Guardrails / claude

Claude reads the instruction file the repository keeps

claude/instructions-symlinked@v1

CLAUDE.md is a symbolic link to AGENTS.md rather than a second copy of it or nothing at all.

Id	claude/instructions-symlinked
Version	v1
Category	claude
Default severity	warning
Interpreter	python3
Timeout	30 seconds
Violations tolerated	0
Collects	files

Collectors

This guardrail gathers nothing itself. It depends on the collectors below, which the CLI runs once per build before any check, and reads what they found out of GUARDRAIL_FACTS. A collector that collects nothing skips this guardrail rather than failing it.

Collector	Gathers	Inputs it is given
files	Presence, size and line counts of the well known files a repository is expected to carry, plus any extra path the guardrail asks for.	paths

The inputs above are this guardrail's own, passed straight through. Configuring one in buildnote.json changes what is collected, and two guardrails configured the same way share the one collection.

Configuration

{
  "guardrails": {
      "failOn": "error",
      "checks": [
          {
              "use": "claude/instructions-symlinked@v1",
              "severity": "warning",
              "with": {
                  "paths": "CLAUDE.md,AGENTS.md"
              },
              "exemptions": []
          }
      ]
  }
}

Inputs

Input	Description	Default	Environment variable
paths	The alias first, then the file it must be a symbolic link to. Both are relative to the directory the CLI runs in.	CLAUDE.md,AGENTS.md	GUARDRAIL_INPUT_PATHS

How to fix

Point the alias at the file the repository actually keeps:

ln -sf AGENTS.md CLAUDE.md
git add CLAUDE.md

Claude Code reads CLAUDE.md; the agents.md convention every other agent follows is AGENTS.md. A repository with only one of them instructs only some of the agents working in it, and a repository with two files instructs them differently within a month.

Check the link is committed as a link: git config core.symlinks must be true on a Windows checkout, or git writes a text file holding the target's name and both agents read that.

More in claude

• claude/cli-safe-flags. No Claude CLI invocation the build executed asked for its approval step to be skipped.
• claude/cli-structured-output. Every headless Claude CLI invocation the build executed asked for structured output.

All 3 claude guardrails