Agent skill

Skill Install Improved

by luongnv89 in luongnv89/asm

Install an improved variant of one named skill: resolve it by local path, repo, or name, run skill-creator's retrofit on a throwaway copy, then install the improved result.

MITAuto-check: notesAgent Workflows

Install Skill Install Improved

skills CLI
$ npx skills add luongnv89/asm --skill skill-install-improved -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install luongnv89/asm skill-install-improved --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/luongnv89/asm.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/skill-install-improved .claude/skills/skill-install-improved && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
skill-install-improved
GitHub stars
953
Token cost
~3.3k tokens
SKILL.md length
1,703 words
Files
6 (incl. references)
Skills in repo
7
Repo updated
First seen
Licence
MIT

At a glance

Install an improved variant of one named skill: resolve it by local path, repo, or name, run skill-creator's retrofit on a throwaway copy, then install the improved result.

  • Works in 5 steps: Resolve the target to one local directory → Delegate to skill-creator's retrofit loop → Harvest artifacts, then clean up → …
  • Improving in place
  • SKILL.md covers When to Use, Prerequisites, Dependency Preflight (mandatory) and Inputs, plus 5 more sections
  • Calls git and gh; reaches github.com

What it does

Skill Install Improved is an agent skill from luongnv89/asm. Install an improved variant of one named skill: resolve it by local path, repo, or name, run skill-creator's retrofit on a throwaway copy, then install the improved result. Don't use for improving in place, upstream PRs, or plain asm install.

Its SKILL.md is about 3.3k tokens, which your agent loads only when the skill is triggered. The skill folder holds 8 other files, including reference files (for example `docs/README.md`, `evals/evals.json` and `references/install-and-report.md`). Compatibility notes: Claude Code; requires asm and git on PATH

It sits in Agent Workflows, covering Skill authoring. The repository describes itself as: The universal skill manager for AI coding agents. The licence is MIT.

When your agent uses it

  • Improving in place
  • Plain asm install

Example prompts

  • “s retrofit on a throwaway copy, then install the improved result. Don”
  • “/skill-install-improved”

Requirements

  • Python 3
  • Node.js
  • Compatibility (from SKILL.md): Claude Code; requires `asm` and `git` on PATH
  • Pre-approved tools (allowed-tools): Bash, Read, Write, Edit, Grep, Glob

Workflow steps

5 steps, taken from the step headings in SKILL.md.

  1. Resolve the target to one local directory
  2. Delegate to skill-creator's retrofit loop
  3. Harvest artifacts, then clean up
  4. Install the improved directory
  5. Report, then clean up

What it can do on your machine

Read from SKILL.md and the folder at commit a01a183. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves these tools, so the agent can use them without asking each time:

    • Bash
    • Read
    • Write
    • Edit
    • Grep
    • Glob

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Shell commands in SKILL.md call:

    • git
    • gh

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    Hosts in commands or code, which the agent is likely to contact:

    • github.com

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

  • Compatibility

    Claude Code; requires `asm` and `git` on PATH

    From compatibility in the SKILL.md frontmatter.

Context cost

Skill Install Improved loads about 3.3k tokens when it runs, and up to ~7.3k if it reads all its reference files. Until then it costs about 66 tokens; SKILL.md has 1,703 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~66
When it runs · the whole SKILL.md, loaded when a task matches
~3.3k
With references · SKILL.md plus every file in references/, read only if the agent opens them
~7.3k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check: notes

The automated check noted patterns worth knowing about, such as sudo or a known installer.

  • NotePre-approves every shell command (allowed-tools: Bash)SKILL.md
    allowed-tools: Bash, Read, Write, Edit, Grep, Glob

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from luongnv89/asm at commit a01a183, republished under its MIT licence (© luongnv89). 1,703 words, ~3,285 tokens.

Download SKILL.mdSave it as .claude/skills/skill-install-improved/SKILL.md (or your agent's skills folder). This skill also uses 5 other files; get the full folder from GitHub.
name
skill-install-improved
description
Install an improved variant of one named skill: resolve it by local path, repo, or name, run skill-creator's retrofit on a throwaway copy, then install the improved result. Don't use for improving in place, upstream PRs, or plain asm install.
allowed-tools
Bash, Read, Write, Edit, Grep, Glob
compatibility
Claude Code; requires `asm` and `git` on PATH
license
MIT
effort
high
dependencies
skill-creator
metadata.version
1.3.0
metadata.author
luongnv89

Skill Install Improved

You install an improved variant of one skill instead of the skill as published: resolve the target to a local directory → run skill-creator's Subpath B1 retrofit loop on it → install the improved result → report what was improved and from what. The improvement is a step inside the install, not a separate errand. One skill per run.

