gwt-execute

gwt-execute

Unified Execute-lane entrypoint for implementation work. Treat every owner as a Work Item: a GitHub Issue number plus labels and artifacts. gwt-spec is a design-required tag, not a separate execution kind.

Execution ownership

Once started, keep moving until one of these is true:

• the scoped tasks are complete
• a real product or scope decision blocks the next task
• a merge conflict or reviewer request cannot be resolved with high confidence
• required auth or tooling is unavailable
• a spec-implementation mismatch is discovered that changes the design surface (not just a typo fix) - use gwt-discussion to investigate and discuss

Use the current user's language for task summaries, completion reports, task check updates, and any user-facing text generated while executing the workflow, unless an existing artifact must keep its established language.

gwtd resolution

Before executing any gwtd ... command from this skill, resolve GWT_BIN first: executable GWT_BIN_PATH, then command -v gwtd, then $GWT_PROJECT_ROOT/target/debug/gwtd or ./target/debug/gwtd. Run the command as "$GWT_BIN" ...; if none exists, stop with an actionable gwtd not found error.

Lifecycle

Use the existing build lifecycle JSON operations for every implementation mode. The operation names and state file remain compatibility surfaces.

• build.start with params.spec:<n> when the owner is a gwt-spec tagged Issue or params.task:<description> for standalone work.
• build.phase with params.label:"red"|"green"|"refactor"|"verify"|"pr" at each TDD milestone.
• build.complete only after verification passed and the Ready PR Gate is satisfied for a releaseable slice.
• build.abort with a concrete reason when implementation cannot proceed.

Linked-owner Execution launches also carry an Execution Control Record (SPEC-3248 P8a) written at launch, and Stop stays blocked until the record is settled — even when build.start was never called (plain-Issue fixes included):

• done and verified: JSON operation execution.complete (a successful build.complete settles the record too), or
• blocked by the environment or missing verification: JSON operation execution.blocked with a non-empty params.reason and optional params.missing_verification. Blocked is not done — report the blocker.

execution.blocked is a terminal outcome, not a pause. Never use it while waiting for a temporary question, owner decision, or verification that can still proceed in the current execution. Continue the discussion/workflow instead.

If the same owning session later resolves a genuine terminal blocker, recover without rewriting trusted state or relaunching: register the changed-surface matrix through verify.plan with params.derive:true, run the entire post-block matrix through verify.run, then call execution.reopen with a non-empty params.reason. Reopen requires the exact immutable derived-plan snapshot, fresh passing evidence that started after the block, the same session/owner/worktree fingerprint, and valid integrity hashes. It appends a recovery audit entry and returns the execution to Active; it does not claim completion. Completed executions remain immutable. Another session must use audited execution.adopt only while the record is Active. Blocked and Completed records are terminal and cannot be adopted; another session must use a fresh linked-owner launch.

Completion and Ready PR handoffs consume tool-generated verification evidence (SPEC-3248 P8b): run the verification matrix through JSON operation verify.plan first, then through verify.run with params.commands:[...] so gwtd itself executes and records it. execution.complete, the execution settlement inside build.complete, non-draft pr.create, and pr.ready refuse when the latest record is missing, failing, stale (worktree changed after the run), from another session, or for another owner. Draft PRs stay available mid-work.

When resuming or recovering another session's execution (crash, closed window), take over the record explicitly with JSON operation execution.adopt and a non-empty params.reason — takeovers are audited as an ownership transfer chain. Adopt requires a valid integrity record. An integrity-failed record cannot be repaired in the same execution lifetime without risking audit loss; use a fresh linked-owner launch instead.

Mode detection

1. If the invocation includes #N or an Issue URL, read the Issue with JSON operations issue.view, issue.comments, and issue.linked_prs.
2. If the Issue has a case-insensitive gwt-spec label, use design-gated mode.
3. If the Issue has no gwt-spec label, use direct mode.
4. If no owner Issue exists, use standalone mode.

design-gated mode

Use this for gwt-spec tagged Issues.

1. Read SPEC sections with JSON operations issue.spec.read or issue.spec.section.
2. If plan or tasks is missing or incomplete, stop before production edits and route to gwt-plan-spec.
3. Run the Board active-claim preflight for the owner before choosing a task slice.
4. Select the next incomplete task in dependency order.
5. Execute strict TDD: write the RED test first, confirm it fails for the expected reason, implement the minimum GREEN change, then refactor.
6. Update the owner Issue tasks section through issue.spec.edit as tasks are completed.

direct mode

Use this for non-gwt-spec Issues.

1. Establish the issue facts from JSON operations issue.view, issue.comments, and issue.linked_prs.
2. For bugs, prove the root cause before editing production code. Do not guess-fix.
3. Classify Spec Status: ALIGNED, IMPLEMENTATION-GAP, SPEC-GAP, or SPEC-AMBIGUOUS.
4. If the intended behavior needs design ownership first, route to gwt-discussion or gwt-register-issue; otherwise continue in direct mode.
5. Execute strict TDD with the same Red-Green-Refactor loop.
6. After verification and PR handoff, post the durable closure comment with root cause, changed files, verification evidence, PR link, and remaining work if any.

standalone mode

Use this when the user provides an approved implementation task without an owner Issue.

1. Capture the task description, acceptance criteria, and target files.
2. Identify the narrowest test location from the existing codebase.
3. Execute strict TDD with the same Red-Green-Refactor loop.
4. Do not create Issues or SPECs unless the work proves to need durable design ownership.

Verification and PR gate

Phase 3 delegates to gwt-verify --mode full. Record the selected commands and results in the evidence bundle. UI-affecting work requires a concrete user verification handoff and a User Verification Result.

PR work goes through gwt-manage-pr. Do not create or update a Ready PR until pre-PR verification passes and the User Verification Result is confirmed or n/a.

Legacy aliases

During the transition, $gwt-build-spec SPEC-N and $gwt-fix-issue #N are accepted only as aliases. Continue as $gwt-execute #N and use the mode rules above. Do not split behavior by SPEC versus Issue once the owner number is known.