Glossary
Short definitions, each backed by numbers we actually counted. No aspirational language; measured write-ups linked where they exist.
Agent instruction drift is the gradual divergence of an AI agent's instruction files across machines and checkouts, until no one can say which copy is correct.
Instruction file sprawl is the accumulation of agent instruction file copies across a machine or team — created by tools, not by decisions.
AGENTS.md is a convention for a single instruction file that multiple AI coding agents read, adopted by Codex CLI, Copilot, Cursor, Windsurf, Amp, and Devin.
CLAUDE.md is the instruction file Claude Code reads before acting — project-level at the repo root and user-level under ~/.claude/.
Read-back verification is re-reading the target system after a write and comparing IDs in both directions, instead of trusting the write's success response.
Zero-touch agent deployment is pushing approved agent instructions, skills, and memory to every machine without anyone editing local copies by hand.
Agent skills are packaged instruction sets (often SKILL.md files) that extend what an AI agent can do for a specific task or domain.
Agent memory files are persistent notes an AI agent keeps between sessions — user preferences, project context, and learned corrections.
AI agent governance is the control layer over what AI agents run on: inventory of their instruction assets, approval before changes ship, deployment, and audit.
An agent asset inventory is a single verified list of every skill, instruction file, and memory an organization's AI agents run on, across tools and machines.
Sync ghost entries are records that remain in a local database after the original item was deleted on the platform — the ledger says it exists; the platform says it never did.
Content-hash deduplication identifies duplicates by hashing the content itself (e.g. sha256) rather than matching titles or metadata.
A single source of truth for agent rules is one canonical rule document from which every tool-specific instruction file is derived — necessary, but not sufficient, to stop drift.
Agent drift detection is comparing the instruction files actually present on each machine against the approved canonical version, and flagging divergence.
An agent approval workflow requires a designated reviewer to approve changes to agent instructions or skills before they deploy anywhere.
An agent configuration audit trail records who changed which agent asset, when, what was approved, and where it deployed — append-only.
Connect one workspace and see every skill, rule, and memory your team has in play — in about ten minutes.
No credit card. Works with what you already run.