Frank
08f90c6df3
Retroactive provenance: the 10 pi-era skill snapshots were byte-identical to upstream. Canonical installs now live in ~/.agents/skills tracked by ~/.agents/.skill-lock.json (mattpocock/skills x38 after grilling removal, find-skills from vercel-labs/skills). Repo tier is authored overrides only: clear all 10 tracked upstream snapshots; omp loads the store directly through the agents provider (skills.enableAgentsUser, default on). grilling returns as a real directory — extracted from the store (CLI + lockfile pruned) so Frank can customize it without skills update resurrecting a twin. Docs: skill-store-wiring.md (+index). Pre-cutover snapshot of the old tracked dirs at /tmp/agent-skills-snap-20261018.
2.5 KiB
type, title, description, tags, timestamp, verified
| type | title | description | tags | timestamp | verified | |||
|---|---|---|---|---|---|---|---|---|
| reference | omp Skill Store Wiring | Division of skill tiers on Dank workstations — skills CLI global store, omp agents provider, and the authored repo tier. |
|
2026-10-18 | 2026-10-18 |
omp Skill Store Wiring
Three tiers, one owner each:
| Tier | Path | Owner | Content |
|---|---|---|---|
| Canonical store | ~/.agents/skills/ |
skills CLI |
upstream skills, 38 as of 2026-10-18 (all mattpocock/skills) |
| Provenance | ~/.agents/.skill-lock.json |
skills CLI |
v3 lockfile: source, skillPath, skillFolderHash per skill |
| Authored | agent/skills/ (this repo) |
Frank, via PR | hand-customized overrides; today only grilling |
How omp loads them
omp is NOT a registered agent in the skills CLI (79-agent registry checked
2026-10-18; see Agent Support) —
skills add never writes anywhere under ~/.omp. Instead omp's own agents
provider reads the store directly: skills.enableAgentsUser (default true)
loads ~/.agents/skills/*/SKILL.md. No per-skill symlinks in the repo — the
repo does not mirror the store; it only holds authored overrides.
Precedence (omp omp://skills.md): authored agent/skills/ (native, priority
100) beats the agents provider; identical copies collapse by realpath, differing
same-named copies survive namespaced (agents/<name>) with a startup warning.
Therefore: when authoring an override, remove the skill from the store
(npx skills remove <name> -g) so the lockfile stays truthful and no
namespaced twin haunts the listing. grilling was extracted this way 2026-10-18.
Updating upstream skills
npx -y skills update -g # refreshes everything the lockfile tracks
npx skills check # drift report vs GitHub tree SHAs
Fresh workstation (omp side is two commands — store restore + authored tier arrives with the repo clone):
npx -y skills add mattpocock/skills --all -g
npx -y skills add vercel-labs/skills --skill find-skills --agent universal -g
npx -y skills remove grilling -g # authored copy ships in the repo
Managed tier
agent/managed-skills/ (auto-learn, priority 5) is separate and untouched by
all of the above; authored skills always win name collisions against it.
Related
- Lock File — hash-based update checking
- Agent Support — full CLI agent registry