Repo workflow and code intelligence

These skills operate on local repositories, code search, cleanup, external source inspection, and evidence-gated public research.

`k-code-quality`

Field	Value
Use when	editing, reviewing, or refactoring implementation code or repository artifacts
Source	`exact_k-code-quality`
Boundary	implementation-quality details, including minimal edit-scope detail, artifact necessity, and semantic dedupe

k-code-quality routes to narrower skills when the surface is present: k-code-quality-react for React/JSX/TSX/hooks, k-code-quality-tests for tests/fixtures/assertions, and k-code-quality-web for HTML/CSS/accessibility/browser UI. Non-code artifacts (config keys, template variables, generated files, instruction sentences, completions, docs) have readers too; the skill requires identifying them and their generated or rendered outputs before editing and updating them in the same change.

`k-code-quality-react`

Field	Value
Use when	editing React, JSX, TSX, hooks, components, props, state, effects, or client-side UI behavior
Source	`exact_k-code-quality-react`

`k-code-quality-tests`

Field	Value
Use when	adding, editing, reviewing, or debugging tests, fixtures, mocks, snapshots, assertions, or coverage
Source	`exact_k-code-quality-tests`

`k-code-quality-web`

Field	Value
Use when	editing HTML, CSS, DOM markup, layout, responsive styles, accessibility, or browser UI
Source	`exact_k-code-quality-web`

`k-codebase-design`

Field	Value
Use when	designing a module interface, deciding a seam, deepening a module, or making code testable
Source	`exact_k-codebase-design`
Boundary	design vocabulary only; SOP owns compatibility/scope; `k-code-quality` owns implementation style
Pivots	resolves necessary in-scope diagnostic seam/design questions; `k-code-quality-tests` owns test mechanics once settled

`k-diagnosing-bugs`

Field	Value
Use when	diagnosing a hard bug, failure, flake, or performance regression from source and observed failure evidence
Source	`exact_k-diagnosing-bugs`
Boundary	routes into SOP §3.5 plus SOP State-Machine Verification; not the runtime-truth chain for "is X set up right"
Pivots	necessary in-scope seam/design question → `k-codebase-design`; authorized regression cases → `k-code-quality-tests`; no automatic post-fix architecture pass

Reuse existing failure evidence. Source investigation does not require a runnable reproduction first. New probes must resolve a material uncertainty; minimization and competing hypotheses are evidence-driven, not mandatory quotas. Production workers return artifacts without private QA; the root owns one integrated final verification.

Failure assessment identifies product, test, infrastructure, mixed, or unresolved causes. A green retry or a test-only patch does not establish a test-only defect; the original product behavior stays in the acceptance criteria.

`k-prototype`

Field	Value
Use when	building throwaway code to answer a design question (logic/state model, or what a UI should be)
Source	`exact_k-prototype`
Boundary	explicit throwaway exception to artifact necessity; delete or absorb when done
Pivots	closes `k-spec` empirical forks — the verdict returns to the packet's Context line

`k-git`

Field	Value
Use when	local git operations: status/diff/log, branching, committing, pushing, rebasing, conflicts
Source	`exact_k-git`
Boundary	no GitHub side effects; no worktree management; no commit without an explicit current-conversation request; configured remote URLs are never printed verbatim because URL userinfo may contain credentials

`k-worktrees`

Field	Value
Use when	`,w`, `,gh-worktree`, PR/issue worktrees, listing/pruning/removing worktrees
Source	`exact_k-worktrees`
Preference	prefer `,w` / `,gh-worktree` over raw `git worktree`

`k-tmux`

Field	Value
Use when	running, probing, or automating tmux commands, panes, sessions, sockets, or interactive CLIs
Source	`exact_k-tmux`
Boundary	never mutate the default tmux server unless the current target is verified or user-requested

`k-semantic-code-search`

Field	Value
Use when	nontrivial diagnosis/implementation impact, review base context, or SCSI index selection
Source	`exact_k-semantic-code-search`
Boundary	not durable memory; use Agent memory for that

Discover and justify the index before querying it, then trace relevant symbols, callers, and consumers. Confirm snapshot findings against the exact local code. An unavailable or absent index, or an explicit opt-out, uses local rg/symbol evidence with a recorded reason; indexed ,sem queries are not part of that fallback. Which repositories a SCSI server indexes is domain-overlay policy. Simple filename lookup and mechanical edits do not require semantic search.

