| name | agent-skills-authoring |
| description | Use when creating or updating repo-local skills for this reading-list repo and you want them aligned with the Agent Skills specification, recent workflow learnings, and reusable references/scripts. |
| version | 1.0.0 |
| author | Bosun |
| license | MIT |
| metadata | {"hermes":{"tags":["skills","authoring","agentskills","repo-maintenance"],"related_skills":["reading-list-repo-maintenance","telegram-reading-flow"]}} |
Agent Skills Authoring
Overview
This skill governs how Bosun should create and maintain repo-local skills in this repo.
It is the meta layer for keeping process documentation clean, reusable, and consistent with the Agent Skills specification at https://agentskills.io/specification.
Use this when adding a new repo-local skill, restructuring an existing one, or syncing local workflow learnings back into the checked-in skills.
When to Use
Use when:
- a workflow lesson should be captured in a repo-local skill
- an existing skill has drifted from how the repo is actually maintained
- a skill should gain reusable
references/ docs or scripts/ helpers
- a local-only skill and its repo-local counterpart need to stay aligned
Do not use when:
- the task is a one-off repo edit with no reusable workflow lesson
- the change belongs only in content/docs and does not alter how Bosun should operate
Workflow
-
Inspect the existing skill surface first.
- Read the current
SKILL.md, linked references, and any sibling skill that overlaps.
- For reading-list workflow changes, check both the repo-local skill and the local
work-reading-dropbox skill when relevant.
- Completion criterion: you can point to the exact files that should change.
-
Align with the Agent Skills spec before adding repo-local conventions.
- Follow the required
SKILL.md frontmatter and directory layout from the spec.
- Prefer putting stable instructions in
SKILL.md, longer rationale in references/, and reusable code in scripts/.
- Completion criterion: the skill shape would still make sense to a reader who only knows the upstream Agent Skills specification.
-
Keep skills progressively disclosed.
- Put trigger conditions, workflow steps, pitfalls, and verification in
SKILL.md.
- Move examples, policy notes, tagging guidance, schema details, or command recipes into
references/ when they are useful but not always needed.
- Add a reusable helper under
scripts/ when a repeated analysis or validation step is likely to come up again.
- Completion criterion: the main skill stays readable without losing important detail.
-
Prefer reusable scripts invoked with go run.
- In this repo, prefer checked-in Go helpers for lightweight reusable scripts because
go is available locally and uvx is currently unavailable.
- If Python is the better fit, keep it as a secondary option and document why.
- Completion criterion: any new helper script has a clear invocation example and a plausible reuse case.
-
Keep local and repo-local workflow skills in sync.
- When the Telegram-to-reading-list workflow changes, patch both the repo-local skill(s) and the local
work-reading-dropbox skill unless the difference is intentionally repo-specific.
- If there is intentional divergence, say so explicitly in the repo skill or reference notes.
- Completion criterion: future Bosun sessions do not learn conflicting versions of the same workflow.
-
Verify the skill as an artifact.
- Check frontmatter shape, linked-file references, and script paths.
- Run the reusable validator script when editing repo-local skills.
- Completion criterion: the skill directory is internally consistent and committed with the repo changes it documents.
References and Scripts
- Read
references/agentskills-spec-notes.md for the repo's distilled interpretation of the Agent Skills specification.
- Use
scripts/validate_skill.go to sanity-check a repo-local skill directory.
Common Pitfalls
- Keeping all knowledge in
SKILL.md and turning it into a wall of text.
- Adding one-off advice without deciding whether it belongs in docs, a skill, or a helper script.
- Forgetting to sync the local
work-reading-dropbox skill after changing the repo-local workflow.
- Creating tags, policies, or file layouts inside one skill that silently disagree with another.
- Writing a helper script with no documented invocation or no realistic reuse path.
Verification Checklist