OpenAI Skills
Build a skill acceptance lab: triggers, permissions and evidence before migration
A proposed learning project for evaluating skill selection and outputs, with revision manifests, negative prompts, source counterexamples and a transparent migration checklist.
What you will learn
- Turn the lessons into a small lab
- Make migration observable and reversible
- Publish evidence, not a score without a denominator
Before you start
- Basic Git, Python and command-line concepts
- An explicit boundary for permitted repository inspection and changes
Explain the chapter’s source behavior and apply its acceptance checklist without confusing a catalog with its host.
Key takeaways
- Separate helper tests, simulated decisions and real host runs.
- Migrate against a preserved output and authority contract.
- Publish denominators and failure categories with any evaluation score.
Turn the lessons into a small lab
A useful follow-up project is a skill acceptance lab rather than another large catalog. Each case should record a fixed skill revision, host build, prompt, permitted tools, expected output and stop condition. Keep an explicit distinction between source-level fixtures, simulated host decisions and real agent runs so a green unit test cannot be mistaken for a successful production workflow.
Start with the counterexamples from this series: a slash-containing ref in a URL, an archive member outside the destination, empty metadata values and a log ending in “0 failures.” These cases exercise parsing and interpretation boundaries. Add workflow cases that require refusal to edit, missing-credential reporting and external-provider handoff without inventing results for tests that have not run.
Make migration observable and reversible
For each existing installation, map the old workflow folder to the intended current distribution, retaining its license and dependencies. Record what changes, what remains local and which prompts depend on the old behavior. A migration is accepted only when the new package meets the same output and authority contract, not when the files merely appear in a different directory.
The proposed dashboard can display a two-dimensional decision map: trigger expected, trigger observed, tool attempted, evidence produced and stop condition met. This sequence is clearer as an accessible SVG or table than as a decorative rotating object. Three-dimensional interaction would need a genuine spatial question to justify its rendering cost, keyboard work and static fallback.
Publish evidence, not a score without a denominator
Separate false triggers from missed triggers, retrieval failures from reasoning failures, and approved actions from attempted overreach. Report the number of cases and the actual acceptance rubric beside any score. Human review remains necessary for whether the cited evidence supports the diagnosis; substring detection cannot establish that by itself.
This lab is an editorial proposal, not an official OpenAI roadmap or a shipped interactive feature. The article diagrams are static explanatory SVGs. The completed work here consists of authored guidance and isolated source probes, while full host runs and migration measurements remain unperformed. That boundary makes the next experiment reproducible without pretending the future project already exists.
Implementation steps
- 1
Create a revision manifest and a small positive/negative case set.
- 2
Run deterministic counterexamples independently of the host.
- 3
Compare old and new workflows under the same permissions.
- 4
Promote only after evidence and rollback checks pass.
Copy-ready example
{"proposal":"skill-acceptance-lab","interactiveDemoShipped":false,"realHostRuns":0,"cases":["slash-ref","zip-traversal","empty-metadata","zero-failures","stop-before-edit"],"migrationAccepted":false}Frequently asked questions
Is this lab already implemented on the page?
No. The page contains original static diagrams and a proposed experiment design, not a running evaluation platform.
Is this OpenAI’s roadmap?
No. It is an independent learning-project proposal grounded in the inspected repository and its migration context.
Sources
- README.mdSource checked 2026-09-08
- skills/.system/skill-installer/SKILL.mdSource checked 2026-09-08
- skills/.system/skill-installer/LICENSE.txtSource checked 2026-09-08
- skills/.system/skill-installer/scripts/install-skill-from-github.pySource checked 2026-09-08
- skills/.system/skill-installer/scripts/github_utils.pySource checked 2026-09-08
- skills/.system/skill-installer/scripts/list-skills.pySource checked 2026-09-08
- skills/.system/skill-creator/scripts/quick_validate.pySource checked 2026-09-08
- skills/.system/skill-creator/references/openai_yaml.mdSource checked 2026-09-08
- skills/.curated/gh-fix-ci/SKILL.mdSource checked 2026-09-08
- skills/.curated/gh-fix-ci/scripts/inspect_pr_checks.pySource checked 2026-09-08
- skills/.curated/gh-fix-ci/agents/openai.yamlSource checked 2026-09-08
- skills/.system/plugin-creator/SKILL.mdSource checked 2026-09-08
- Current official skills guideSource checked 2026-09-08