skillview-cli

Quick-reference card for driving the skillview CLI from inside an agent loop. The companion document agents_onboarding.md has the full agent-facing recipe book — read it once, then come back here for the cheat sheet.

Install

bash
cargo install skillview

Commands by task

| task | command | |---|---| | see what's here | skillview agents · skillview roots · skillview stats --pretty | | find a skill | skillview list --name <q> --format ids | | inspect one | skillview show <id\|name\|path-substring> | | find duplicates | skillview dups --sort size | | rank by usage | skillview usage --top 20 | | copy-pasteable recipes | skillview examples | | per-command help | skillview <cmd> --help |

Filter cheatsheet (list)

Filters compose with AND semantics:

--agent claude|codex|cursor|agents|unknown
--tier  primary|secondary
--root-kind claude-global|claude-project|codex|cursor|agents-generic|unknown
--name <substring>                  # case-insensitive
--has-usage | --min-usage N
--min-tokens N | --max-tokens N
--validation-failed
--dups-only [--dup-kind exact|near]
--sort agent-name|name|agent|tier|usage|tokens|sessions|path
--limit N
--format json|jsonl|tsv|ids|paths|names

dups accepts: --exact, --near, --min-size N, --agent, --root-kind, --sort {size|similarity|kind}, --limit, --format.

usage accepts: --agent, --min-mentions N, --top N, --include-low, --sort {mentions|sessions|recent|name}, --format (no paths here).

Identifying skills across calls

• name — what humans say; not unique (two dup-recipe skills can share a name across agents).
• id (s_<n>) — stable across runs against the same tree (sorted by path),

but a new file inserted lexicographically earlier will shift downstream ids.

• path — absolute filesystem path. Globally unique. Use this if you need a

durable handle.

Speed tips

A full scan of $HOME can take minutes. If you don't need clustering or usage data, always pass:

bash
skillview --no-similarity --no-usage <subcommand>

This drops scan time to sub-second on typical machines.

Output shape

The default invocation (and scan) emits:

jsonc
{
  "schema_version": 2,
  "generated_at": "2026-05-13T...Z",
  "roots":    [ { "id": "r_0", "kind": "claude-global", "path": "..." } ],
  "skills":   [ /* full skill records w/ frontmatter, tokens, validation, usage */ ],
  "clusters": [ { "id": "c_0", "kind": "exact|near", "similarity": 1.0, "members": [...] } ],
  "stats":    { /* counts + timings */ }
}

Schema: <https://github.com/galElmalah/skillview/blob/main/schema/skillview.schema.json>.

When you're stuck

bash
skillview --help                  # top-level overview
skillview <cmd> --help            # filters + flags for that subcommand
skillview examples                # curated recipe book (no scan, instant)

Everything the CLI can do is reachable from those three commands without guessing.