External Hermes Skills Lifecycle

External Hermes Skills Lifecycle

Overview

Use this umbrella skill when the task is about the lifecycle of a skill outside Hermes core:

• adopting a third-party skill into ~/.hermes/skills
• preserving linked support files and category layout
• making the installed skill discoverable and verifiable
• packaging a reusable skill or methodology repo for public release so other agents can adopt it correctly

A human maintainer would not keep separate micro-skills for "install" and "package" when they are two halves of the same lifecycle. This umbrella keeps both directions together:

• inbound adoption -> install into a local Hermes environment
• outbound release -> package a local skill/methodology repo for open-source reuse

When to Use

Use when:

• a user wants a third-party skill from GitHub or another repo installed into Hermes
• a skill repo must be copied into ~/.hermes/skills with correct category placement
• you need to verify that support files (references/, templates/, scripts/, assets/) came across intact
• a reusable skill/methodology repo is being prepared for public release
• README, AGENTS, runtime guidance, and deployment docs need clearer separation

Do not use when:

• the task is authoring one local skill from scratch with no import/release dimension
• the task is only a tiny wording tweak inside an already-installed local skill
• the repo is not intended to be reused outside the current workspace

Lifecycle Split

Inbound: adopt an external skill locally

Goal: install a third-party skill into ~/.hermes/skills without modifying Hermes core.

Outbound: package a reusable skill repo for others

Goal: ship a repo that another human or agent can adopt systematically without guessing structure, rollout order, or memory boundaries.

Part A — Installing External Skills Into Hermes

1. Inspect the live local skill layout first

Before copying anything:

• inspect ~/.hermes/skills
• determine category conventions
• confirm whether the destination should be category/skill-name/

Do not guess taxonomy from the upstream repo alone.

2. Fetch or update the upstream source

• clone or update the source repo in the requested workspace
• record the exact commit or revision used

3. Inspect the upstream skill directory before copying

Read:

• SKILL.md
• linked files under references/, scripts/, templates/, assets/

Decide install category from:

• live local taxonomy
• the skill's actual purpose

4. Install only the requested skill directory

• copy just the requested skill directory
• preserve upstream files exactly unless a minimal compatibility fix is required
• do not drag in unrelated siblings from the repo

5. Verify exact-copy integrity and discovery

Verification should include:

• destination path under ~/.hermes/skills
• file existence and integrity
• skills_list visibility
• skill_view(name) load success
• support-file access when relevant

6. Separate discoverability from runtime readiness

A skill can be installed and discoverable even if optional scripts or dependencies are not yet usable. Report missing runtime dependencies separately instead of calling the install a failure.

Part B — Packaging Skills or Methodology Repos for Open Source

1. De-privateize first

Before publishing:

• remove machine-specific paths, usernames, vault names, and private defaults
• move environment-specific assumptions into documented config or integration docs
• make clear what the package is not if it could be mistaken for a plugin, product, or daemon

2. Separate package surfaces by role

Common surfaces include:

• README -> human-facing overview and adoption hook
• multi-language README entry links when more than one language is shipped
• optional agent loader docs such as AGENTS.md
• optional compressed runtime guidance such as SOUL.md
• memory-boundary guidance when relevant
• reusable assets under skills/, templates/, scripts/
• deployment or implementation docs that explain rollout order

Not every repo needs every surface, but every shipped surface must have a clear purpose.

When packaging a unified public repository from an existing local skill library:

• organize skills into a small number of class-level bundles rather than dumping a flat directory tree
• copy the public package from the source skill directories, not the other way around; keep the local/private originals untouched
• preserve support files with each skill (references/, templates/, scripts/, assets/)
• normalize private examples in the copied public version: machine paths, wrapper binaries, private vault names, host-specific policy wording
• add repo-level surfaces that explain the collection, not just the individual skills; a strong default is README.md, AGENTS.md, a bundle map, portability notes, and a source map

For a compact packaging checklist and recommended bundle/doc layout, see references/public-skill-bundle-packaging.md.

After the first public push succeeds, do a public-repo polish pass before calling the package truly handoff-ready:

• add community-health surfaces such as CONTRIBUTING.md, .github/ISSUE_TEMPLATE/*, and .github/pull_request_template.md
• add governance surfaces such as SECURITY.md and CODE_OF_CONDUCT.md when the repo is meant for outside collaboration
• add a CHANGELOG.md and cut an explicit initial release tag instead of leaving the repo in an unversioned “just pushed” state
• add bundle-level README.md entrypoints when the repo groups many skills into thematic directories
• add a social-preview asset to the repo itself (for example assets/social-preview.png) plus a short doc explaining intended usage
• remember that GitHub's repository social preview card may still require a manual upload in repository settings even when the image asset is committed in-repo; do not assume the README image automatically becomes the repo card
• update repo metadata after structure stabilizes: description, topics, and release notes

3. Make README hybrid, not overloaded

README should:

• explain what the package is
• surface the distinctive value proposition early
• cite upstream conceptual lineage when relevant
• keep detailed rollout procedures in deeper docs
• optionally include a concise For Agents tail
• when multiple languages exist, put neutral language-switch links near the top of the README instead of special-casing one language as an afterthought

README should not:

• become one giant system prompt
• duplicate the full implementation guide
• hide the real package value behind only structure/install text

4. Separate package layers from rollout order

Implementation/deployment guidance should clearly distinguish:

• what artifacts exist
• in what order to land them into a host system

Also provide:

• capability-based landing guidance
• minimum / standard / full adoption profiles when helpful
• smoke tests or post-install verification

Shared Pitfalls Across the Lifecycle

• guessing category placement instead of checking the live local taxonomy
• copying whole repos when the request is only for one skill
• losing support files because only SKILL.md was copied
• calling a discoverable install broken just because an optional script dependency is missing
• publishing a repo with local-only assumptions still embedded
• letting README, AGENTS, runtime docs, and deployment docs duplicate each other

Verification Checklist

Before claiming success, confirm the relevant half of the lifecycle:

For installs

• correct destination path under ~/.hermes/skills
• requested skill copied without unrelated extras
• support files preserved
• skill discoverable via skills_list
• skill loadable via skill_view
• runtime caveats reported separately

For packaging

• README is human-readable first and not overloaded
• deployment docs separate structure from rollout order
• reusable assets are actually portable
• local/private assumptions were removed or documented
• there is at least one concrete adoption/verification path

Expected Outcome

A good run of this skill should leave the user with either:

• a correctly installed external skill that Hermes can discover and load

or

• a reusable skill/methodology repo that another agent can adopt without structural guesswork.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: xiaohei-info
• Source: xiaohei-info/oh-my-agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.