Profile My skills Saved Collections Edit profile Submit a skill

← all publishers

patebry

@patebry source repo

3 published skills

  1. ▌
    Distill · patebry
    Turn a body of source material into a short, defensible document. Not for reviewing code, diffs or pull requests, tasks ending in a code change, or exploring and documenting a codebase — use code-review, security-review, simplify or init. Not for a single short document or a straight summary. Use for contract bundles, RFPs, spec bundles, research corpora, interview transcripts, incident evidence — dense material whose correctness cannot be assumed, where the output will be acted on and being wrong is expensive. Triggers on "distill", "synthesise these", "what do these documents actually say", "produce a memo from this", "verify this memo against its source corpus".
    0
    installs
  2. ▌
    Humanize · patebry
    Rewrite an outbound message to a person (Telegram, Slack, email, a PR or issue comment, a customer or partner update) so it reads as written by a human rather than generated, and cut it to the length it deserves. Not for code comments, documentation, README or spec prose, commit messages, or the assistant's own replies in the terminal; those keep their own conventions. Not for translating, proofreading for grammar only, or drafting a message from scratch with no source text. Use when a draft exists and reads stiff, padded, over-structured, or machine-written, and a real person is about to receive it. Triggers on "humanize", "make this sound human", "cut this message in half", "remove the em dashes", "make it less AI", "just give me a quick message", "shorten this before I send it".
    0
    installs
  3. ▌
    Review Pull Request · patebry
    Review a GitHub pull request or code diff for actionable correctness, security, data-integrity, lifecycle, contract, architecture, and test defects, then publish one terse final review when permitted. Use for PR review, audit, judgment, or assessment; request changes only for proven merge blockers, distinguish defects from verification gaps, use comment-only reviews for self-authored PRs, and do not publish clean results.
    0
    installs