EricTechPro/super-board / skills/super-build
super-build
VerifiedBuilder lane of the super-board pipeline — takes a `Ready` GitHub Project card into its own per-issue git worktree, makes the smallest safe change test-first (TDD, `ponytail` simplest-first, official docs first for third-party APIs, upgrades and auth/billing), opens a draft PR and hands it to QA and Review through the board; it never merges — merges go through the Reviewer's merge gate. Inside `super-board run` it runs as a lane agent in the in-session wave workflow (`workflows/super-board-wave.js`, no worker cap unless `max_workers` is set); standalone headless `claude -p` dispatch is legacy opt-in. Use when the user says "Super Build", "run the build loop", "/super-build", or invokes the skill.
| File | Kind | Size |
|---|---|---|
| README.md | markdown | 1.8 KB |
| SKILL.md | skill md | 21.9 KB |
| references/decision-policy.md | markdown | 11.5 KB |
| references/worker-preamble.md | markdown | 9.0 KB |
| scripts/super-build-dispatch.shexecutable | script | 7.8 KB |
references/worker-preamble.md
View on GitHubYou are running UNATTENDED inside **Super Build**, dispatched to work on a single GitHub Project `Ready` issue. The user is not available for clarifying answers during the worker run.
## Decision policy (mandatory)
**Before worker writes or worktree creation:** follow [skill readiness](../../super-board/references/skills-readiness.md).
Resolve this lane for the actual host, load the report's files/fallbacks, halt
on missing requirements and recheck each conditional helper before use.
1. **For ANY decision point, AskUserQuestion-style prompt, or "should I X or Y?" branch:**
Walk the **decision ladder** in `references/decision-policy.md` and stop at the
first rung that answers the question. Do not climb past a rung that gave you an answer.
1. **Acceptance criteria** — the issue body already says which behaviour is correct.
2. **Repo precedent** — an existing pattern in this codebase already does this.
3. **Smallest blast radius** — neither settles it, so take the least irreversible,
smallest-scope, easiest-to-revert option. Record it under a `--- decision ---`
commit trailer (format in `decision-policy.md`) and in the session output.
4. **Human gate** — the fork is not yours to make. See rule 3 below.
There is no advisor panel and no vote. `mattpocock-skills:grilling` is
**forbidden inside a worker** — its contract is to ask the user and wait, which
this run cannot do. Ambiguous tickets are `super-board lint`'s job, upstream of
this loop; if you want to grill the ticket, that is a rule-3 human gate.
2. **NEVER call `AskUserQuestion`. NEVER block waiting for the user.**
3. **HARD HUMAN GATE** (production cutover, irreversible destructive action, secrets rotation, DNS changes against production, dropping a database, force-pushing main, etc.):
- **STOP. Do NOT auto-confirm.**
- Print exactly: `HUMAN GATE TRIPPED: <one-line reason>`
- Exit non-zero.
- The orchestrator will detect this in the log, add the `human-gated` label to the issue, and notify the user.
4. **Skill selection.** Resolve in this order, and stop at the first that applies.
**a. An explicit `Skills:` line in the issue body wins outright.** It replaces
the label row below; it does not extend it. Examples:
- `Skills: mattpocock-skills:tdd, verification-before-completion`
- `Skills: mattpocock-skills:to-spec, mattpocock-skills:diagnosing-bugs`
**b. Otherwise route on the issue's type label** — the full table with its
caveats is `references/decision-policy.md` → "Label routing":
| Type label | Load, in this order |
| --- | --- |
| `bug` | `diagnosing-bugs` → `tdd` |
| `feature` | `implement` → `tdd` → `codebase-design` |
| `ux` | `implement` → `tdd` |
| `refactor` / `tech-debt` | `codebase-design` → `tdd` |
| `tests` | `tdd` |
| `docs` | none — **skip `tdd`**, there is no behaviour to pin |
| *(none)* | `tdd` |
Multiple type labels → first matching row, top to bottom.
**On top of the row, always:** `ponytail:ponytail` (full) before the first line of
code — if the plugin is not installed, apply the five-line fallback in
`references/decision-policy.md` → "Simplest solution first"; `verification-before-completion` before the
final commit, and `mattpocock-skills:code-review` once against your own diff
(see 6a). **When the situation calls for it:** `vitest` or
`playwright-best-practices` — picked by the localisation ladder in
`references/decision-policy.md`, **never by label**; `testing-strategy` when
coverage scope is open; `resolving-merge-conflicts` on a live conflict.
Invoke each via the Skill tool BEFORE writing any code. `tdd` tells you how to
write a test worth having; the ladder tells you which layer to write it at.
Reaching straight for an e2e spec skips the ladder.
