Claude Config

Create a Decision Record (DR) in MADR v4 format with automatic numbering. Its subject is not limited to architecture; it covers every decision that is hard to reverse and surprising without context.

Analyzes branch changes and opens a draft pull request. Detects the base branch from where this one was cut, refines the body through a prose review before it goes up, and captures a screenshot and attaches it to the PR when the change touches the UI.

Fix bugs confined to 1-3 files in a development environment. Hand it a filed issue number and that fix carries straight through. Do NOT use for new feature implementation or changes spanning 4 or more files (write the plan via /think and /issue, then hand the number to the build workflow).

Generate a GitHub Issue with a structured title and body. When challenge / research artifacts exist in the conversation, they serve as the body's evidence. When a /think plan draft exists, it moves into the `## Plan` section. Given an issue number, it transfers a plan into a filed issue that has no `## Plan` section.

Break a plan / spec / PRD into independently-grabbable tracer-bullet vertical-slice issues and publish them to GitHub in dependency order. Each issue is one thin slice cutting through every layer. Do NOT use to file a single request (use /issue instead).

Design exploration with adversarial critique by critic-design. Assembles the surviving approach into a structured plan, self-checks it, and returns it to the caller. The issue's Plan section is the plan's only persistent home. Do NOT use for codebase investigation without planning intent (use /research instead).

Removes one harness element at a time, reruns, and judges whether that element moves the result. Elements that do not move it are listed as delete candidates.

Discover design decisions that exist in code but have no DR, and produce a DR promotion candidate list ranked by impact and reversibility. Pairs with adrift, which scans existing DRs for drift against code.

Analyze Git diff, generate a Conventional Commits format message, and run the commit.

Extract recurring patterns from past closed PRs/issues and the research findings in .claude/workspace/research/, verify them against the latest code, and propose them to docs/wiki/ via PR.

Generates and updates .claude/OUTCOME.md interactively. When the file is absent or empty (Behavior blank or TBD only), collects content via AskUserQuestion and writes the stub; when present, shows the current state and applies updates.

Matches a PR's diff against the issue's `## Plan` section and returns unimplemented units, missing tests, and out-of-scope changes.

Inspect whether an issue is in shape to hand to build, returning a verdict (build-ready / needs-plan / needs-fix / needs-split) and the findings. Do NOT use to file an issue (use /issue) or to match a PR against its plan (use /preview).

Analyze Git changes and create a new branch with an appropriate name.

Probe project and technical questions. Findings are positions to be challenged with explicit sources, not conclusions. Phase 6 advisor pass argues against the synthesis before it lands. Do NOT use for design planning or plan generation (use /think instead).

Two-phase challenge that judges whether a discovered problem is real and whether a proposed idea is usable. Phase 1 loops subagent verification and advisor judgment over evidence (OUTCOME.md + parallel subagents) to self-resolve design-tree branches. It asks the user only the irreversible branches that remain and proceeds on stated assumptions for the rest. Phase 2 spawns two critic-design subagents (internal attack / OUTCOME.md attack) as devil's advocate input. The verdict leads the output as a simple GO / NO-GO. Do NOT use for code review findings (use the audit workflow) or outcome assertion (use /assert which has built-in adversarial testing).

Reads a spreadsheet (xlsx, ods, csv) that the Read tool cannot open, converts merged-cell business documents into Markdown, and verifies that no cell was lost.

Web search, page fetch, and GitHub repo exploration via scout CLI.

Reads Google Sheets and Docs through gsheet and gdoc. Authentication comes from a gcloud access token.

Search past Claude Code/Codex sessions via recall CLI.

Structural search and rewrite by AST pattern via ast-grep CLI.

Symbol-level code structure queries via codegraph CLI. Callers, callees, change impact, and symbol source trails.

Internal helper for /pr. Reads PR body (Preview URL + How to Test), drives agent-browser to capture a screenshot (1 step) or a video (2+ steps), and returns the artifact path that /pr passes to gh pr create --attach.

Silent failure detection. Do NOT use for security (use-context-reviewer-security) or readability (use-context-reviewer-readability).

OWASP Top 10 security review. Do NOT use for readability (use-context-reviewer-readability) or test design (use-context-reviewer-testability).

Root cause analysis by eliminating hypotheses.

Code readability review. Do NOT use for security (use-context-reviewer-security), error handling (use-context-reviewer-silence), or test design (use-context-reviewer-testability).

Testable code design review. Do NOT use for security (use-context-reviewer-security) or readability (use-context-reviewer-readability).