skill-auto-improver

A repair workflow for an existing SKILL.md file, the instruction file that defines how a coding-agent skill works. It brings older or manually written skills toward the skill-creator standard and tests whether they meet required quality checks.

In plain words
What is it for?
Retrofitting an existing skill, running quick validation and frontmatter checks, evaluating its instructions, and reporting blockers. It also performs a separate advisory review of how predictably the skill guides decisions.
Why use it?
It helps find and fix structural or quality problems in legacy skills. A skill must pass both the validation checks and the supplementary evaluation threshold before the workflow considers it finished.

Skill for Claude CodeCodex

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/luongnv89/asm/skill-auto-improver
Any agent
npx skills add luongnv89/asm --skill skill-auto-improver
Clone the repo
git clone --depth 1 https://github.com/luongnv89/asm

Made for: Claude Code, Codex.

Per session 58 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 5,179 The whole file, excluding the scripts and references it only reads on demand.
Security scan B 1 finding. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00058 $0.05179
Opus 5 $0.00029 $0.02589
Sonnet 5 $0.00012 $0.01036
Haiku 4.5 $0.00006 $0.00518

Measured yesterday against content hash 49de3172c094, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade B, and why

skill-auto-improver scanned grade B with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Recursive force deletemediumDestructive command

rm -rf with a variable or a broad path is one typo away from removing the wrong tree.

- **Destructive action**: never `rm -rf` the skill directory. `asm eval --fix` creates `SKILL.md.bak` — leave it until the user explicitly cleans up.

Downgraded: this mod is about security review, or the phrase is quoted, so it is likely naming the pattern rather than instructing it.

skills/skill-auto-improver/SKILL.md · 303 lines

How it starts

The opening of the file, as written. The whole thing — 303 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Skill Auto-Improver

You run an eval-driven loop that retrofits an existing SKILL.md to the current skill-creator standard. It is the remediation tool for skills that did not go through skill-creator — external, legacy, manually-authored, or drifted. Fresh skill-creator output is publish-ready by construction and should not normally need it.

The target clears two hard gates, then gets one advisory audit:

  1. Gate 1 — skill-creator standard (must-pass floor)quick_validate clean, Frontmatter Audit passes, within the size caps.
  2. Gate 2 — asm-eval floor (supplementary)overallScore > 85 AND every category >= 8.
  3. Advisory — predictability audit (Phase 2b) — judgment-based findings against skill-creator's rubric, reported separately, never blocking.

A skill that scores 92 but fails quick_validate.py is not done; one that passes it but scores 70 is not done either. Both gates must clear, or the loop reports a blocker — open predictability findings alone never make one.

Two modes

Pick one before Phase 0 — they do not share a workflow.

  • Mode 1 — retrofit (default). Bring the target to the skill-creator standard via the Phase 0–7 loop below. Every "improve", "fix", "level up", or "bring up to standard" request is Mode 1.
  • Mode 2 — delegation conversion (opt-in). Restructure the target's steps onto per-step context delegation: each heavy step names the slice of its own references/ tree its worker needs and hands it over as that worker's Input. Runs outside the Phase 6 loop, on a target that already clears Gate 1, and only once the user confirms. A Phase 2b delegability finding routes here but never starts a conversion by itself. Procedure: references/delegation-conversion.md.

Dependency Preflight (mandatory)

This skill invokes skill-creator: it runs that skill's quick_validate.py (required — the Gate 1 validator) and reads its predictability-rubric.md (fail-soft — a local copy may predate the rubric, and a missing one only degrades Phase 2b to a warning). Resolve both before the repo sync below, the first step that changes anything:

Read the full file on GitHub · 303 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 303 lines · 58 tokens per session scan B 49de3172c094

Subscribe to this mod's changes

skill-auto-improver is a skill published in the GitHub repository luongnv89/asm (902 stars, last pushed 2d ago), licensed MIT. It adds 58 tokens to every session and 5,179 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it B with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Related

Other skills, from other repositories

sentry-setup-ai-monitoring

Setup Sentry AI Agent Monitoring in any project. Use this when asked to add AI monitoring, track LLM calls, monitor AI agents, or instrument OpenAI/Anthropic/Vercel AI/LangChain/Google GenAI. Automatically detects installed AI SDKs and configures the appropriate Sentry integration.

troykelly/claude-skills · 70 tokens

Skill Development

This skill should be used when the user wants to "create a skill", "add a skill to plugin", "write a new skill", "improve skill description", "organize skill content", or needs guidance on skill structure, progressive disclosure, or skill development best practices for Claude Code plugins.

troykelly/claude-skills · 63 tokens

Hook Development

This skill should be used when the user asks to "create a hook", "add a PreToolUse/PostToolUse/Stop hook", "validate tool use", "implement prompt-based hooks", "use ${CLAUDEPLUGINROOT}", "set up event-driven automation", "block dangerous commands", or mentions hook events (PreToolUse, PostToolUse, Stop, SubagentStop…

troykelly/claude-skills · 117 tokens

sentry-setup-metrics

Setup Sentry Metrics in any project. Use this when asked to add Sentry metrics, track custom metrics, setup counters/gauges/distributions, or instrument application performance metrics. Supports JavaScript, TypeScript, Python, React, Next.js, and Node.js.

troykelly/claude-skills · 61 tokens

sentry-setup-tracing

Setup Sentry Tracing (Performance Monitoring) in any project. Use this when asked to add performance monitoring, enable tracing, track transactions/spans, or instrument application performance. Supports JavaScript, TypeScript, Python, Ruby, React, Next.js, and Node.js.

troykelly/claude-skills · 62 tokens

database-architecture

MANDATORY when designing schemas, writing migrations, creating indexes, or making architectural database decisions - enforces PostgreSQL 18 best practices including AIO, UUIDv7, temporal constraints, and modern indexing strategies.

troykelly/claude-skills · 45 tokens