最近新增
刚加入收藏的新技能，抢先试用。
measure-agents-honestly
musterd-io
How to measure whether a multi-agent setup actually helps — a pre-registered manifest whose ruler cannot bend to fit the result, wasted work reconstructed from git alone, results that diagnose rather than rank, and the denominator that decides whether your headline is a finding or a sales pitch. Use before running any agent comparison, when writing up one, or when someone quotes a multiplier at you.
mast-checklist
musterd-io
Read your own multi-agent coordination log against the failure shapes that research has catalogued — broadcast-journalling, unanswered requests, one-sided threads, silent participants — with a screen that names where to read rather than scoring you. Cites the taxonomies it draws on instead of paraphrasing them. Use after running a multi-agent setup for a while, when coordination feels busy but nothing closes, or before claiming a team is working.
hooks-that-reach-the-model
musterd-io
A hook that runs and a hook whose output the model actually sees are different facts, and most harnesses only tell you the first. The per-harness reachability table with a falsifier per row, the house style that keeps a hook from breaking someone's turn, and a canary that measures your own harness instead of trusting a table. Use when wiring any agent-harness hook, when an injected reminder is being ignored, or when a hook "works" and nothing changes.
emitted-is-not-published
musterd-io
Storing a teammate's prose so their teammates can read it is the product; publishing it is a separate act with its own permission. Structural fields ship pseudonymised per release, prose bodies are omitted by default, consent is per-author and fails closed — and there is deliberately no scrubber, because regex PII-stripping over free prose gives false safety. Use before releasing a dataset, opening a corpus, or sharing anything a colleague wrote without meaning it for strangers.
dreaming
musterd-io
How a seat consolidates overnight — save the day's traps and numbers as insights, promote anything seen twice into a wiki page with a date and a falsifier, as a PR that wiki:check gates. Use when Presence provenance is dream, or when you hold the dreaming write path.
handover-flush
yotamleo-org
Session-end sweep across handover/* branches — push, open missing PRs, report merged. Use for /handover-flush.
context-hop
yotamleo-org
Jump mid-session to a fresh claude session when context nears the soft budget. Use for /context-hop.
nemotron-policy-generator
toxicwind
"Generates BYO custom safety policies for NVIDIA Nemotron content-safety guardrails — Nemotron-Content-Safety-Reasoning-4B (text) and multimodal Nemotron-3-Content-Safety. Produces a Markdown policy, JSON taxonomy, and drop-in inference prompts. Maps rough words or an existing policy to V2 categories, adding custom categories or topic-following rules."
ast-bm25-racer
toxicwind
Estate AST-BM25 Racer, local-heavy edition: hybrid code retrieval plus nanosecond candidate racing. Triggers on: code retrieval, candidate racing, AST search.
phone-backup
toxicwind
Bulk backup and storage triage for User's Pixel 9 Pro XL via ADB on yote. Benchmarked fastest pull (parallel adb pull, NOT tar-stream), phone storage map, the toybox find quirk, 8TB archive layout, and the future-forward Syncthing continuous-sync direction. Use when User asks about phone storage, pulling files off the phone, or freeing phone space. Photos (DCIM/Pictures) are excluded by standing direction — this skill handles code, archives, repos, and documents only.
issuelooper-shared
kazunori279
"Shared reference for this repository's issue loop (issuelooper): the roles (worker, merger, tester, monitor, aide), how the supervisor ccpool.py runs them, the label scheme and who may set what, the merge gates a pull request must pass, comment signing, and the limits every session works under. Read it first, then .issuelooper/project.md, then the skill for your CC_ROLE."
deckcraft
dnntrn
Craft beautiful technical presentation slide decks using Slidev. Activate when the user asks for slides, a presentation, a deck, a talk, conference material, training content, or demos. Active when working in deck/ or creating Slidev markdown files.
security-portable
musterd-io
The security role's playbook for threat modeling, security review, authorized testing, validating and rating findings, handling secrets, fixing vulnerabilities, and reviewing dependencies and releases. Use when a team asks for security evidence or a security decision.
implement-spec
lirtual
"Implement the result of /to-spec and /to-tickets in code."
implement
lirtual
"Implement a piece of work based on a spec or set of tickets."
designer
musterd-io
The designer role — the member who owns what people see (product UI, live and real-time views, the marketing site, command-line output, design files) and judges every visual change before it ships. Use when designing or changing any surface a person sees, writing a spec or tokens for an implementer, capturing and checking a UI, reviewing someone else's UI for acceptance, or checking accessibility, motion or the performance budget of a page.
definition-of-done
musterd-io
What "done" has to mean before anyone says it — docs, traces and an eval as peers of tests, landed verified against the trunk rather than assumed, and a second party claiming it back. Use when finishing a task, defining a team's done, or when "done" and "merged" have started to drift apart.
decision-records
musterd-io
The whole lifecycle of a numbered decision record — taking a number against everything in flight rather than everything that landed, publishing it before you write, freezing the Decision on accept so amendments are dated and append-only, and naming what would tell you it was wrong. Ships a checker that refuses a rewritten Decision, including across a rename. Use when adopting ADRs, when two records collided on a number, or when someone edits a decision that already shipped.
courier
musterd-io
The courier role — the member who moves team material between the team's machines and networks (files, artifacts, datasets) when git and the sync surface are the wrong carrier, and refuses what should not move. Use when you hold the courier role and a teammate asks for something carried, or when you need something moved and must find who can do it.
controls-in-force
musterd-io
Every guard you believe is protecting you carries the date someone last watched it work, whether it has ever actually caught anything, and an honest answer to "would it have caught the incident that motivated it?" — with a hundred-line checker that fails the build when the evidence goes stale. Use when adopting a guard you did not write, when a gate has been green a suspiciously long time, or when someone cites a control as satisfied.
comms
musterd-io
How a team speaks in public, and the evidence every public word rests on — messaging, launch narrative, release notes, press, public docs and READMEs, UI-copy specs, positioning and GTM proposals, user research synthesis, charts, pitch decks, investor and accelerator writing. Use for any writing that leaves the repo, any figure that carries a number, any claim about users or the market, and any UI-copy spec handed to the designer. Enforces the brand voice, the canonical glossary and one evidence gate; per-craft procedures live in references/.
claims-ledger
musterd-io
When a claim someone put on the record turns out to be wrong, the correction mints one entry — who claimed it, who caught it, through which channel, how long it stood, what it cost. Minted by the corrector riding an act they are already performing, never by a sweep; no bare rates, ever; and self-correction scores best. Use when a team keeps being confidently wrong, when you want to know whether review catches anything, or before building any "quality" metric over people.
building-a-role-skill
musterd-io
How to build or rebuild the skill a musterd role carries, by one measured method — read the charter and sources, research the craft, survey upstream skills under a licence check, write a lean core with references, evaluate with and without the skill graded blind, fix and re-run, and have a holder of the role accept it. Use when creating a custom role's skill, rebuilding an existing one, or deciding whether a role's skill is good enough to ship.
a-finding-is-not-a-fix-request
musterd-io
A review finding is REQUIRED only if the spec would have demanded it before anyone opened the diff — discovery adds information, not obligation. Four categories earn a block; everything else is a note that routes to the board under the finder's name, and complying with an out-of-scope REQUIRED is the failure mode, not the polite option. Use when reviewing a change, when a review has grown the work, or when a reviewer and an author are stuck.