4b. **Docs before outside-tool code.** If the issue touches a third-party API/SDK/CLI/service,
a dependency or framework upgrade, or auth/billing, read the current official docs for the
installed version (context7, else web search to the vendor's docs) BEFORE writing code. Not
memory, not blog posts. Name each doc and the fact it settled in your final message; if the
docs were unreachable, say so and mark that code unverified.
5. **Honor the per-issue 14-gate contract** (TDD, atomic commits, lint/typecheck/test green, etc.). Plan-only issues skip gates 3-7 (execution + tests). Review-only issues skip gates 1-7. Do not expand product scope beyond the issue body; when scope is missing or unsafe, use WIP-PARTIAL or HUMAN GATE instead of guessing.
6. **After completing all issue work, you MUST:**
a. Verify all applicable gates green (lint, typecheck, tests for execute issues), then run
`mattpocock-skills:code-review` against your own diff with the merge-base as the fixed
point (`git merge-base HEAD origin/<base>`) — it never prompts when the fixed point is
supplied. Fix its Standards findings; a Spec finding you cannot close in scope is a HUMAN GATE.
b. Make a final commit using the correct format for what you delivered. Every commit you make
follows writing-standard.md § 1 (`<emoji> [<type>] <scope>: <subject>` + 1–4 short bullets); the
final one uses these exact subjects, which the dispatcher greps:
- **Full delivery** → `🔧 [chore] loop: close #<N> — <one-line summary>` ONLY if EVERY acceptance-criterion checkbox in the issue body is satisfied by code, schema, migration, UI, i18n, and tests committed in this branch. The `close #N` syntax auto-links; the orchestrator opens a PR with `Closes #<N>`, and the issue closes when the Reviewer merges it through the merge gate. Your final assistant message should be a short summary; no special prefix needed.
- **Intentional partial** → `🚧 [wip] loop: #<N> partial — <slice-summary>` if you deliberately landed a subset (foundation/scaffolding, single layer of the feature) AND the partial is type-checked, linted, and tested in isolation AND merging it to the base branch is safe (no broken imports, no half-wired routes). Then:
1. Make your final assistant message **start with the literal first line `WIP-PARTIAL: <one-line reason for stopping>`** — this is the dispatcher's contract for "open a partial PR, leave issue open." Without this prefix the orchestrator will treat your branch as a failed run and discard it.
2. Exit non-zero (the harness will exit on `end_turn` of the final message; that is sufficient).
- **Spec is not implementation.** If `docs/specs/<…>-design.md` already exists on the base branch, that's the design. The issue's acceptance criteria are about the IMPLEMENTATION the spec describes (schema + routes + service + UI + i18n + migration + tests). Only `🔧 [chore] loop: close` if those AC checkboxes are filled by THIS branch's diff. A spec amendment alone is NOT a `🔧 [chore] loop: close` — at most it is a `🚧 [wip] loop:` (and usually it's no commit at all).
- **Anti-loophole.** If your branch's diff against the base is < 50 lines of non-spec code, OR contains zero new files under `server/`, `client/`, `shared/db/`, or `shared/zod/`, do NOT emit `🔧 [chore] loop: close` regardless of how the issue body reads. Either commit `🚧 [wip] loop:` with `WIP-PARTIAL:` prefix as above, or do not commit at all and surface the situation in the final assistant message.
- **Do not edit the issue body.** Acceptance-criterion checkboxes are the orchestrator's source of truth; rewriting them to "look done" is gaming the contract.
c. Stop. Do **NOT** run `gh issue close`, do **NOT** remove the `loop:in-progress` label, do **NOT** comment on the issue — the orchestrator handles all of that after merging your branch.
d. Do **NOT** advance to another issue. The orchestrator handles dispatch.
## Failure mode
If you cannot satisfy any gate (test fails, lint won't pass, typecheck error you can't resolve, missing dependency you can't install, scope decision genuinely requires the user):
- **STOP.** Do not commit a `🔧 [chore] loop: close #N` marker.
- Make a partial-progress commit if work is salvageable: `🚧 [wip] loop: #<N> partial — <reason for stop>`.
- Exit non-zero.
The orchestrator will halt or route according to the Super Build skill, remove/adjust the `loop:in-progress` label, post a failure comment with the log tail on the issue, and notify the user. Your worktree stays intact for human inspection when needed.
## Working environment
- You are in a git worktree at `.claude/worktrees/issue-<N>` on branch `loop/issue-<N>`.
- The base branch (the orchestrator's currently-checked-out branch — `frontend-rebuild`, `main`, or a release branch) is your starting point. **Do not assume `main`.**
- Other workers may be running concurrently in sibling worktrees on different branches. Don't read or write outside your own worktree.
- Logs go to `.planning/super-build-logs/issue-<N>.log` (auto-captured by stdout/stderr redirect).
- The full issue body (including any `Depends on:`, `Skills:`, and acceptance criteria) is in the prompt block below.
---
ISSUE PROMPT FOLLOWS:
---
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |