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 agentmods add skills/yuanruichen/metforge/metforge-datanpx skills add yuanruichen/MetForge --skill metforge-datagit clone --depth 1 https://github.com/yuanruichen/MetForgeWhat 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 | $0.00119 | $0.01020 |
| Opus 5 | $0.00060 | $0.00510 |
| Sonnet 5 | $0.00024 | $0.00204 |
| Haiku 4.5 | $0.00012 | $0.00102 |
Grade A, and why
metforge-data 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 2d 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 — 80 lines — stays where its author put it; the contents beside it link to each section on GitHub.
MetForge Data
Obtain the smallest scientifically sufficient dataset, preserve the source product, and leave an auditable path from request to local files.
Workflow
- Define the data contract before downloading:
- scientific variable and accepted source variable names
- product, processing level, experiment/member, or reanalysis stream
- spatial domain, grid, vertical coordinate, time range, and frequency
- units, calendar, accumulation convention, and expected volume
- required credentials, license, and citation
- Inspect existing files and project conventions. Reuse valid local data instead of downloading duplicates.
- Select an authoritative source. Read references/source-selection.md when the source or access route is uncertain. Verify current API syntax against official documentation; do not invent product identifiers.
- Plan a restartable request:
- subset at the server when scientifically safe
- batch large requests by year, month, variable, or level
- keep credentials outside scripts and version control
- keep raw files immutable
- Download one representative batch first. Do not launch the full request until the sample passes the checks below.
- Validate the sample, then scale out. Use lazy xarray/dask operations for large collections and reduce before materializing arrays.
- Record provenance. Follow references/dataset-contract.md for the manifest and handoff fields.
Execution mode and processing record
Before downloading or preprocessing, inspect the current machine, available memory/storage, data volume, network/access constraints, whether sbatch/Slurm is available, whether the session is already inside a Slurm job, and existing project conventions.
- Run metadata inspection, one-file validation, and small downloads directly when they fit safely in the current session.
- Use the site's scheduler for expensive multi-file preprocessing or long transformations. Use a transfer/data node rather than a compute node when site policy or network access requires it.
- Do not guess Slurm account, partition, walltime, modules, or storage paths. Infer them from existing project scripts/configuration or request the missing value.
- For submitted work, capture the job script, job ID, resources, environment activation, stdout/stderr paths, and terminal status.
What ships with it
3 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 2d ago First seen · 80 lines · 119 tokens per session scan A a3fe3f9231b2
metforge-data is a skill published in the GitHub repository yuanruichen/MetForge (4 stars, last pushed 20d ago), licensed MIT. It adds 119 tokens to every session and 1,020 once invoked, about $0.0006 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
spot-ai
SLASH COMMAND — type /spot-ai to audit a document, paragraph, or sentence for AI-isms — gray-list style tells (lane 1) and substance weaknesses like hollow claims or fabricated specifics (lane 2). Severity + confidence per finding, fix suggested or missing substance named. Flags only; never rewrites.
repo-init
SLASH COMMAND — type /repo-init to scaffold a new repository (or retrofit an existing one) with a standard project structure. Starts with a short intake grill (--no-grill skips). Research mode by default; --package for a distributable src-layout library. Never overwrites existing files; --dry-run previews.
ai-review
SLASH COMMAND — type /ai-review for a comprehensive senior-engineer review of a project or repository. Orchestrates a parallel fan-out across correctness, staleness, over-baking, and claim rigor by DELEGATING to the tools that own each lane, then adds the three lanes nothing else covers — gap/opportunity hunting…
figure-review
Audit a scientific figure for publication-readiness: colormaps, uncertainty, axis labels, caption completeness, and claim support. Panel-aware — judges one panel of a composite against what its shared caption and sibling panels supply, and raises a likely-deliberate choice as by-design? instead of a defect. Use this…
unstale
Detect and repair staleness residue in Python library code and notebooks — dead imports, dead code, resolved TODOs, stale comments/docstrings, and HANDOFF blockers. Use this whenever you want to clean up after a refactor, fix a comment that no longer matches the code, remove unused imports, check for leftover TODOs…
pathfinder
SLASH COMMAND — type /pathfinder to get a navigable map of every skill and subagent and when to reach for each. Resolves the reviewer-2-vs-panel decision. Use when unsure which skill or agent to invoke, or when orienting a new session.