When to Use

  • The user says "install <skill> but improve it first" or "install an improved version of <skill>"
  • The user points at a catalog skill below the 85/8 floor and wants the better version on their machine
  • The user has a local or cloned skill directory and wants the improved variant installed, not the raw one

Do not trigger for: improving without installing (skill-creator), contributing upstream (skill-upstream-pr), authoring from scratch (skill-creator), plain installs (asm install <source>), or bulk-improving many skills at once.

Prerequisites

Verify each before resolving anything. If one fails, stop and tell the user which one.

  • asm and git on PATH, and Python 3 (skill-creator's validator needs it)
  • Network access to GitHub, for the repo and name forms
  • Write access to the install target for the chosen tool and scope (see asm list --json for existing install paths)
The user's own files are never modified (design decision)

Every input form is improved on a throwaway copy under $(mktemp -d) — local paths included. An install that rewrites the user's working copy is a side effect they did not ask for, and the retrofit loop's git pull --rebase would rebase their branch.

Repo sync is skipped for every form. A local-path copy has no origin. A repo or name clone does have origin, but it is a throwaway --depth 1 checkout that may sit on a detached --ref, and a pull could move it off the requested ref. The retrofit loop allows this skip for a caller's throwaway copy. Say so when you skip it.

Trade-off: the improvement lands only in the installed copy. To persist it, point the user at skill-creator (their own path) or skill-upstream-pr (the source repo).

Dependency Preflight (mandatory)

This skill invokes skill-creator, declared in frontmatter dependencies. Before Phase 0, verify asm is available:

bash
command -v asm >/dev/null || {
  echo "Missing installer: npm install -g agent-skill-manager" >&2
  exit 1
}
asm deps discover skill-install-improved --json

When execution reaches Phase 1, the main agent runs:

bash
asm deps acquire skill-creator --session <caller-session-id> --json

<caller-session-id> is a unique id the main agent picks once per run, e.g. skill-install-improved-<epoch-seconds>; reuse it for the release.

Set SKILL_CREATOR_DIR to the directory holding the returned skillMdPath, and use it immediately; do not wait for a provider catalog rescan. The retrofit loop is $SKILL_CREATOR_DIR/references/retrofit-loop.md; its validator is $SKILL_CREATOR_DIR/scripts/quick_validate.py. If the acquire fails, stop before Phase 1 and report the error — this skill cannot improve anything without it.

The main agent releases the lease from its own finally/shutdown handling — after Phase 4, or at any earlier stop:

bash
asm deps release --session <caller-session-id> --json

If Phase 0 stops before Phase 1, do not acquire.

Inputs

The user identifies one target skill, in any of four forms — a local path (skills/foo), a local SKILL.md file path, a repo (https://github.com/owner/repo, owner/repo), or a skill name (code-review). Phase 0 normalizes all of them to one local directory, $SKILL_PATH.

Also needed before Phase 3: the install scope (global or project) and the target tool (-p/--tool — claude, codex, agents, …). If the user did not state one, ask; never guess. asm install hard-fails without --tool in a non-interactive run, and -y does not cover it.

Optional: a --name <alt> intent.

Before Phase 0, record the directory the user invoked you from as CALLER_DIR (pwd). Project-scope paths resolve from the current directory, so Phase 3 must run there, not inside $WORK.

Workflow

Execute phases in order. Do not skip or reorder.

Phase 0 — Resolve the target to one local directory

Full contract in references/target-resolution.md. In short, with WORK="$(mktemp -d)":

  • Local path — cp -R into $WORK. A SKILL.md file path folds to its parent first; the retrofit loop takes a directory, not a file.
  • Repo — plain git clone into $WORK. Never gh repo fork — this skill performs no public GitHub action.
  • Skill name — asm search "<term>" --available --json, then copy the string after asm install from the chosen result's installCommand verbatim. Never hand-construct github:owner/repo:path. Clone what it names.

Set $SKILL_PATH to the directory holding the chosen SKILL.md. If the checkout holds several and no subpath was given, list them (find "$WORK" -maxdepth 5 -name SKILL.md -type f) and ask which one. Never guess.

Record for the report: the supplied identifier, the resolved install or clone URL, and the upstream commit SHA.

Phase 1 — Delegate to skill-creator's retrofit loop

This skill does not reimplement the improvement loop, and it reads that loop only now, so the loop's references stay out of the context budget until Phase 1. Acquire skill-creator (Dependency Preflight), then follow $SKILL_CREATOR_DIR/references/retrofit-loop.md (Subpath B1) with $SKILL_PATH as the target, through its Phase 7 report.

