04-builder

04-builder is a skill for Claude Code, Codex from IncommensurableHubris/fullstack-director. It costs 256 tokens per session (4,360 once invoked), scanned A, original, MIT.

A sequential implementation workflow that turns feature specifications and verification contracts into working code and tests. TDD, or test-driven development, means writing a failing test before implementing the behavior that should make it pass.

In plain words
What is it for?
Use it to build a sprint's vertical slice—a small end-to-end piece of functionality—under src/, write tests, track requirement coverage, and prepare evidence for review.
Why use it?
It gives the builder explicit requirements and tests, then creates a handoff that an independent reviewer can inspect without relying on the build conversation.

Skill for Claude CodeCodex

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/incommensurablehubris/fullstack-director/04-builder
Any agent
npx skills add IncommensurableHubris/fullstack-director --skill 04-builder
Clone the repo
git clone --depth 1 https://github.com/IncommensurableHubris/fullstack-director

Made for: Claude Code, Codex.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for 04-builder

README.md
[![agentmods](https://agentmods.dev/badge/skills/incommensurablehubris/fullstack-director/04-builder.svg)](https://agentmods.dev/skills/incommensurablehubris/fullstack-director/04-builder)
Your own site
<a href="https://agentmods.dev/skills/incommensurablehubris/fullstack-director/04-builder"><img src="https://agentmods.dev/badge/skills/incommensurablehubris/fullstack-director/04-builder.svg" alt="Measured on agentmods" height="20"></a>
Per session 256 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 4,360 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00256 $0.04360
Opus 5 $0.00128 $0.02180
Sonnet 5 $0.00051 $0.00872
Haiku 4.5 $0.00026 $0.00436

Measured 5d ago against content hash c5c26398847c, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-05, from the pricing page.

Security

Grade A, and why

04-builder scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.

The scan reads SKILL.md. This mod also ships 8 executable files (evals/check_build.py, evals/fixtures/cases/fix-pass/src/digest.js, evals/fixtures/cases/fix-pass/test/digest.test.js, …), listed below but not scanned — reading those needs a real analyzer, not pattern matching.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.agents/skills/04-builder/SKILL.md · 166 lines

How it starts

The opening of the file, as written. The whole thing — 166 lines — stays where its author put it; the contents beside it link to each section on GitHub.

04 · Builder — build

One mode. 04-builder sprint N turns the sprint's realization layer into a working vertical slice + tests and emits the build-handoff — the sole seed for the context-isolated reviewer (05). Unlike 02/03 (two modes, a Reconcile gate), 04 is one sequential pass, no gate, no subagents: the isolation that matters is 05's, downstream. Your graded value is not "working code" (a strong builder writes that too) — it is the cold-reviewable, honest handoff + non-tautological tests that let a fresh reviewer verify without inheriting your build bias.

Operating principle — build from realizations, hand off for isolated review

  • Build (the funnel). Read 03's per-feature specs + Verification Contracts, 02's design contract, the slice's "Done When", system.md/ADRs as the ambient contract. Implement src/** sequentially, TDD RED-first for testable behaviors, happy-path (hardening is 05). Never re-interpret the raw spine.
  • Hand off (evidence, not narrative). Emit _artifacts/exports/build-handoff-sprint-NN.md: a frozen diff baseline + the per-behavior evidence state + a REQ coverage map + attestations, so 05 can reconstruct the diff, re-run the oracles, and check coverage with zero access to this session.
  • Non-amender, honest escalator. 04 appends no amendment-log.json row (that stays 00/02/03/08). It never silently patches code or spec to fake a pass — build-time drift is surfaced and recorded in the handoff.

The flow — one sequential pass (craft lives in the references; load each as its step begins)

