build-executor
GitHub基于已批准的执行契约,指导受控的代码实现。支持TDD、批量实施及隔离分支检查,确保逻辑合规与范围受控,适用于Full/Hotfix/Tweak等模式。
Trigger Scenarios
Install
npx skills add MageByte-Zero/spec-superflow --skill build-executor -g -y
SKILL.md
Frontmatter
{
"name": "build-executor",
"description": "Govern implementation from an approved execution contract. Invoke when execution-contract.md is approved and the user wants disciplined build work, TDD execution, or guarded batch-by-batch implementation."
}
Build Executor
Controls the implementation phase. Uses execution-contract.md as the workflow authority.
Required Inputs
For Full or legacy Hotfix, read execution-contract.md, tasks.md, relevant specs/, and relevant design.md. Quick, direct incident Hotfix, and Tweak require only their receipt, request boundary, changed files, and verification command.
Check workflow mode and receipt first. Tweak → direct edit mode. Quick or a valid direct incident Hotfix → Direct Quick and Hotfix. Full or legacy Hotfix → standard contract-first discipline.
Branch/worktree preflight before ANY implementation edit (mandatory — do not skip):
- Run the isolation check:
This script enforces git isolation: if you are onnpx --yes --package spec-superflow@0.12.1 ssf isolate <change-dir>main/masterit creates a git worktree (preferred) or a new branch, and exits non-zero if it cannot and you have not approved--force. - If
npx --yes --package spec-superflow@0.12.1 ssf isolateexits non-zero: STOP. Do not editmain/masterin place. Ask the user for explicit approval (and re-run withnpx --yes --package spec-superflow@0.12.1 ssf isolate <change-dir> --forceonly after they approve). - If it succeeds, report the chosen branch/worktree and make all implementation edits there.
Core Laws
Law 1: Contract First (Full and legacy Hotfix)
For Full and legacy Hotfix, the execution contract is the approved handoff artifact, not chat history. Direct Quick and incident Hotfix use their valid direct receipt plus bounded verification instead; they must not create or require a contract.
Law 2: TDD Iron Law — Full and legacy Hotfix
RED (write test, see it fail) → GREEN (write minimal code, see it pass) → REFACTOR (clean up, suite stays green).
Quick follows the verification strategy persisted in its receipt: tdd, new-test, or bounded (targeted test, syntax/static check, or other stated evidence). A direct Hotfix must still demonstrate that the original symptom is gone.
Red Flags: ignoring the selected verification strategy, reporting a manual check as if it were automated evidence, or silently expanding a bounded Quick change. Full and legacy Hotfix still require RED → GREEN → REFACTOR.
Law 3: Review Before Drift
Block on: logic defects, spec violations, missing required tests, unintended scope expansion.
Law 4: Rewind on Contract Break — Full and legacy Hotfix
Return to specifying or bridging if: new behavior appears, interfaces change materially, design assumptions fail, artifacts no longer define intended implementation.
For Quick/direct Hotfix, stop instead of creating or rewinding a contract; refresh
workflow recommend with the observed risk, then select full --confirm.
Execution Mode Selection
For Full or legacy Hotfix, generate proposed waves from the approved contract, then use the recommendation as a decision aid rather than silently defaulting a mode:
npx --yes --package spec-superflow@0.12.1 ssf execution recommend <change-dir> \
--wave <wave-id>:<parallel|serial>:<task,...>[:<depends-on,...>] --json
# Show every available mode, the observed facts, and the recommendation to the user.
# The command writes a receipt tied to the artifacts, contract, and waves. After the user chooses, record that explicit confirmation:
npx --yes --package spec-superflow@0.12.1 ssf execution plan <change-dir> \
--mode <selected-mode> --confirm --reason "user-selected execution mode" \
--wave <wave-id>:<parallel|serial>:<task,...>[:<depends-on,...>]
# Add --acknowledge-recommendation when the selection differs from the recommendation.
npx --yes --package spec-superflow@0.12.1 ssf execution show <change-dir> --json
The optional fourth --wave segment names prerequisite wave IDs. execution show --json reports current, plus each wave's depends_on, receipt, blockers, retryable, and eligible status. A wave with retryable: true has a current fail receipt and is eligible only for its focused repair and re-review; its dependents remain blocked until its replacement pass receipt. Report the saved plan revision, selected mode, ordered waves, dependencies, and whether every parallel wave can actually be dispatched concurrently on the current platform. If concurrency is unavailable, state the capability and reason plainly; retain the planned parallel strategy and do not silently execute it as a serial or Batch Inline plan.