Two adaptations, because the target is a throwaway copy:

  • Skip its "Repo Sync Before Edits" step. Log — repo sync skipped (throwaway copy) and continue.
  • .asm-improver/ is written relative to the current working directory. Run the loop with cwd inside $SKILL_PATH, or Phase 2 has nothing to harvest.
  • If the loop asks how to handle a target with no frontmatter, answer "abort" (see Edge Cases). If it offers a Subpath B3 delegation conversion, decline: this run installs, it does not restructure.

If the baseline already clears both gates, the retrofit loop stops without editing — a valid outcome, see Edge Cases.

Phase 2 — Harvest artifacts, then clean up

Harvest before any cleanup. Read the files under $SKILL_PATH/.asm-improver/ that references/install-and-report.md lists, and extract its fields. An early-exit baseline writes no iter-N.json; then the baseline is also the final state.

Then confirm $SKILL_PATH is still inside $WORK, and remove the two artifacts that must not ship (asm install copies the source recursively):

bash
rm -f "$SKILL_PATH/SKILL.md.bak"
rm -rf "$SKILL_PATH/.asm-improver" "$SKILL_PATH/.git"

Never point either command at a user-supplied path.

Show full SKILL.md (750 more words)Show less
Phase 3 — Install the improved directory

The retrofit loop never renames, so the improved variant keeps the original frontmatter name. asm install never refuses on a collision: with -y it deletes and replaces the target directory without a prompt. So probe before invoking it:

Run this phase from CALLER_DIR:

bash
cd "$CALLER_DIR"
asm list --json    # is this skill's name or directory already installed for $TOOL / $SCOPE?

Read references/install-and-report.md → Collision policy and take the path it gives for the probe result. The rule that never changes: if the probe shows any match — frontmatter name or directory name — name the target path and what occupies it, and get explicit confirmation before running asm install. Use --name <alt> only when the user asks for it, after the duplicate-trigger warning.

Only after the probe is read and any collision is confirmed:

bash
asm install "$SKILL_PATH" -p "$TOOL" --scope "$SCOPE" --json -y

Install $SKILL_PATH, never the original source. Take the installed path from the command's --json output (.path); never assume ~/.claude/skills/.

Phase 4 — Report, then clean up

Fill in the report template in references/install-and-report.md. It covers the four items the user needs: the result, the evidence, what stays untested, and the next decision.

Remove $WORK only after the report is printed, then release the dependency lease.

Step Completion Reports (mandatory)

After each phase, print a compact status block: √ pass, × fail, — context.

◆ Phase N — [phase name] ([N of 5])
··································································
  [check 1]:         √ pass
  [check 2]:         × fail — [reason]
  Result:            PASS | FAIL | PARTIAL

Checks per phase:

  • Phase 0 — Form identified, Copied to temp, SKILL_PATH unambiguous, Provenance recorded
  • Phase 1 — skill-creator acquired, Baseline captured, Repo sync skipped, Gates cleared or stop reason known
  • Phase 2 — Metrics harvested, .bak removed, .asm-improver removed, Deletions confined to temp
  • Phase 3 — Collision probed before install, Overwrite confirmed (if any), Tool and scope supplied, Install succeeded
  • Phase 4 — Report printed, Temp dir removed, Lease released

Acceptance Criteria

The expected output of a successful run, each item checkable:

  • Exactly one target resolved; $SKILL_PATH is a directory with a SKILL.md, under $WORK; the user's files are unmodified
  • Name resolution copied installCommand verbatim; repo resolution used plain git clone — no fork, no push
  • The retrofit loop ran from the acquired skill-creator; baseline.json, iter-N.json, and report.md were read before cleanup
  • The installed skill contains no .asm-improver/ and no SKILL.md.bak
  • asm install targeted $SKILL_PATH with both -p/--tool and --scope, after a collision probe; any overwrite was confirmed first
  • The report states the installed path, provenance, and before → after numbers
  • $WORK was removed after the report, and the dependency lease was released

When reviewing a run's output (by eval or by a human), also check that a reader can:

  • Find the main result — the first line says an improved variant was installed, the original was installed unchanged, or the run stopped.
  • Separate facts from assumptions — measured scores are distinct from the untested runtime behavior of the improved skill.
  • Trace claims — before → after numbers map to baseline.json and the last iter-N.json; the install path maps to install --json .path.
  • See the next decision — the report names any action left (restart the agent, persist the change upstream) or states that none is needed.

An agent's own review cannot confirm human understanding. A reviewer who received no human feedback records understanding as unconfirmed.