  1. FIX-PASS CHECK — if a FIX REQUIRED docs/quality/qa-report-sprint-NN.md exists for this sprint, you are the maker-half of the build↔review loop (05 is the checker): read its Findings + the committed reviewer RED tests as an added funnel input and run the fix pass (references/build-discipline.md) — drive each failing behavior to green (make a reviewer RED test pass, never editing it — the anti-circular rule; full TDD-for-bugs for a prose finding), then re-emit the handoff (fixed rows EXECUTED + a deviations note + fresh final_commit/ spec_slice_hash). Scope the pass from the findings ∪ a sweep of every VC row in the in-scope feature specs (each PRESENT, or a disposition row) — /03 runs docs-parallel, so a contract can land mid-loop and no input carries it to you (2026-07-28, FB-009). Leave any REQ named in convergence_fired alone — it survived ≥2 rounds and belongs to /03//00, not to another widen (FB-010). If no such report exists, this is inert — proceed to the normal build. On a fix pass, record fix_round: N + repeat_findings: in the re-emitted handoff — you read the report, 05 may not, so this history is the convergence guard's only input (FB-010). PATCH-FUNNEL CHECK — if the dispatch is a certified patch (a docs/planning/patches/patch-NNN.md record with an open ## Patches ledger row), the funnel narrows: inputs = the patch record + the existing realizations its owning REQs already have (no new specs, no sprint file); discipline = TDD-for-bugs (reproducing RED test first — the fix ships its regression test); scope = the record's behaviors, nothing wider. Flip the ledger row planned → in-progress at build start (your one routine ledger write). The handoff is patch-keyed: _artifacts/exports/build-handoff-patch-NNN.md with review_mode: patch, patch: patch-NNN, and spec_slice_path = the patch record (spec_slice_hash computed over it). P3/P4 are HALT conditions: a new dependency (P3) or the certified size budget exceeded (P4) → HALT, mark the ledger row escalated, re-enter the normal chain — never silently widen. (A small fix arriving here without a patch record? Propose patch classification and route to /01-planner patch — requester ≠ authorizer.) If neither applies, proceed normally.
  2. READ FUNNEL — load the realizations as ambient contract: docs/architecture/specs/<feature>.md (+ Verification Contracts) · system.md · docs/architecture/adr/** · docs/design/approved/sprint-NN/manifest.md (if a UI slice) · docs/design/design-system.md (tokens) · docs/planning/sprints/sprint-NN.md (the slice + "Done When" + frozen Gherkin) · .claude/rules/quality-guardrails.md (if present). Direct reads — no Explore subagent (build stays sequential).
  3. CAPTURE baseline_commitgit rev-parse HEAD before any edit, into the handoff frontmatter. It anchors the reviewer's exact diff. (Fresh project with no commit? Make the initial commit first, then capture.)
  4. REVIEW VERIFICATION CONTRACTS — for each VC row, confirm it is buildable in this environment. Unbuildable → INFERRED with a cited reason (e.g. a browser row with no runtime) — honest scoping, not silent dropping. This is the builder↔reviewer negotiation.
  5. IMPLEMENTATION ORDER — consume the spec/slice order (foundation → domain → API → UI → wiring). Do not re-derive the slice.
  6. BUILD (sequential, per behavior)references/build-discipline.md:
    • TDD RED-first, scoped to unit/api-contract rows: write the test, observe it fail (record it — the anti-tautology capture), minimal code to green, refactor. static-conformance/browser rows use their own oracle.
    • eval-suite rows (Profile: agent-system): eval-first RED — the RED is a failing eval case observed before the fix (the harness over the in-spine dataset is the oracle); and the grader must bite — a degenerate output must fail the grader before the row counts EXECUTED (the bite rule, shared/agentic-profile.md). The handoff's eval-suite row carries both the RED-note and the grader-bites line. Writing a new judge for a row is evals-operations (shared/agentic-profile.md §eval-suite) — its validation record must exist before the row counts EXECUTED.
    • Anti-tautology gate before marking a row EXECUTED: RED evidence or the introduce-a-bug answer. Reject assertion-free / assert-the-mock / oracle-copied-from-impl tests.
    • Reuse-first (read the files a task touches; no duplicate modules) · hardened dependency safety (verify existence + reputation before install; pin + lockfile + audit; a new dep beyond the envelope is a HALT) · outcomes, not ceremony (readable · SRP · no dead code; no SOLID-acronym audit — style is 05/08's seat). Small, atomic conventional commits.
  7. TRACEABILITY — every VC row → test:loc; every in-scope REQ → FULL/PARTIAL/NONE; every DM-ID (if a manifest) → file:line.
  8. HANDOFF — write the evidence-bearing build-handoff (references/build-handoff.md · templates/build-handoff.md): frontmatter (baseline_commit/final_commit/spec_slice_path/spec_slice_files/spec_slice_hash/ review_mode: sprint — 04 emits sprint|patch only; full is 05's all-sprints lane, FB-006; on a fix pass also fix_round: N + repeat_findings: [REQ ids the report raised in an earlier round too] — 05 cannot read the prior report, so this history is the convergence guard's only input, FB-010) + File List + VC carry-forward (evidence state + repro command + RED-note + oracle hash + every non-EXECUTED reason/Unknown)
    • the REQ coverage map + attestations. Compute spec_slice_hash mechanically (the one-liner in references/build-handoff.md) so 05 can recompute + compare — a mismatch is 05's BLOCK. Capture final_commit last.

Read the full file on GitHub · 166 lines

Files

What ships with it

45 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 5d ago First seen · 166 lines · 256 tokens per session scan A c5c26398847c

Subscribe to this mod's changes

04-builder is a skill published in the GitHub repository IncommensurableHubris/fullstack-director (5 stars, last pushed 25d ago), licensed MIT. It adds 256 tokens to every session and 4,360 once invoked, about $0.0013 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other skills, from other repositories