The recommendation uses task count, configured execution.inlineThreshold, and declared wave strategy. It never auto-selects: present every available mode and the recommendation to the user. --confirm records any user-selected mode; a choice that differs from the recommendation requires --acknowledge-recommendation so the plan captures an informed risk decision.
| Mode | Criteria |
|---|---|
| SDD | Recommended for parallel waves, multiple waves, or work beyond the inline threshold |
| Inline | Recommended for a single sequential task; always available for a user-confirmed choice |
| Batch Inline | Recommended for a bounded sequential batch; it remains serial and is never presented as parallel |
Do not transition to executing until execution show reports current: true and the phase guard passes. A revised plan must repeat npx --yes --package spec-superflow@0.12.1 ssf execution recommend and use npx --yes --package spec-superflow@0.12.1 ssf execution revise --confirm; it creates a new revision and invalidates receipts from the prior revision.
Batch Inline Execution
Only when the user explicitly confirms batch-inline after seeing the recommendation. Current agent executes directly and serially. TDD Iron Law still applies.
Procedure: announce mode → write failing test → confirm failure → implement → run suite → refactor → lightweight checkpoint (files exist, no placeholders, test passed, no unintended changes) → report.
Boundaries: if any task touches >1 module, involves schema/API/config changes, or has open questions → downgrade to Inline or SDD.
SDD Workflow
For Full/legacy Hotfix by default. Dispatch according to the persisted plan, review each planned wave, and run a final broad review after all waves.
Planned-Wave Loop
- Read the current plan with
npx --yes --package spec-superflow@0.12.1 ssf execution show <change-dir> --json; only waves shown withcurrent: trueandeligible: truemay start. Aretryable: truewave may only be repaired and re-reviewed; do not dispatch its dependents until its replacement receipt ispass. The CLI encodes dependencies in--wave <id>:<strategy>:<tasks>[:<depends-on,...>]and rejects a review receipt for a wave whose prerequisites lack currentpassreceipts. - A
parallelwave may dispatch independent tasks simultaneously only when the platform supports concurrent dispatch. If it does not, disclose the unavailable capability and execute the same wave one task at a time without changing its stored strategy. - A
serialwave dispatches one task at a time in listed order. - After every wave, write a non-empty persisted regular-file review report (separate from the implementer's report), then record exactly one receipt that names that review report:
Do not begin a dependent wave until its predecessor receipt isnpx --yes --package spec-superflow@0.12.1 ssf execution review <change-dir> \ --wave <wave-id> --base <sha> --head <sha> --report <review-report-path> --verdict <pass|fail>pass. - Critical/Important findings require a
failreceipt, a focused repair, re-review, then a replacementpassreceipt. Never advance or close with a missing or failed receipt.
Per-Task Loop
- Dispatch implementer: Load the template with
npx --yes --package spec-superflow@0.12.1 ssf runtime asset read skills/build-executor/implementer-prompt.md. Extract task brief withscripts/task-brief PLAN_FILE N. Include: where task fits, brief path, interfaces from prior tasks, report file path. - Handle response: DONE → generate review package + dispatch reviewer. DONE_WITH_CONCERNS → assess. NEEDS_CONTEXT → provide context. BLOCKED → re-dispatch with better model or escalate.
