit: an explicit kaizen request, accepting an end-of-effort or session-start offer to reflect on captured friction, or applying a pending kaizen brief in the kit repo. Jotting a single friction note does not need this skill; the global capture rule covers that.

Validate and repair this machine's claude-kit installation. Use when the kit was just installed or updated on a machine, when a kit capability misbehaves (hooks, memory tooling, doctrine not loading), or when I ask to run the doctor, check the install, or verify kit setup.

Arm or clear a project-scoped completion leash for a plan run or an ordered sequence of them. Use when I type /kit-goal <plan path>... to hold an autonomous run to completion across a session swap, /kit-goal clear to release it, or /kit-goal to see what is armed. The kit-native, deterministic alternative to native /goal for plan-based runs.

My core operating doctrine: how to think, decide, build, verify, communicate, and stay in scope. Consult at the START of ANY non-trivial task (coding, design, debugging, review, or writing) and whenever unsure how I want work approached. This is the always-apply house ruleset, not an optional reference.

Use when reading the roster of live sessions, messaging another session, or acting on a message one sent. Triggers: the ListAgents or SendMessage tools, cross-session coordination, a sibling session in this or another repo, coordinating with a session working the same tree, a warm consult of a peer session with loaded context, handoff questions to a predecessor session, a compaction boundary a role seat declares or an operator-consent release it is asked to write, or a notify_when_idle subscription. Not a replacement for durable handoffs (the doc still stands alone) and not the consult skill's fresh-context judge.

Use when a review agent returns findings, when I give feedback or a correction, or before implementing a suggestion from either, especially when it seems wrong, unclear, or larger than the problem. Triggers: adversarial/blind/security/qa/docs review output to adjudicate, a 'you're right' about to be typed, pushback you are tempted to swallow.

Use when taking a seat with /role <Seat>, writing or reading a session registry entry, claiming or checking the machine's heavy-process slot, or resolving whether a seat holds the operator's standing delegation. Triggers: /role, a seat takeover or handoff, a file under ~/.claude/coordinator/, scoped direction from a senior seat whose standing is in question.

Guide to writing in the style of Scott Applefeld. Use whenever asked for a document, draft, or other written output that will be sent by Scott Applefeld or intended to mimic his style.

My T-SQL house style. Use whenever writing or modifying ANY SQL: stored procedures, tables, functions, indexes, install or deployment scripts, or ad-hoc queries. Signature traits: shell-then-ALTER deployment, banner-comment headers, leading commas, tab-aligned columns, leading semicolons, audit-logging CATCH blocks that do not re-throw. Trigger on any SQL work even when style is not named.

Use when a session watches a live system it does not own on a repeating loop, waking on a timer or an operator message, checking, intervening, and sleeping again. Triggers: a watch or babysitter session, a loop armed with /loop or a self-authored wake prompt, a runbook to follow each pass, a ledger to carry the board between passes, an operator asking you to keep an eye on a running system.

Root-cause debugging discipline. Use whenever investigating a bug, failure, unexpected behavior, failing test, or production incident - BEFORE proposing any fix. Triggers include 'bug', 'broken', 'failing', 'why is this happening', error reports, and especially any situation where a previous fix attempt did not work. Skip only for trivial fixes where the cause is directly visible.

Use when writing a test, deciding whether a change earns one, choosing which tests to run after a fix, or reading a red or a wall-clock figure. Triggers: a new test or test file, a fix round tempting a full-suite re-run, a suite that reds only beside a busy neighbor, a wall clock that grew, a test reaching for a port, a spawn, or shared state.

Use when creating a skill for this kit, editing one, or deciding whether a wording change to a behavior-shaping skill will actually change behavior. Also use when amending curated prose the kit ships: a skill, an agent charter, the output style, a README, a plan doc, or a doc under docs/. Triggers: adding a new SKILL.md, reworking a skill's rules, correcting a claim a curated document states, a skill that reads well but agents ignore under pressure, or a kaizen change to the kit's own skills.