skillforgeSAFE
Grounded training-data pipeline and evaluation tools for agent-skill selection and abstention. CPU pipeline implemented; model training pending.
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
SkillForge
Turn the open agent-skill ecosystem into supervised fine-tuning data — and measure whether a small model can actually use it.
Every task, plan and check is mined from the skill's own text. Nothing is invented — a dataset you have to caveat is a dataset nobody can build on.
📐 Design · 🧪 Benchmark · ⚖️ Licensing · 📚 Related work
[](https://github.com/mkd-hika/skillforge/actions/workflows/ci.yml)
Reproduce the fixture report
git clone https://github.com/mkd-hika/skillforge.git cd skillforge python examples/quickstart.py
View the recorded output and its limitations. No installation or GPU is required.
The gap
Agent skills — a SKILL.md with a description that decides when it fires and a body disclosed only after it is chosen — went from a convention to an ecosystem in 2026. Thousands are published. Three things are still missing:
- Skill use is never a training target. Skills are prompt-engineered around
models, not traine
66ddd0a57fe4OBSERVED · 2026-10-13Install
Commands as the repository documents them. They are shown, not run.
git clone https://github.com/mkd-hika/skillforge.git
git clone https://github.com/mkd-hika/skillforge.git
What it tells the agent
The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.
--- name: mystery-skill license: Apache-2.0 --- # Mystery Has a name but no description. Inventing one would fabricate the trigger surface.
Trust audit
SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | WARN |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- none-observed
- Network
- none-observed
- Shell
- none-observed
- Dependencies
- pinned
- Secrets in source
- none-found
Findings (2)
normalized = text.replace("\r\n", "\n").lstrip("")meta, _ = parse("---\r\nname: x\r\ndescription: y\r\n---\r\n\r\nbody\r\n")Gates applied: no_behavioural_pass.
66ddd0a57fe4full audit observations/trust-audit/skill/mkd-hika__skillforge.json · Report an issue / request a re-scanAudit history
Every audit this skill has had.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-13 | 66ddd0a57fe4 | SAFE | B | 89 | first audit |
Questions
What does the skillforge skill do?
Grounded training-data pipeline and evaluation tools for agent-skill selection and abstention. CPU pipeline implemented; model training pending.
Is skillforge safe to install?
The audit found nothing in the source that contradicts what it says it does, and graded it B (89/100). Grade A is held back for packages that have also passed a sandboxed behavioural run, which is why a clean skill reads B.
What can skillforge access on my machine?
The audit observed no filesystem, network or shell use at all in its source.
How current is this page?
The grade is for one exact copy of the source (66ddd0a57fe4), read on 2026-10-13. The repository is watched, and a new audit runs when it changes — this is the first audit.