- Review: Load
npx --yes --package spec-superflow@0.12.1 ssf runtime asset read skills/build-executor/task-reviewer-prompt.md. Reviewer returns spec compliance + code quality verdicts with the wave ID, git range, report path, andpass/failreceipt command. - Fix: If Critical or Important issues, write the
failreceipt, dispatch fix subagent, re-review, and write the replacementpassreceipt. - Mark complete: Append to
.superpowers/sdd/progress.md:Task N: complete (commits <base7>..<head7>, review clean)
Model Selection
Use the configured profile that matches the task role. Resolve it before dispatch:
npx --yes --package spec-superflow@0.12.1 ssf runtime config --resolve-model <profile>
| Profile | Role |
|---|---|
mechanical |
Cheap, routine edits |
standard |
Integration and judgment work |
strong |
Architecture, design, and final review |
review |
Review that matches the diff |
For platforms whose dispatch supports a model field, explicitly pass the resolved model value. If the result is configured: false, automatic selection is unavailable: do not invent a provider model and do not bypass the existing requirement to specify model explicitly. Resolution only reads configuration; it does not switch models.
Progress Ledger
Track in .superpowers/sdd/progress.md. Check for existing ledger — completed tasks are done. After each batch: npx --yes --package spec-superflow@0.12.1 ssf state set <change-dir> batches_completed <N>.
Inline Execution Mode
Only after a user-confirmed inline selection is recorded by npx --yes --package spec-superflow@0.12.1 ssf execution plan --confirm; a non-recommended selection also records --acknowledge-recommendation. Executes in the current session and still writes one review receipt per planned wave.
Per-task: extract brief → write failing test → confirm failure → implement → confirm green → checkpoint review (done-when criteria, SHALL/MUST verification) → commit → save a task-level recovery checkpoint when another task remains → append to progress ledger.
After a task is committed and reviewed, when another task remains, save the recovery context with real evidence:
npx --yes --package spec-superflow@0.12.1 ssf checkpoint save <change-dir> \
--task <completed-task-id> --next "<next task>" --completed "<completed work>" \
--verification "<verification report path>" --review "<review report path>" \
--risk "<open risk or None>" --commit-start <base-sha> --commit-end <head-sha>
This augments .superpowers/sdd/progress.md; it does not replace the progress
ledger or add a new core workflow state. Do not claim a checkpoint is current
when npx --yes --package spec-superflow@0.12.1 ssf checkpoint list reports it as stale.
If task hits BLOCKED (3+ fix failures or changes outside declared scope), escalate to SDD.
Tweak Mode
Skip TDD. Apply changes directly. Verify file integrity (exists, non-empty, valid syntax). No batch execution — sequential changes.
Direct Quick and Hotfix
Quick direct execution requires the valid receipt, a bounded diff, the receipt's selected verification strategy, and a persisted test_result: pass; do not create a contract, execution plan, wave review, DP-6, or DP-7. Direct Hotfix follows the same route only for an incident-backed receipt and must run a regression that demonstrates the original symptom is fixed. If scope grows or risk appears, refresh workflow recommend, show the revised risk, and wait for the user to choose Quick or Full before resuming. A legacy Hotfix without a direct receipt remains subject to the contract, DP-3, execution plan, and review receipts.
DP Records
DP-4 is written by npx --yes --package spec-superflow@0.12.1 ssf execution plan; do not write it with raw state set.
DP-5 (debug escalation): npx --yes --package spec-superflow@0.12.1 ssf state set <change-dir> dp_5_result "<resolution>" + timestamp.
Completion Standard
For Full or legacy Hotfix, do not report completion until tests pass, contract obligations are satisfied, review blockers resolved, every planned wave has a current pass receipt, and final review is complete. For Quick/direct Hotfix/Tweak, report completion only after bounded verification and persisted test_result: pass; do not require contract or review receipts.
Exception Handling
- Parse failures: Stop and report exact line/format issue. Route back to
contract-builder. - Missing artifacts: Route back to appropriate upstream skill. Don't guess.
- User interruption: Progress ledger enables recovery. Check ledger on resume.
Version History
- 1970fe7 Current 2026-07-30 20:20


