agentleFS
SantiagoVqzSantiagoVqz/claude-skills/skills/2-build/build/SKILL.md

build

Skill2 starsChanged 12 days agoMITView on GitHub

Plan and build one ticket in this session: read it and the code, check the hard stops, get one approval, pick the approach, and implement. Also takes the next ticket of a spec, or a change in words.

169 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use backprop, finances and 9 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • JuliusBrusseeSkill

    backprop

    Bug → spec protocol. When a bug is found or a test fails, trace the cause, decide whether a new §V invariant would catch recurrence, append to §B. This is the one non-obvious thing SDD does that plan-then-execute doesn't. Triggers on test failure, bug report, post-mortem, or explicit user ask.

    Also in 4 projects that use this

  • GerardoRdz96Skill

    finances

    Use when Gera talks about his money — logging an expense ("I spent X on Y"), checking where he stands this pay-period, closing a period, planning debt payoff, onboarding his financial picture, or asking for budgeting/finance opinions. Servy's financial brain. Triggers — "I spent", "log this expense", "where am I this period", "close the period", "how's my budget", "debt plan", "should I pay off", "finances onboard", "/finances".

    Also in 3 projects that use this

  • cpiprintSkill

    favicon

    Generate favicons from a source image

    Also in 3 projects that use this

  • madeehahassanSkill

    Go-Live

    Pre-launch checklist. Use before making any app public, sharing a link externally, or announcing a product launch. Adapts to Claude Code, claude.ai, and Cowork.

    Also in 3 projects that use this

  • TrevokeSkill

    define

    Use when a new feature request, bug report, or work request needs requirements defined before implementation

    Also in 3 projects that use this

  • HAR5HA-7663Skill

    shape

    Plan the UX and UI for a feature before writing code. Runs a structured discovery interview, then produces a design brief that guides implementation. Use during the planning phase to establish design direction, constraints, and strategy before any code is written.

    Also in 6 projects that use this

The file

We split this skill into the rules it gives an agent

8 rules. 0 are common, 3 are distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 155 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read the ## Agent skills block in AGENTS.md, then docs/agents/tracker.md.

    line 14
  2. Very specific:

    Spec or project with open sub-issues: stop. Name each open sub-issue with no open blocker.

    line 18
  3. Very specific:

    Idea: stop. The next step is /spec <n>.

    line 19
  4. Distinctive:

    Read the code that the ticket touches, and git log on that area.

    line 26
  5. Distinctive:

    Write the plan in your message, in about 15 lines.

    line 49
  6. Very specific:

    When the parent spec has a Domain changes section that the trunk does not have yet, write it with DOCS-FORMAT.md, and commit: Record terms …

    line 73
  7. Distinctive:

    When the work shows a hard stop that the plan does not name, stop, and ask: "Stay light", with a reason, or "Full plan".

    line 92
  8. Very specific:

    Run the typecheck, the lint, and the tests that the diff adds or touches.

    line 100

“Common” is measured: kept by at least 6 in 10 of the 155 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Build

/build takes one ticket to one commit on a branch, in this session, ready for /ship. The user sees the work as it happens, and can change it at any point. It is light by default: a short plan and one approval. A hard stop gets the full plan, in this same session.

1. Read

Read the ## Agent skills block in AGENTS.md, then docs/agents/tracker.md. When the block is missing, read NO-SETUP.md in the folder of the setup-skills skill, and go on with its sources. Read the issue with its comments and its parent.

Find what the argument is. A label such as Remissions changes nothing here.

  • Spec or project with open sub-issues: stop. Name each open sub-issue with no open blocker. The next step is /build <that issue>.
  • Idea: stop. The next step is /spec <n>.
  • Spec with no sub-issues: one ticket, with its stories as the criteria.
  • Words with no issue: one ticket. The plan is the record, so write nothing to the tracker, and the branch is fix/<slug>.
  • Any other issue: one ticket, also with no criteria yet.

Stop when an issue in blockedBy is open, and name it. When someone else has the issue, ask with the question tool: "Take over" or "Stop". When the branch exists and the issue has a ### Plan of #<n> comment, switch to the branch, and resume at step 5 from its commits.

Read the code that the ticket touches, and git log on that area. Ask whether the ticket is a symptom of a cause one layer down.

Completion: you can name the current behavior or the cause, the fix, and the code that the fix changes.

2. Hard stops

A hard stop is a change that needs more than a short plan:

  • a migration or a backfill
  • auth or permission code
  • a write to production data
  • a new or materially changed user-visible UI element
  • an API contract that another repo or app consumes
  • the third issue on the same symptom: search all issues and git log for it

Each item is a class of past incident. Change the list only from /retro evidence.

