Why this collection
Many Skills work well only in the environment where they were written. They silently assume a tool name, a local path, network access, write permission, or a powerful model that can fill in missing steps.
Portable Agent Skills takes a narrower approach: keep the Skill core host-neutral, make decisions explicit, load details progressively, and require evidence before declaring success. The result is easier to reuse across Codex, Hermes Agent, and other hosts that support the SKILL.md format—including setups that use DeepSeek or other models that benefit from clearer execution structure.
The three Skills
| Skill | What it turns into a repeatable workflow | Built-in guardrail |
|---|---|---|
| 🔎 topic-research | A vague research question → a traceable decision brief | Separates facts, source claims, inference, and unknowns |
| 🚀 github-project-deployer | A source repository → a verified local or hosted result | Read-only inspection and an explicit execution gate before code runs |
| 🛡️ skill-safety-checker | An unfamiliar Skill → an enable / isolate / fix / reject decision | Static findings remain candidates until context confirms them |
Prompt examples
Use topic-research to compare three approaches to this market and recommend one.
Use github-project-deployer to inspect this repository, show the execution plan,
then run it in the smallest reversible environment available.
Use skill-safety-checker to audit this downloaded Skill before I enable it.
Quick start
Open Agent Skills installer
Install one Skill:
npx skills add ch1109/portable-agent-skills --skill topic-research
npx skills add ch1109/portable-agent-skills --skill github-project-deployer
npx skills add ch1109/portable-agent-skills --skill skill-safety-checker
Or select interactively:
npx skills add ch1109/portable-agent-skills
Hermes Agent
Add the repository as a tap:
hermes skills tap add ch1109/portable-agent-skills
Or install a Skill directly:
hermes skills install ch1109/portable-agent-skills/skills/topic-research
hermes skills install ch1109/portable-agent-skills/skills/github-project-deployer
hermes skills install ch1109/portable-agent-skills/skills/skill-safety-checker
Codex and other compatible hosts
This repository includes a Codex plugin manifest and a standard skills/ directory. Use your host's plugin or Skill installation flow, or copy the selected directory intact into a Skill location recognized by that host.
Keep each Skill directory intact: its references/ and scripts/ folders are part of the Skill.
What makes the Skills portable
- Capability discovery first. The agent checks what the current host can actually do instead of assuming tools, network, storage, or permission.
- Host-neutral core. Platform-specific installation belongs in this README and adapters, not in the reusable Skill instructions.
- Progressive disclosure. The main instruction stays compact; detailed checklists and templates live in references loaded only when useful.
- Explicit decision gates. High-impact actions, weak evidence, and incomplete safety coverage cannot be silently treated as success.
- Evidence-based completion. “Done”, “deployed”, and “safe” require visible verification appropriate to the claim.
Read Why portable Skills for the design rationale and Repository benchmark for the public projects that informed this presentation.
Compatibility
The Skill core uses the minimal name and description frontmatter plus ordinary Markdown, references, and an optional read-only Python scanner.
| Environment | Distribution path | Notes |
|---|---|---|
| Hermes Agent | GitHub tap or direct repository install | Repository follows the documented skills/<name>/SKILL.md tap layout |
| Codex | Plugin manifest, installer, or manual Skill install | Host-specific metadata stays outside the Skill core |
| Other Agent Skills hosts | Installer or manual copy | Expected to work when the host supports standard SKILL.md discovery and bundled files |
Compatibility means the instructions can be discovered and loaded. Actual capabilities still depend on the host, enabled tools, permissions, and model.
Validation
Every change is checked for:
- valid frontmatter and directory naming;
- broken relative links and missing bundled files;
- host names, personal paths, and fixed permission assumptions in the Skill core;
- scanner readability and complete static-scan coverage;
- accidental secrets and common sensitive-file patterns.
Run the local checks with:
python3 -m unittest -v tests/test_repository.py
The safety scanner produces candidates, not automatic verdicts:
python3 skills/skill-safety-checker/scripts/scan_safety.py \
skills/topic-research --pretty
Repository layout
portable-agent-skills/
├── skills/
│ ├── topic-research/
│ ├── github-project-deployer/
│ └── skill-safety-checker/
├── docs/
├── tests/
├── assets/
└── .codex-plugin/
Safety note
No Skill can make an unsafe host safe. Review third-party content, use the least privilege available, and keep human approval for actions that publish, delete, expose data, spend money, or affect production. A static PASS is not a guarantee that unseen or runtime behavior is safe.
Contributing
Small, evidence-backed improvements are welcome. Please read CONTRIBUTING.md before opening a pull request. Security-sensitive reports should follow SECURITY.md.
License
MIT © 2026 ch1109
No comments yet
Be the first to share your take.