`k-sem` skill (`,sem` CLI)

Field	Value
Use when	entity history across moves, per-entity blame, entity diff classification
Source	`exact_k-sem`
Tool	`,sem` CLI

Index-free diff, log, and blame are the default surface: diff --format json proves a mechanical-only change (renames, moves, structuralChange: false), and log <entity> follows an entity across file moves where git log -L stops. Indexed queries (impact, context, find, callers, refs, grep, entities) build a multi-gigabyte per-worktree index and do not resolve @kbn/* package specifiers, so the skill forbids them unless the user explicitly asks.

`k-weave`

Field	Value
Use when	preparing merges, previewing/resolving conflicts at function/class granularity
Source	`exact_k-weave`
Tool	`weave` CLI

Setup omits dedicated patterns for .vue, .svelte, .erb, and .hs. Compound names such as .svelte.ts can still match *.ts; preview and direct-driver calls can also attempt entity merging.

`k-public-sources`

Field	Value
Use when	inspecting public source, or synthesizing factual claims across multiple public sources
Source	`exact_k-public-sources`
Boundary	explicit repo URLs stay source-first; multi-source synthesis carries primary evidence for one batched final judgment

The multi-source branch collects and drafts the requested synthesis in one research assignment, retaining exact source identity, supporting passages and uncertainty. The final Verify stage judges material claims together using those artifacts. Unsupported claims are omitted or qualified; there are no per-claim verifiers, post-verification deepening loops or duplicate source fetches for independence. Final judgment preserves the required capability and reports reduced independence when applicable. Verified reusable insights enter the root's final ,ai-kb learning batch, not leaf memory workflows.

`k-jscpd`

Field	Value
Use when	duplicate-code detection during refactor or cleanup
Source	`exact_k-jscpd`
Tool	`jscpd`

`k-knip`

Field	Value
Use when	unused files, dependencies, and exports in JS/TS projects
Source	`exact_k-knip`
Tool	`knip`

`k-cli-skills`

Field	Value
Use when	creating or upgrading CLI tool skills
Source	`exact_k-cli-skills`
Boundary	CLI-tool skill mechanics; general skill craft lives in `k-writing-great-skills`

`k-instruction-boundaries`

Field	Value
Use when	writing or refactoring AI-facing instructions that need hard prohibition boundaries
Source	`exact_k-instruction-boundaries`
Routing	user-invoked (`disable-model-invocation: true`); referenced by `k-writing-great-skills`
Boundary	hard prohibitions by default; affirmative wording only where it sharpens execution

`k-writing-great-skills`

Field	Value
Use when	authoring or refactoring any skill: invocation choice, information hierarchy, leading words, pruning
Source	`exact_k-writing-great-skills`
Routing	model-invoked; auto-loads when authoring/refactoring a skill
Boundary	general skill craft; `k-cli-skills` owns CLI-tool-specific mechanics

`k-walkthrough`

Field	Value
Use when	interactive codebase exploration, architecture tracing, diagrams
Source	`exact_k-walkthrough`
Routing	manual

k-code-quality​

k-code-quality-react​

k-code-quality-tests​

k-code-quality-web​

k-codebase-design​

k-diagnosing-bugs​

k-prototype​

k-git​

k-worktrees​

k-tmux​

k-semantic-code-search​

k-sem skill (,sem CLI)​

k-weave​

k-public-sources​

k-jscpd​

k-knip​

k-cli-skills​

k-instruction-boundaries​

k-writing-great-skills​

k-walkthrough​