record-decision

Repository context. Gather first

Collect these with individual Bash calls, one command per call, never combined into a single invocation:

• Current branch, git branch --show-current
• Working tree status (empty = clean), git status --porcelain | head -10

The pipe is the bound and belongs in the command. A read-time cap ("read only the first 10 entries") is not a bound.

• Pointer: when deciding how much of a command's output reaches context, fetch tools reference § Output limits live. As of: 2026-10-11. Recheck trigger: that section changes how much of a valid result reaches context inline.

Treat a failure (not a repository, git unavailable) as an unknown value and carry on. Keep these as separate body Bash calls, one plain command each, rather than pre-compute lines.

• Pointer: when deciding whether a git command can run as a pre-compute line or in a worktree-isolated session, fetch skills § How injected commands run and worktrees § How Claude Code enforces isolation live. As of: 2026-10-11. Recheck trigger: either section changes how a skill's injected commands run or which command shapes an isolated session refuses.

Variables

Arguments: $ARGUMENTS

Purpose

Record one architecture decision into whatever ADR convention the repository already has. The convention is discovered and followed, never prescribed: this skill owns discovery, the next identifier, the record shape, and the single write. Where no convention exists it reports what it searched, offers common shapes, and defers to the human rather than inventing one.

Input contract

The decision comes from $ARGUMENTS and the conversation. A file path the user names is read verbatim. This skill never reads design-threads.md, PLAN.md, or any other planning artifact on its own; those formats belong to the planning capability, and callers pass the decision and its rationale in the invocation text. When context, decision, or consequences are missing, ask for them. Never invent them.