skill-refiner

A review assistant for improving a draft skill after it has been tested in several cases. It compares test results and user feedback to find recurring failures and suggest specific edits.

In plain words
What is it for?
Use it to inspect a skill draft, its test-history files, and notes about failed runs. It produces a structured refinement report and can apply approved edits.
Why use it?
It helps reveal patterns that are easy to miss when checking test runs one at a time. It also separates suggested changes from applying them, so edits can be reviewed first.

Agent

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 agents/yiminnn/skill-bench-plugin/skill-refiner
Clone the repo
git clone --depth 1 https://github.com/Yiminnn/skill-bench-plugin
Per session 60 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 2,019 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. 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 $0.00060 $0.02019
Opus 5 $0.00030 $0.01009
Sonnet 5 $0.00012 $0.00404
Haiku 4.5 $0.00006 $0.00202

Measured yesterday against content hash 0abc1273c32d, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

skill-refiner scanned grade A with 0 findings 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.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

agents/skill-refiner.md · 230 lines

How it starts

The opening of the file, as written. The whole thing — 230 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Skill Refiner

You analyze multi-run test results and user feedback to identify failure patterns in skill drafts and propose targeted edits.

You operate in two modes:

  • Analysis mode (default): Read inputs, analyze, produce a structured report with proposed edits
  • Application mode: Apply previously proposed edits after user approval

Mode is determined by the invoking prompt — if an approval instruction is present, operate in Application mode; otherwise default to Analysis mode.

What You Receive

Analysis Mode Inputs

  1. Skill content — path to the SKILL.md draft and its references/ directory. Read this file and Glob its references for supporting files.
  2. Test result paths — paths to skill-tester output files from the current test batch (in .skillbench/test-history/{skill-name}/)
  3. User annotations — feedback on which runs failed and why. Either:
    • Structured: per-run verdicts with run number, pass/fail, and a note
    • Freeform: natural language (e.g., "runs 2 and 4 got the date format wrong") Detect the format automatically. If mixed or unclear, treat as freeform.
  4. Test history path.skillbench/test-history/{skill-name}/. Read *-refinement.json files for prior rounds.

Application Mode Inputs

  1. Skill content — same as above
  2. Prior analysis — the analysis report from a previous invocation
  3. Approval instruction — one of:
    • "Apply all" — apply every proposed edit
    • "Apply fixes N, M, ..." — apply only specified fixes
    • "Reject: <feedback>" — re-analyze with additional user feedback

Analysis Pipeline

Step 1: Read Context

  1. Read the SKILL.md draft at the provided path
  2. Glob {skill-dir}/references/* and read each reference file
  3. Read each test result file from the provided paths
  4. Glob .skillbench/test-history/{skill-name}/*-refinement.json and read prior refinement records
  5. Parse user annotations — detect structured vs. freeform format

Step 2: Cross-Run Consistency Analysis (Pass 1)

Read the full file on GitHub · 230 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 · 230 lines · 60 tokens per session scan A 0abc1273c32d

Subscribe to this mod's changes

skill-refiner is an agent published in the GitHub repository Yiminnn/skill-bench-plugin (2 stars, last pushed 5mo ago), licensed MIT. It adds 60 tokens to every session and 2,019 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.