How do I install the Verify Skl skill?

Run npx skillmds@latest add x0ba/verify-skl in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

What does the Verify Skl skill do?

Drive the skl CLI and TUI the way a user does. Use when proving create, use, capture, unuse, list, status, or doctor against a real skl binary, or when a change might break library or project-projection behavior. It is listed under Coding & Dev Tools on SkillMD.

Is Verify Skl safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Verify Skl?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Verify Skl free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Verify Skl?

x0ba (@x0ba) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Verify Skl by x0ba — Agent Skill

Name: Verify Skl
Author: x0ba

Verify skl

skl is a personal agent-skill library. The write path is the skl CLI and the two-pane TUI (skl on a TTY, or skl tui / skl ui). Web (/skills, /devices) and HTTP /v1 are secondary: they list, preview, approve devices, and sync. They do not create, edit, activate, or capture. This skill drives the CLI first. Use the TUI recipe only when the map names a TUI entry.

Never drive the operator's real library (~/.local/share/skl) or this repo checkout as --project. The checkout already has a skills.toml for file-pr / frontend-design / greploop / html-communication.

Launch

skl is a short-lived binary. There is no server to keep alive for CLI proof. Build once, then start each drive in the isolate this helper creates.

.cursor/skills/verify-skl/helpers/verify-skl.sh launch

Launch prints an absolute source /tmp/skl-verify-<id>/session.env line. Copy that line into a later shell. doctor / cli / cleanup also read /tmp/skl-verify-latest (the last launch's session file), so a fresh shell can run doctor without $RUN_ID set. Concurrent runs should pass VERIFY_SKL_RUN_ID instead of relying on latest.

Ready when stdout contains ready version=skl 0.4.2, root /tmp/skl-verify-<id>, and the source line. The helper runs cargo build -p skl from the repo root and uses target/debug/skl. Debug api_base defaults to http://localhost:8787; launch overrides that with http://127.0.0.1:1 and writes [sync] auto = false so piggyback sync never leaves the isolate.

Isolate layout:

Path	Role
`$ROOT/data`	`SKL_DATA_DIR` — library at `$ROOT/data/skills`, index at `$ROOT/data/state.db`
`$ROOT/config`	`SKL_CONFIG_DIR` — `config.toml`
`$ROOT/home`	`HOME` so `skl doctor` / `skl init` scan empty agent roots, not `~/.claude`
`$ROOT/project`	disposable `--project` (and TUI cwd)
`/tmp/skl-verify-evidence/$RUN_ID`	proof artifacts; cleanup does not delete this

Launch also sets SKL_NO_TUI=1, SKL_NO_PROMPT=1, CI=1, VISUAL=true, EDITOR=true, and unsets SKL_TOKEN / SKL_TOKEN_FILE. true is the non-interactive editor: skl create still writes SKILL.md and returns 0 (editor.rs; create treats editor failure as a warning).

Teardown is helpers/verify-skl.sh cleanup (see Cleanup). Two isolates can run side by side if each has its own VERIFY_SKL_RUN_ID. Do not reuse a session you did not start.

Doctor

Run this first whenever anything looks off. It is read-only besides printing.

.cursor/skills/verify-skl/helpers/verify-skl.sh doctor

The helper requires skl 0.4.2, isolated SKL_DATA_DIR / SKL_CONFIG_DIR / HOME, then runs skl status and skl doctor from $VERIFY_SKL_PROJECT so the project line is the isolate, not this repo.

Worth driving only when doctor prints doctor ok and both reports show:

api_base http://127.0.0.1:1
state.db $ROOT/data/state.db
config $ROOT/config/config.toml
auto_sync off (status)
token no / token absent (local-only isolate)
== API, == Auth, == State, == Agent skill roots, == Linking
roles personal library is canonical; use materializes (copy) into .agents/skills
no /workspace/skills.toml on the project line

skl doctor always exits 0 after printing (apps/docs/docs/reference/doctor.mdx). Health unreachable on http://127.0.0.1:1 is expected. A live GET /v1/health is not required for local create/use/capture/unuse.

Refuse to continue if SKL_DATA_DIR is unset or is the default XDG path.

Drive

Prefer the helper. It already exports the isolate and sets cwd to $VERIFY_SKL_PROJECT.

.cursor/skills/verify-skl/helpers/verify-skl.sh cli -- create verify-greeter
.cursor/skills/verify-skl/helpers/verify-skl.sh cli -- use verify-greeter --project "$VERIFY_SKL_PROJECT"
.cursor/skills/verify-skl/helpers/verify-skl.sh cli -- use --project "$VERIFY_SKL_PROJECT"
.cursor/skills/verify-skl/helpers/verify-skl.sh cli -- list
.cursor/skills/verify-skl/helpers/verify-skl.sh cli -- doctor

Stable handles (do not use coordinates or tab order):