Edge Cases

  • Baseline already passes both gates — the retrofit loop stops without editing, but it has already written .asm-improver/. Still run Phase 2's cleanup and Phase 3's probe. Install the original, unchanged, and report that no improvement was needed, naming the baseline score. Never imply a delta that did not happen.
  • Retrofit loop ends in BLOCKER — show the blocker list and ask whether to install the partial result or abort. Never install one silently.
  • skill-creator cannot be acquired — stop before Phase 1, remove $WORK, and report the asm deps acquire error.
  • Several SKILL.md files in a clone — list them and ask. Never batch, never guess.
  • asm search returns nothing, or several equally-plausible matches — show the candidates and ask; never install the first hit.
  • Target has no frontmatter — asm eval --fix cannot add it, and this skill does not scaffold one. Answer "abort" to the retrofit loop, report, and stop; do not install an unimprovable skill as if it were improved.
  • Collision with a same-named skill from another source — the install would replace someone else's skill with no prompt. Name the target directory and what the probe shows there (name, dirName, provider, scope), confirm, and offer --name <alt> — all before asm install runs.
  • Clone or disk failure mid-resolve — stop, remove $WORK, report. Never install a partial checkout.

References

  • references/target-resolution.md — the input forms normalized to one local $SKILL_PATH
  • references/install-and-report.md — harvested fields, install flags, collision policy, report template
  • $SKILL_CREATOR_DIR/references/retrofit-loop.md — the improvement loop this skill delegates to (acquired at run time)
  • asm install --help and asm search --help — flag references

© luongnv89, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 5 other files (references) in skills/skill-install-improved of luongnv89/asm.

  • SKILL.md
  • LICENSE.txt
  • docs/README.md
  • evals/evals.json
  • references/install-and-report.md
  • references/target-resolution.md

Open the folder on GitHubat commit a01a183

Compare with similar skills

Skill Install Improved next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Skill Install Improved compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Skill Install Improved this skillluongnv89/asm953—~3.3kAutomated safety check: NotesMIT
Skill CreatorAzure/azqr79589 repos~8.2kAutomated safety check: PassApache-2.0
Claude Code Skill Developer Guidediet103/claude-code-infrastructure-showcase10k11 repos~3.5kAutomated safety check: PassMIT
Darwin Skill Optimizeralchaincyf/darwin-skill6.2k1 repos~4.7kAutomated safety check: PassMIT
Claude Code Command Developmentanthropics/claude-plugins-official38k10 repos~4.8kAutomated safety check: PassApache-2.0
Claude Code Plugin Structureanthropics/claude-plugins-official38k10 repos~3.4kAutomated safety check: PassApache-2.0

Similar skills

  • Skill Creator

    Azure/azqr

    Official

    Create new skills, modify and improve existing skills, and measure skill performance.

    795 GitHub starsUsed in 89 repos~8.2k tokens
    Agent WorkflowsAuto-check passed
  • Claude Code Skill Developer Guide

    diet103/claude-code-infrastructure-showcase

    A guide to creating and managing Claude Code skills with auto-activation: skill-rules.json triggers, hooks, enforcement levels, YAML frontmatter and progressive disclosure.

    10k GitHub starsUsed in 11 repos~3.5k tokens
    Agent WorkflowsAuto-check passed
  • Darwin Skill Optimizer

    alchaincyf/darwin-skill

    Scores SKILL.md files on a nine-dimension rubric, then improves them in a keep-or-revert loop with independent judge agents, test prompts, git history and human checkpoints.

    6.2k GitHub starsUsed in 1 repo~4.7k tokens
    Agent WorkflowsAuto-check passed
  • Claude Code Command Development

    anthropics/claude-plugins-official

    Official

    Explains how to write Claude Code slash commands: Markdown files with YAML frontmatter, arguments, file references, bash context and interactive prompts.

    38k GitHub starsUsed in 10 repos~4.8k tokens
    Agent WorkflowsAuto-check passed
  • Claude Code Plugin Structure

    anthropics/claude-plugins-official

    Official

    Explains the directory layout, plugin.json manifest and component organization of a Claude Code plugin, including auto-discovery and portable paths.

    38k GitHub starsUsed in 10 repos~3.4k tokens
    Agent WorkflowsAuto-check passed
  • Skill Release Gate

    rohitg00/ai-engineering-from-scratch

    Evaluates an Agent Skill bundle before release for structure, trigger quality, artifact improvement, script correctness, safety, installed-tree integrity and host portability.

