ZachArticulateV

build

"Implement planned work with proof at every step: execute a plan or tickets unit by unit, running independent units in parallel; refine code just written; or improve a measured metric, keeping only the wins that clear the noise. Use when the user asks to implement, build, execute a plan or ticket, clean up freshly written code, or make something faster or smaller."

ZachArticulateV 1 Actualitzat fa 2 h

Recursos

2
GitHub

Instal·la

npx skillscat add zacharticulatev/compound-memory/build

Instal·la des del registre de SkillsCat.

SKILL.md

Build

Make the change the plan describes, prove each piece as it lands, and leave a
history someone can follow commit by commit.

Done when: every unit you built has a check suited to it that you ran and
saw give the right answer (new behavior: it failed before the change and
passes after; cleanup: the same results before and after; configuration or
documents: they parse, build, or lint), the project's own suite passes, no
temporary debug code is left, and the work is committed.

Modes

$0 Use it to Reference
work Implement a plan, tickets, or a request, unit by unit references/work.md
simplify Refine code just written before it goes to review references/simplify.md
optimize Improve one metric through measured experiments references/optimize.md

Pick the mode the request describes. With no mode word and nothing in the
request to decide it, use work. Read only the reference for the mode you run.

How it works

  1. Ground. Read the plan or ticket, memory/ when it exists, and the
    lessons for the paths you will touch:

    node "${CLAUDE_SKILL_DIR}/../../scripts/lessons.mjs" --paths <dir>,<dir>

    On the repository's default branch, create a feature branch before the first
    commit. When the working tree already holds edits, snapshot it before your
    first change, so the commit check can tell the user's edits from yours:

    node "${CLAUDE_SKILL_DIR}/../../scripts/changes.mjs" snapshot

    When the project has a verification skill (.claude/skills/verify-* or
    .agents/skills/verify-*), note it: verify uses it to prove user-facing
    changes.

  2. Run the mode from its reference.

  3. Keep the plan true. When the plan turns out wrong, fix the plan file and
    say what changed, rather than quietly building something else. A decision
    the plan records as the user's is not yours to change: when it proves
    unworkable, stop and ask.

  4. Hand off. Run verify for user-facing changes and review before
    shipping. When the work crosses skills or sessions, keep the check results
    in a run record (../../references/artifacts.md) so the next stage reads
    them instead of re-running them blind; a one-sitting change does not need
    one.

The habits in ../../references/principles.md apply throughout, especially
building what was asked and proving it works.

Works with

  • Reads: a plan or tickets from plan, lessons, a prototype's validated
    module, the project's verification skill.
  • Hands off: commits to review and ship; finished tickets marked done;
    a lesson worth keeping to learn capture.
  • Alone: given a plain request and no plan, write a short plan in the reply
    (units, each with a check) and build from that.

Paths in this skill and its references are relative to this skill's directory
(${CLAUDE_SKILL_DIR}). Outside Claude Code, that is the directory holding this
file, and $0 is the first word after the skill name.