manage-skills

Imported from s2b-dev/smart-second-brain (src/skills/defaults/manage-skills/SKILL.md). Install upstream with npx skills add s2b-dev/smart-second-brain --skill manage-skills. Copyright stays with the author.

Managing Skills

Every operation below applies immediately — there is no staging or review step. Be deliberate; by the time you get a result back, the change has already happened.

Skills are where task knowledge lives: how a kind of task is done here, the pitfalls, and the user's corrections about that work. Memory holds facts about the user and pointers into the vault; it is not the place for procedures.

Revising a skill — load it with load_skill first. Patches and updates are refused until you have read the skill in this conversation, so the change is written against its current text.

• Revise the skill you used whenever it misled you or lacked a step, and whenever the user corrects how this kind of task should be done. Do it before you finish the task; those corrections live here, not in memory.
• Do this to make verified knowledge permanent: after you discover how a task actually works (e.g. a plugin's real API), fold the concrete methods, arguments, and patterns into the skill so future runs skip re-discovery.
• Fix in place: change the sentence that was wrong. Never append an "Update:" paragraph under it.
• Write lessons, not logs: one rule per point, imperative plus one clause of why. No narration of what happened, no dates.
• Only write instructions you have actually confirmed — never speculative or unverified steps. A skill full of guesses is worse than none.
• A skill's name and plugin link are locked; you can change the body and description only.

Patch — the normal way to revise: give the exact passage to replace, copied from the loaded skill, and its replacement. The passage must match exactly once, so include enough surrounding text to make it unique. An empty replacement deletes the passage. Frontmatter cannot be patched.

Update — replace the whole body (and optionally the description). Use it only when the skill needs restructuring; for anything smaller, patch.

Create — author a brand-new skill: give it a name, a one-line description, and instructions (body). You may request a small set of built-in tools for it via allowedTools; only tools from an allowed subset are granted, others are silently dropped.

• Create when the workflow has no home in an attached skill; prefer extending a skill that already covers the area.
• A skill you create is attached and usable immediately — there is no separate enable step.
• Keep new skills narrow and instructions concrete — write down only what you'd actually want to remember doing again.

Delete — remove a skill you created, immediately and without confirmation. Built-in core skills cannot be deleted.