Skills & Built-ins

Memory and steering were already shared across vendors. Skills are the third shared leg: reusable capabilities — a name, a when-to-use description, a procedure, optionally helper scripts — authored once and usable by any agent.

How the layer works

• Neutral source of truth. A committed agent-skills/<name>/SKILL.md — vendor-neutral markdown — is the single definition; it travels with the repo, like memory/.
• Universal runtime. The AGENTS.md "Skills" section is the baseline: when a task matches a skill's description, the agent reads and follows that SKILL.md. Because the agent is the runtime, this works on any vendor with no engine.
• Thin per-vendor adapters. For runtimes with a native skill/command system, the tool regenerates pointers across six vendor targets for native auto-trigger. Adapters are gitignored and regenerated (never copies), so the neutral skill never drifts.
• Migration promotes, never flattens. A vendor's existing skills (e.g. .claude/skills/) are promoted into agent-skills/ (originals preserved under legacy/), not folded into steering — skills are procedures, not rules.

The six adapter targets: .claude/skills/, .gemini/commands/, .cursor/rules/, .kiro/, .github/skills/ (GitHub Copilot), and .agents/skills/ (Google Antigravity). Regeneration is itself a runnable script (sync-adapters), so it never depends on an agent improvising the recipe.

The judgment-vs-arithmetic boundary

Real cross-vendor use surfaced a recurring failure class: a capable agent silently does only part of a multi-step ritual. The fix pattern, applied every time, became a core principle:

Mechanize the arithmetic, not the judgment.

The deterministic, mechanical parts of a memory ritual become runnable helpers so an agent can't silently skip them; every act of judgment stays with the agent and the human.

The seven built-ins

Installed into every enabled repo and tool-managed (marked provenance: agent-memory-builtin). They sort cleanly along that boundary:

Built-in	Role	Side
memory-lint	Deterministic integrity verifier (nine checks; Python and Node)	arithmetic — read-only
refresh-metadata	Recompute every fact's tier/usage from the ledger	arithmetic
archive-fact	Perform the archive move safely (truncation-proof)	arithmetic
sync-adapters	Regenerate the six vendor adapters	arithmetic
harvest-knowledge	Distil durable facts from the team's docs	judgment-assisted
second-opinion	Snapshot for a clean-memory reviewer	judgment-assisted
apply-critique	Bounded, human-gated apply loop	judgment-assisted

Zero overhead by default

The built-ins are tool-managed (fork under a new name to customize, or upstream a genuine fix). Like everything else, they are installed ≠ run.

To write your own, see Author a Skill. For the full design, see the Skills Layer design note.