skilldrop
Agent skills that are measured against the agent without them, and retired when they lose. Skills for the work people ship: ADRs, PRDs, runbooks, threat models, decks, SQL reviews, DPIAs and Terraform plan reviews. Each one is a folder you copy into Claude Code, Cursor, Kiro, Codex, Antigravity or Copilot.
npx skilldrop-cli install --pack dev-team # or product-manager, solution-architect, sre-oncall …
The same request, with and without a skill
From a held-out eval, written blind from the skill's description:
"We're taking ClaimLens to the investment committee on Thursday… Can you build a lean canvas for it and set out our riskiest assumptions and how we'd test them?"
| Plain Claude Code | With lean-canvas |
|---|
| Finds the corrected break-even (~225 seats), then recommends "scope the ask to Claims-only Phase 1". | "Corrected, break-even is 225 seats — more than all three departments combined (215), so full rollout still doesn't cover running costs at £40/seat." |
The blind judge picked the skill's answer: "A misses this and recommends a Claims-only phase that cannot break even."
The receipts
Every skill runs as a real agent session twice: once with the skill, once without. The results are published either way. In the last full run (23 October 2026: 117 skills, 520 evals, claude-sonnet-5):
- With a skill, answers met 73% of their checks; without one, 53%. The checks were written by each skill's author, so the fairer numbers are the next ones.
- Two blind judges preferred the skill's answer in 62% and 61% of pairs. Each judge saw only the request and both answers, in shuffled order. On the 176 held-out evals that no revision or admission was based on, it was 49% and 48%: no preference.
- On Sonnet 5.5, now the standard tier, the judges preferred the plain agent. On all 308 held-out evals the skill won 40% and 43% of pairs (what the benchmark showed).
- 4 skills that didn't beat the plain agent were retired, and they're listed. Each new skill has to win on held-out evals before it ships.
Get started · Browse the packs · Search all skills · Docs · What stays stable in 1.x · Contribute
Start in one command
npx skilldrop-cli install --pack dev-team
The install prints a starter prompt for the pack. For dev-team, ask your agent:
Is this branch safe to merge? Run a pre-merge review on the current diff.
A good result runs your lint, typecheck and tests through a gate script, then a reviewer panel, and ends with READY or NOT READY. To see any pack's starter, run npx skilldrop-cli info --pack <name>.
In Claude Code you can use the plugin marketplace instead: run /plugin marketplace add sananthanarayan/skilldrop, then /plugin install dev-team@skilldrop. Every install route covers single skills, other IDEs, loops, reviewer subagents, profiles and updates.
Install one pack. Every role pack also brings core (intake, critique, review council, output hygiene, and the decisions behind a codebase). Each pack page has a starter prompt to paste and says what a good result looks like.
- Product managers:
product-manager — PR/FAQs, strategy, OKRs, business cases, PRDs, success metrics. Open pack
- Architects:
solution-architect — diagrams, ADRs, design docs, API and data contracts, threat models. Open pack
- Software teams:
dev-team — story splitting, implementation with adversarial review, merge gates, PR descriptions, a tech-debt register, launch readiness, release notes. Open pack
- SRE and on-call:
sre-oncall — observability, runbooks, incident comms, postmortems, capacity. Open pack
- Explaining things to decision-makers:
stakeholder-comms — audience profiles, exec summaries, decision logs, guides. Open pack
- On-brand decks and flyers:
design — capture your brand once, then build .pptx decks and print-ready flyers with your logo, colours and fonts. Open pack
- Adopting and building AI:
ai-engineering — readiness, use-case triage, policy, agent loops, budgets, agent threat models, evals. Open pack
- Building on the Claude API:
api-builder — prompt caching, token budgets, eval generation, tool schemas. Open pack
- Moving between formats:
converters — Markdown to Word, Excel and HTML, files back to Markdown, and Mermaid diagrams checked and rendered.
How it works
- A skill produces one file you own. Each ships a quality bar, named anti-patterns and acceptance evals, so the output is an artifact, not a conversation. See every skill, by category.
- Loops order skills, and a gate decides when work moves on. Five loops cover the lifecycle (
discover, design, build, release, operate), plus the ship-a-draft wrapper for any document and apply for a job application. Each gate is a script, a review panel or a person, chosen by how expensive the mistake is to undo. See the operating model for every stage and gate on one page, the loops, stage by stage, and why loops.
- Measured, not asserted.
run_bench.py runs each acceptance eval with the skill and without it, in a sandbox, and reports the lift, a blind preference from two judges, and the cost per run. The results are published whether or not they flatter the skills. See how skills are checked and RFC-0040.
- Copy, never transform. A skill is a plain
SKILL.md folder in the Agent Skills format, at packs/<pack>/skills/<name>/. What runs in your agent is byte-identical to what is reviewed here. There is no runtime, and the tooling has zero dependencies.
- Updates keep your edits.
npx skilldrop-cli update replaces files you haven't touched. For any file you changed, it leaves the new version beside it as <file>.upstream. See upgrade installed skills.
Go deeper
License
MIT or Apache-2.0, at your option. See LICENSE.