A hard stop that the parent spec already settled stays light. Name it on the verdict line. For any other hard stop, or when the user asks for it, follow PLAN.md, then go to step 4. The user can say "stay light", with a reason.

Completion: you have the verdict line: Hard stops: none, or Hard stops: <which> (<settled in #n, or override: reason>).

3. Plan

Write the plan in your message, in about 15 lines. A plan that needs more goes to PLAN.md.

Issue: two or three sentences in plain terms, with no symbol names.
Cause: the cause or the current behavior, and your confidence in it.
Fix: the change, and the code that it touches.
Done when:
- each criterion, with the edge cases
Out of scope: what stays as it is.
Defaults: each choice that you took without asking.
Approach: from the table in step 5, with the reason in a few words.
Hard stops: the verdict line.

When the ticket already has criteria, use them word for word. Ask with the question tool: "Approve", "Change" or "Full plan". When the plan hinges on a real choice, add it as a second question in the same call, with your recommendation first. Two questions is the limit. Take a fact from the code, and take a default when the repo's rules settle the choice.

Completion: the user picked "Approve".

4. Set up

With an issue, as tracker.md says: add ## Acceptance criteria at the end of the body when it has none, and keep the text that the person wrote. Post the plan as a comment that starts with ### Plan of #<n>. Assign the issue to the user.

The branch is fix/<n>-<slug> from the trunk. Find the trunk as the worktree skill does. Make the branch in the primary checkout when that checkout is clean and on the trunk. Call the Skill tool with worktree when the checkout is busy, when the plan has a migration, or when the user asks: only then does the work need its own folder and databases.

When the parent spec has a Domain changes section that the trunk does not have yet, write it with DOCS-FORMAT.md, and commit: Record terms and decisions for #<n>.

Completion: the session is on the branch, and the issue has the plan.

5. Build

Pick the approach from the work, and call the Skill tool with its skill. Run nothing that the approach does not need.

The workApproach
Behavior with a clear test seamtdd
A defect whose cause is unknowndiagnose
Structure only, and the behavior stays the samerefactor
Look, copy or config, with no seam that can test itBuild it directly, then run it and look

A ticket can mix them, such as tdd for the logic and a direct build for the layout.

Work in this session. Delegate to one implementer subagent only when the work is large: over about 800 changed lines, or reads that would flood this context. Give it the shell rules below, the plan word for word, and the approach. A question from it comes to the user, and the answer goes back to the same subagent with SendMessage.

When the work shows a hard stop that the plan does not name, stop, and ask: "Stay light", with a reason, or "Full plan".

For a plan with phases, run this step and step 6 for each phase. After a phase of a multi-file migration or refactor, run verify in a new subagent on the diff of that phase. Other phases wait for the one cold read at /ship.

Completion: each criterion has its code.

6. Check and land

Run the typecheck, the lint, and the tests that the diff adds or touches. Leave the full suite to /ship. A failure is yours to fix now.

Squash into one commit, with one -m flag per paragraph:

  • <title> (#<n>)
  • one line for each criterion: <criterion> · <test name, or what you saw>
  • Hard stops: <the verdict>, in the last paragraph, so that /ship can read it as a trailer. Any Co-Authored-By line goes right below it.

Then write your report, and stop. The user asks for changes in the chat: make each one, run this step again, and squash it into the same commit. Run ui-check only when the user asks.

Completion: the checks pass, the branch has one commit, and the report is in the chat.

Shell rules

These apply in a worktree, and in each subagent prompt:

Run one plain shell command per call, from the worktree root, on paths in the worktree. For a subfolder, use a flag such as pnpm --dir or uv --directory, because a cd stays in effect for the rest of the session. Change a file with the Edit tool, and make a new file with the Write tool. Quote each argument that holds *, [ or ]. The guard refuses only some other forms, such as a && chain, $(...) or a heredoc, so a command that ran is no proof that its form is safe. Print no value from an env file, because a line such as PROD_DATABASE_URL holds a password. To see which variables a file sets, list the names only: grep -oE '^[A-Z_]+=' <file>. Run each test command in the foreground, with a timeout of up to 10 minutes. Only a run that needs longer goes in the background with its output in a file: wait for its exit notification, and read the result before you report. Ask before a host install, and stop an install that shows no progress for 5 minutes.

Report

Each criterion with its test, or what you saw · the commit · the checks that ran · up to three steps for the user to check · next step: /ship <n>.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.