skill-pack-curator
Skill0 starsChanged 39 days agoMITView on GitHub
Designs the membership and structure of a skill pack - persona, member selection, install sequence, cross-links, and what to add or cut. Use when someone asks "which skills should go in this skill pack", "design the membership of a new skill pack", "decide which skills to add or cut from a pack", or is turning a vertical (an industry or role) into a coherent bundle. Do NOT use to grade or rewrite a finished pack to a quality bar (use skill-auditor) - this designs composition.
In this repository4 other files it gives its agents.
- skill-auditorSkill
- skill-creatorSkill
- skill-description-writerSkill
- skill-testerSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 5 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 46
Do not admit a skill because it is good - admit it because the persona hits that stage of the job.
- Distinctive: line 47
Do not ship members that do not reference each other.
- Distinctive: line 48
Do not let two members define the same threshold differently.
- Distinctive: line 49
Do not pad to a target size, and do not span two personas because both are adjacent - split instead.
- Distinctive: line 50
Do not grade or rewrite member quality here.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Skill Pack Curator
A pack is not a dump of related skills. It is a curated operating system for one persona or one workflow, where every member earns its place, the members route to each other, and the whole covers the job end to end. The costly mistake this prevents: shipping a "collection" - ten adjacent skills with no sequence, no cross-links, and no reason to buy the bundle over any one member.
Inputs to collect
- The persona or workflow, in one sentence: "This pack is for [persona] who need to [job]."
- The existing catalog - which candidate members already exist versus need creating.
- The methodology, if any, the pack encodes (a named framework, a book, a practitioner playbook).
Persona packs vs workflow packs
A persona pack serves one role across many task types ("Engineering Manager Toolkit"). A workflow pack covers one end-to-end workflow ("Database Migration Workflow"). Choose one mode and hold it; mixing both makes membership decisions arbitrary. Default: vertical/industry packs are persona packs; process packs are workflow packs.
Operating procedure
- Write the persona sentence and the end-to-end job map: every stage the persona moves through, in order (e.g. for a gym owner: know the math → build the offer → fill it → sell it → keep them → scale). The job map is the pack skeleton; members exist to cover stages.
- Slot candidates against the map. For each stage, name the skill that covers it - existing catalog skill (reuse), existing skill needing uplift, or a gap needing creation. A stage with no skill is a coverage hole; a skill covering no stage is a cut candidate.
- Enforce the overlap budget. No two members should fire on the same prompt more than roughly one time in five. Test with each member's positive trigger scenarios: if another member would also fire on more than one in five, merge them or cut the weaker. Coordinated handoffs (one skill's deliverable feeds the next) are not overlap - they are the point.
- Size the pack. Minimum 3 members (below that it is one skill with sections), maximum 12 (above that, split), sweet spot 5-9. Never pad to hit a number: a weak member taxes the pack's credibility more than a gap does.
- Design the connective tissue - what makes a pack worth more than its members:
- An install sequence with reasoning: which skill first and why ("install gym-money-model first; every other skill spends money and this one says whether the spend pays").
- Cross-links by plain slug mention in each member's body, routing at the natural handoff ("if conversion is under 40 percent, go to closer-sales-script").
- One worked example threaded through every member - the same example business, project, or person throughout, so the skills visibly compose.
- Shared thresholds: a number defined in one skill (a target, a rule) is cited identically by the others, never contradicted.
- Write the pack description. Tagline (90 chars max) names the audience and the outcome: "For gym owners: fill, sell, and scale using the money-model playbook." Description sells when to reach for the pack and what running it end-to-end produces - never a contents list.
Deliverable
A pack manifest: persona sentence, job map, the member list annotated reuse / uplift / create with each member's one-line trigger, the install sequence with reasoning, the cross-link plan (which member routes to which, at what condition), the worked example that threads through, and the tagline + description.
Do NOT
- Do not admit a skill because it is good - admit it because the persona hits that stage of the job. Off-persona members belong in another pack.
- Do not ship members that do not reference each other; an unlinked pack is a discount bundle, not a product.
- Do not let two members define the same threshold differently; shared numbers are defined once and cited.
- Do not pad to a target size, and do not span two personas because both are adjacent - split instead.
- Do not grade or rewrite member quality here; route finished composition to skill-auditor for the quality pass.
Quality bar
The composition is done when: every stage of the job map has exactly one owning member; install order is stated and justified; every member cross-links at least one other at a concrete handoff condition; one worked example appears in every member; no overlap pair exceeds the budget; and the tagline names audience + outcome in under 90 chars.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

