btm-skills

Plugin btm-skills from marketplace btm-skills in BTreeMap/SKILLs. Commands read from the repo's manifest — 21 days ago.

Skills (21)

• advisor

  Reads research project like a principal investigator: which prior ideas it combines and adds to, which setup parts are outdated, what your lab can afford, cheapest experiment proving or killing each claim. Every current-practice claim carries a source or is marked as memory. Use when user shares a p

• aesthete

  Designs, builds, reviews web interfaces meeting WCAG 2.2, following supplied brand or design system, avoiding templated look of generated UI, on marketing, editorial, product surfaces. Use when designing, building, auditing, or redesigning any web interface, applying a design system, or designing co

• asd-ste100

  Writes, rewrites, checks text in ASD-STE100 Simplified Technical English against Issue 9 dictionary and writing rules, naming each unapproved word with approved alternatives. Use when asked for STE, controlled English, or an STE check. Requires uv and a full SKILLs repository checkout. The first run

• author-skill

  Turns procedure or task history into a SKILL.md another agent can follow with no memory of original session; reviews existing skills against Agent Skills standard. Use when asked to create, refactor, review, or distill a skill. argument-hint: "[skill name or path]

• caveman

  Compresses replies into terse phrasing keeping every technical fact: code, numbers, units, negations, error strings stay exact. Terseness ranges from tightened prose to one-word answers, in English or classical Chinese. Use when user asks for caveman mode, "be brief", fewer tokens, or longer-lived c

• draft-paper

  Drafts conference, workshop, journal, survey, or demo papers, answers reviews; each empirical claim traces to an artifact, each citation to a retrieved record. Use when turning idea or results into a draft, retargeting one, or rebutting. Requires uv and a full SKILLs repository checkout. Venue and l

• fact-check

  Checks a document claim by claim against sources it retrieves, reporting each verdict with verbatim quotes, URLs, access dates. Never verifies from memory; changes no text until user approves that correction. Use when asked to fact-check a document, verify claims, specs, statistics, or version numbe

• finance

  Keeps a private record of slow-money stock bets, each with stated odds and kill date, measured by deflated Sharpe, PBO, and purged walk-forward at the book's asset cap, sized by evidence under a ruin bound; operator replies in chat and sends every trade by hand. Use when setting up or continuing an

• git-commit

  Drafts and reviews Conventional Commits messages, why over what; can commit and push in one step. Use when writing or reviewing a commit message, or asked to commit or push. argument-hint: "[basic|full|maximum] [push]

• humanize

  Rewrites AI-sounding prose so it reads like its writer, keeping every claim, number, citation at original strength, inventing nothing. Writing sample or style file outranks its catalogue of AI writing signs. Pasted text comes back rewritten; named file edited in place. Use when asked to humanize, de

• lit-review

  Produces literature review whose every citation traces to a paper retrieved from OpenAlex, arXiv, or Crossref, never memory; criteria fixed before first search, each exclusion keeps its reason, abstracts never pass as full text. Rigor ranges from scoping pass to PRISMA-style. Use when asked for a li

• peer-review

  Reviews a paper as adverse referee: contribution claims taken verbatim, objections accepted only when they quote a page and cite dated prior work from its own literature search, recommendation following from what survives. Use when asked to review, referee, red-team, or find weaknesses in a paper, m

• pl-theorist

  Brings programming-languages theorist's discipline to design, code, review, tests: immutable data, algebraic types behind smart constructors, illegal states unrepresentable, explicit effects, data structures fit to dominant operation. Tuned for Python, JavaScript, TypeScript, Rust, Go, Haskell, C, C

• ponder

  Answers open question and shows its work: every load-bearing claim carries a source, inferred conclusions marked, strongest rival explanation tested, what stays unsettled reported open. Use when user asks open question needing researched, sourced answer. Requires uv, retrieval (the harness's web sea

• ponytail

  Forces laziest solution that works: asks whether task needs doing, then standard library before custom code, native platform features before a dependency; review and audit look only at over-engineering; every shortcut recorded as debt. Use when writing, refactoring, reviewing, or designing code, cho

• read-pdf

  Extracts text and metadata from a PDF file or URL, answers questions about it with page-cited evidence. Only reads, writes nothing, no OCR: scanned page reported as having no text. Use when asked to read, summarize, search, quote, or answer questions about a PDF. Requires uv, and a full SKILLs repos

• reframe

  Turns design or planning discussion into testable direction judgment: thesis, binding constraint, target worth aiming at, what to stop doing, three costed routes from conservative to clean, evidence that would prove it wrong. Use when asked to think bigger, escape incrementalism, reconsider legacy c

• search-web

  Searches web, Wikipedia, scholarly record; pulls readable text out of a page; papers come with DOI, year, citation count. Use when harness has no search or fetch tool of its own, or a question needs papers by DOI. Requires uv, network access, and a full SKILLs repository checkout. The first run buil

• setup-env

  Installs a project's development toolchain in userspace: no sudo, no docker, nothing assumed but uv, all under one disposable root leaving HOME and caches untouched. Tags such as [email protected], kotlin:android, or rust name what project needs; tools built for another CPU architecture still run. Use whe

• summon

  Hands task to another agent: whether to delegate, what to tell delegate, how parallel agents stay apart, how to judge output as untrusted text. Covers Claude Code, GitHub Copilot, Google Antigravity, OpenAI Codex, DeepSeek Harness, OpenAI Agents SDK. Use when writing a subagent prompt, deciding whet

• thematic-analysis

  Develops themes from qualitative text under one named school: choice recorded, every unit coded against bounded codebook, each theme backed by verbatim extracts and counts. Defaults suit feedback, tickets, reviews, usability sessions, interviews. Use when asked to find themes in qualitative data, co

Similar plugins