Surface	Handle
CLI create	stderr `created <name> (<data>/skills/<name>)`
CLI use	stderr `using <name> (agents …)` then `copy agents <project>/.agents/skills/<name>` then `updated <project>/skills.toml`
CLI use list	stdout header `name` / `source` / `mode`; rows use source `library` and mode `copy`
CLI unuse	stderr `unused <name>` then `removed agents …` or `absent`
CLI capture	stderr `captured <name> (<data>/skills/<name>)`
CLI list	stdout columns `name`, `tree_hash`, `source`, `path`, `remote`; library source is `agents`
CLI list empty	stderr `no local skill index; run \`skl init` first`or stdout`(no local skills)`
TUI chrome	header `library: N skill(s) · last sync: never · project: …`; list title `skills`; empty preview `No local skills.`
TUI create	overlay title `new skill`; hint `Enter opens $EDITOR with name + description frontmatter.`
TUI keys	`n` create, `/` search, `u` use, `U` unuse, `e` edit, `d` delete, `s` sync, `r` refresh, `?` help, `q`/`Esc` quit

TUI (only when a feature file names it). Needs a real TTY. Do not set SKL_NO_TUI. TERM=dumb degrades (tui/launch.rs).

.cursor/skills/verify-skl/helpers/verify-skl.sh tui-start
tmux -f /exec-daemon/tmux.portal.conf send-keys -t "$VERIFY_SKL_TMUX:0.0" 'n'   # or the key the map names
.cursor/skills/verify-skl/helpers/verify-skl.sh tui-capture
.cursor/skills/verify-skl/helpers/verify-skl.sh tui-stop

If /exec-daemon/tmux.portal.conf is missing, the helper calls plain tmux. Capture the pane after the UI settles; do not assert on a fixed sleep alone.

Read features/README.md and the matching feature file before driving. A proof that uses one convenient entry is incomplete when that file lists others.

Evidence

Write proof under /tmp/skl-verify-evidence/$RUN_ID/<feature-id>/ via:

.cursor/skills/verify-skl/helpers/verify-skl.sh evidence use-skill

That directory gets cli-last.txt, cli-history.txt (every cli -- in the run), doctor.txt, status.txt, dest trees, and skills.toml. Cleanup never deletes it. Copy elsewhere if a reviewer needs a durable attachment; do not commit run artifacts into .cursor/skills/verify-skl/.

Proof standards:

Exercise the real user command (skl create, skl use, …) or the TUI key that maps to it. Do not insert rows into state.db or write .agents/skills by hand and call that a pass.
Capture the action and the resulting state: command + exit code + stderr/stdout, then a second read (skl list, skl use with no names, cat of SKILL.md / skills.toml, find of dests).
Mutation proof includes on-disk side effects: library tree, dest file type (directory vs symlink), and skills.toml name / source / mode.
source = "library" in skills.toml is the portable token. Library index rows use source agents. Do not "fix" that (library.rs).
Default use is a real directory (mode = "copy"). --link is a symlink (mode = "symlink"). Capture without --keep-copy rewrites the project dest to a symlink.
Local-only isolate: skip skl sync, skl login, and skl delete (delete requires a device token and DELETE /v1/skills/:name). Report those paths as skipped with the unmet precondition. Do not claim them verified via create/use.
Auto-sync is disabled in the isolate. If a recipe turns it on, observe network / last_sync rather than trusting the name.

Cleanup

.cursor/skills/verify-skl/helpers/verify-skl.sh cleanup

Stops the tmux session this run started (by session name / pane pid, never pkill -f skl), then deletes only /tmp/skl-verify-$RUN_ID for that run. It refuses /tmp/skl-verify-evidence and any other isolate. Evidence stays at /tmp/skl-verify-evidence/$RUN_ID.

After a failed iteration, run the same cleanup before the next launch so the next tui-start does not collide.

Helpers

helpers/verify-skl.sh is executable. Every invocation in this file is literal.

Command	What it does
`launch`	`cargo build -p skl`, isolate dirs, session.env, `/tmp/skl-verify-latest`, version gate
`doctor`	isolate + `skl status` + `skl doctor` from the isolate project
`cli -- <args>`	`skl <args>` with isolate env, cwd = isolate project
`plant <name>`	write an unmanaged `.agents/skills/<name>/SKILL.md` in the isolate project
`tui-start` / `tui-capture` / `tui-stop`	one tmux session named `skl-verify-$RUN_ID`
`evidence <id>`	copy logs, trees, and skill files into the evidence dir
`cleanup`	tear down this run's isolate and tmux; keep evidence

Reuse a run in another shell by copying the printed source /tmp/skl-verify-<id>/session.env line, or run helper commands as-is (they follow /tmp/skl-verify-latest). VERIFY_SKL_RUN_ID=... pins a concurrent run. A later launch with a new VERIFY_SKL_RUN_ID ignores leftover VERIFY_SKL_ROOT / VERIFY_SKL_EVIDENCE from that source.

Maintenance

When commands, dest layout, or TUI keys change, update this skill and features/ with /maintain-verification-skill.

Verify Skl

File contents

Verify skl

Launch

Doctor

Drive

Evidence

Cleanup

Helpers

Maintenance

Frequently asked questions

Verify Skl

File contents

Verify skl

Launch

Doctor

Drive

Evidence

Cleanup

Helpers

Maintenance

Related skills

Frequently asked questions

How do I install the Verify Skl skill?

What does the Verify Skl skill do?

Is Verify Skl safe to use?

Which AI agents work with Verify Skl?

Is Verify Skl free to use?

Who published Verify Skl?