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.
npx skills add augchan42/inkstone --skill gloss-drift-auditgit clone --depth 1 https://github.com/augchan42/inkstoneWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/augchan42/inkstone/gloss-drift-audit)<a href="https://agentmods.dev/skills/augchan42/inkstone/gloss-drift-audit"><img src="https://agentmods.dev/badge/skills/augchan42/inkstone/gloss-drift-audit/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/augchan42/inkstone/gloss-drift-audit"><img src="https://agentmods.dev/badge/skills/augchan42/inkstone/gloss-drift-audit.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00091 | $0.01615 |
| Opus 5 | $0.00046 | $0.00807 |
| Sonnet 5 | $0.00018 | $0.00323 |
| Haiku 4.5 | $0.00009 | $0.00161 |
Grade A, and why
gloss-drift-audit 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 9d ago.
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.
How it starts
The opening of the file, as written. The whole thing — 100 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Gloss Drift Audit
When a source text is corrected, everything derived from it inherits the old reading. This skill is the order of operations for finding what inherited it — and, just as important, for not re-deriving the 96% that is fine.
The chain
source ──► gloss ──► prompt ──► rendered asset
free free free PAID
Every arrow can change meaning. Cost is flat until the last one. That asymmetry dictates the whole order: settle meaning first, move prompts only where meaning actually moved, render once at the end.
Two failure classes — do not conflate them
| Movement staleness | Meaning error | |
|---|---|---|
| What happened | source moved under the gloss | gloss was wrong the day it was written |
| Detectable? | yes — stamp the source digest per axis | no — nothing in the text reveals it |
| Self-clearing? | yes, on restamp | no |
| Typical measure | sha(source) != stamp |
none exists by default |
The trap: a corpus with a green staleness report can be full of mistranslations. A real case — a verse glossed 祈父 (an office: Minister of War, from 《詩經·小雅·祈父》) as "his father" — was never stale for a moment. Its source had not moved in years. Every gate said PASS.
If your only instrument is a digest comparison, you are measuring bookkeeping and calling it correctness.
Order of operations
- Re-measure the baseline first. A baseline taken after a fix cannot show the fix worked. Expect the number to have moved since it was last written — derived edits ride along with unrelated passes.
- Exclude dead data. Rows superseded downstream (editorial rewrites, overrides) often keep a stale field that nothing renders. Comparing it manufactures findings no reader will ever see. In one audit, 9 of the 10 worst-scoring rows were dead data.
- Run the free comparison (below), knowing its coverage.
- Run the register (below). This is the instrument with reach.
- Correct glosses. Restamp that axis only.
- Re-judge prompts for exactly the corrected rows — never all of them. Prompt and gloss were authored together from the same reading, so correcting one convicts the other. But check: a generic prompt may be right either way and cost nothing.
- Render last, once.
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.
- 9d ago First seen · 100 lines · 91 tokens per session scan A 88d9557d3d9c
gloss-drift-audit is a skill published in the GitHub repository augchan42/inkstone (4 stars, last pushed 13d ago), licensed MIT. It adds 91 tokens to every session and 1,615 once invoked, about $0.0005 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.
Other skills, from other repositories
tiktok-shop-cross-border
Cross-border selling on TikTok Shop. Market selection, logistics setup, localization strategy, compliance requirements, and international expansion. Use when the user asks about TikTok Shop international selling, cross-border ecommerce, global expansion, or selling in multiple countries.
shopify-international
Shopify Markets — multi-currency, translation, duties/taxes, localized pricing, market-specific content.
localization-testing
AI-powered localization and internationalization testing skill for e-commerce sites. Designs multi-language QA frameworks, currency validation checks, shipping info verification, and regional compliance audits.
translation
Translate text between languages with cultural adaptation, terminology consistency, and support for multiple document types including technical docs, marketing copy, and UI strings. Use when the user requests translation or provides relevant inputs for this workflow.
global-toolbar-controls
Quick global settings — currency, language, region, units — belong in a persistent, low-profile location such as a header toolbar or footer. These controls are frequent but not primary, so they use small typography and stay out of the main content hierarchy. Use when designing global selectors, locale switchers, or…
fec-drawio-studio
An editable workflow for making technical diagrams in draw.io (also called diagrams.net), with the original .drawio file saved alongside exported images or documents. It covers architecture, database, UML, sequence, flow, machine-learning, and code-structure diagrams.