    66k GitHub stars~1k tokensUpdated yesterday
    Agent WorkflowsAuto-check passed

More from luongnv89/asm

  • Skill Shortener

    luongnv89/asm

    Refactor a too-long SKILL.md by progressive disclosure: measure token cost, classify every section KEEP/CUT/MOVE, shorten the body into references/ and scripts/, verify nothing was lost.

    953 GitHub stars~3.8k tokensUpdated 2 days ago
    Auto-check: notes
  • Skill Creator

    luongnv89/asm

    Create a skill or bring an existing one up to the same standard (validate + asm eval fix loop); run evals, tune triggering.

    953 GitHub stars~5.3k tokensUpdated 2 days ago
    Auto-check passed
  • Find Me Skills

    luongnv89/asm

    Find Agent Skills for a goal the user cannot name yet, then export an installable bundle.

    953 GitHub stars~2.8k tokensUpdated 2 days ago
    Auto-check passed
  • Refresh Index

    luongnv89/asm

    Sync every enabled repo in the curated skill index and open a confirmation-gated PR.

    953 GitHub stars~3.4k tokensUpdated 2 days ago
    Auto-check: notes
  • Skill Index Updater

    luongnv89/asm

    Add GitHub skill repos to the ASM index: clone, audit, eval, regenerate index, rebuild catalog, open PR.

    953 GitHub stars~4.3k tokensUpdated 2 days ago
    Auto-check: notes
  • Skill Upstream PR

    luongnv89/asm

    Improve an open-source GitHub skill and open a friendly suggestion PR upstream: fork, run skill-creator's retrofit, attach asm eval before/after metrics.

    953 GitHub stars~2.4k tokensUpdated 2 days ago
    Auto-check: notes

Categories

Questions about Skill Install Improved

What does Skill Install Improved do?

Install an improved variant of one named skill: resolve it by local path, repo, or name, run skill-creator's retrofit on a throwaway copy, then install the improved result. Skill Install Improved is an agent skill from luongnv89/asm. Install an improved variant of one named skill: resolve it by local path, repo, or name, run skill-creator's retrofit on a throwaway copy, then install the improved result.

When should I use Skill Install Improved?

Skill Install Improved fits situations like: improving in place; plain asm install.

How do I install Skill Install Improved in Claude Code?

Run `npx skills add luongnv89/asm --skill skill-install-improved -a claude-code`. Or copy the skill folder (skills/skill-install-improved in luongnv89/asm) into .claude/skills/skill-install-improved in your project. Claude Code loads it when a task matches its description.

How do I install Skill Install Improved in Codex?

Run `npx skills add luongnv89/asm --skill skill-install-improved -a codex`. Or copy the skill folder (skills/skill-install-improved in luongnv89/asm) into .agents/skills/skill-install-improved in your project. Codex loads it when a task matches its description.

Can I use Skill Install Improved in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add luongnv89/asm --skill skill-install-improved -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/skill-install-improved, .gemini/skills/skill-install-improved, .github/skills/skill-install-improved and .opencode/skills/skill-install-improved in your project.

What does Skill Install Improved need to run?

Going by SKILL.md and its folder, Skill Install Improved needs the command-line tools its instructions call (git and gh). Our summary lists: Python 3; Node.js. Its frontmatter pre-approves these tools: Bash, Read, Write, Edit, Grep, Glob. Compatibility (from SKILL.md): Claude Code; requires `asm` and `git` on PATH.

Does Skill Install Improved access the network?

SKILL.md names 1 domain. In commands or code: github.com; the agent is likely to contact it when it follows the instructions. This is read from the text; nothing was executed.

Is Skill Install Improved safe to install?

Our automated static check of SKILL.md found notes only (pre-approves every shell command (allowed-tools: bash)), nothing it rates as a warning. It is not a guarantee. Review the folder before installing.

What licence does Skill Install Improved use?

Skill Install Improved is published under the MIT licence (declared in SKILL.md). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Skill Install Improved use?

About 3.3k tokens (SKILL.md is roughly 13k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 4.1k tokens, read only when the agent opens those files.

What are the alternatives to Skill Install Improved?

Skills that share tags, products or a category with Skill Install Improved: Skill Creator (Azure/azqr, 795 stars), Claude Code Skill Developer Guide (diet103/claude-code-infrastructure-showcase, 10k stars), Darwin Skill Optimizer (alchaincyf/darwin-skill, 6.2k stars) and Claude Code Command Development (anthropics/claude-plugins-official, 38k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Skill Install Improved?

luongnv89 (a GitHub user) maintains it in luongnv89/asm, which has 953 GitHub stars. The repository holds 7 skills in this directory. The repository was last updated on October 6, 2026.

Source: luongnv89/asm on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.