Files and safety

Agentbriefer is local-first. It inspects project files, stores configuration and reusable data on your machine, and writes the instruction outputs you select.

Project files

Path	Ownership
agentbriefer.yaml	Human-reviewed Agentbriefer configuration
AGENTS.md	Generated or synchronized output
CLAUDE.md	Generated or synchronized output
.cursor/rules/agentbriefer.mdc	Generated or synchronized output
.github/copilot-instructions.md	Generated or synchronized output

Project manifests and lockfiles are read for stack detection. Agentbriefer does not modify them.

User data

Installed skills and reusable developer or skill profiles live in the operating system's standard per-user application data directory. The exact base directory follows the platform conventions resolved for Agentbriefer, rather than assuming a hard-coded home path.

Back up custom skills and profiles if they are important organizational assets. Prefer keeping their source in a version-controlled repository and installing from that source.

Preservation rules

• generate owns the complete selected output.
• sync changes only Agentbriefer's marked section and retains surrounding content.
• Required parent directories are created when needed.
• Symbolic-link output paths are refused to avoid writing through an unexpected target.
• Invalid YAML, missing profiles, and unresolved skills are reported instead of silently rewriting intent.

An unknown skill is excluded from rendering and reported by validation. Run doctor before relying on newly generated output.

Source-control checklist

Commit agentbriefer.yaml and the generated files your team consumes. Do not commit private custom skill content or profile data unless it is intentionally safe for every repository reader.

Before merging a policy change:

agentbriefer doctor
agentbriefer sync
git diff --check

Review the semantic change in the generated instructions—not just whether generation succeeded.