/implement

name: implement
description: Implement changes with a linear plan-build-verify workflow and measurable quality gates.

<!-- ../../assets/images/implement-arch.svg diagrams the lightweight and detailed paths, owner fan-out, joins and gates. Documentation only, never loaded — update it in the same commit as any change to step order, gate placement, or agent fan-out. -->

> Before asking, read [User Questions](../../shared/codex-user-questions.md).

Implement

See the [fixed recurrence and root-cause policy](../../shared/native-skill-contract.md#recurrence-and-root-cause-policy) and [reasoning-progress escalation policy](../../shared/native-skill-contract.md#reasoning-progress-escalation) for repeated-obstacle handling; record and validate `reasoning-progress.json` before another cycle after escalation trigger.

Run linear implementation with strict gates.

Keep setup and review artifacts subordinate to the user's acceptance condition under the primary-goal rules in Reasoning-Progress Escalation. Once the fix is established, write the minimal implementation and regression before another reviewer-setup cycle. One failed bounded setup repair stops that auxiliary route; continue safe authorized implementation and retain the coverage limit without claiming clean completion.

An ordinary bounded plan or implementation challenge is one independent pass. Use `challenge-resolve` and `../../shared/adversarial-loop.md` for an explicitly requested iterative loop; this workflow still owns its normal completion gates.

A request to repeat review and fix until a reviewer reports no findings or no security findings, or to resolve all findings; a second independent review round about to be dispatched for the same change; or a skill workflow repeating its review then fix cycle counts as an explicitly invoked loop from that moment: before the next reviewer dispatch, start ledger; follow `../../shared/adversarial-loop.md` (helper output, stop/plateau rules, user's round budget).

Input Schema

{
  "goal": "required implementation objective",
  "mode": "feature|fix|refactor|config|spike",
  "constraints": [
    "optional constraints"
  ],
  "execution": "optional auto|serial|parallel-read|parallel-write; default auto",
  "done_when": "required acceptance statement"
}

Lightweight Local Work

<!-- policy-sibling: shared/native-skill-contract.md §Proportional Execution, skills/manage/SKILL.md §Lightweight Local Work and §Execution Route Selection; intentional divergence: implementation may use native disjoint writers, while management mutations remain parent-serial. -->

> Read and apply [Proportional Execution](../../shared/native-skill-contract.md#proportional-execution) before choosing the direct or detailed path.

Use a parent-owned implementation path for a small, understood, reversible local change with bounded impact and existing task authorization. A code fix plus its regression test remains one change. Routing examples are optional; tests and documentation for one change do not create separate domains.

1. Inspect the affected flow and applicable instructions; define a concise plan with core acceptance and deferred scope. Investigate an unknown cause before editing. Perform a brief parent self-challenge (not independent); accept the challenged plan under existing authority. Assign execution to the parent for indivisible work or useful bounded disjoint owners. 2. Make the authorized change with the smallest coherent core. Consolidate the actual result and handoffs; reconcile conflicts and integrate before result review. Obtain one bounded independent challenge of the changed behavior and its relevant source; retain its findings and actual independence limits. 3. Resolve feasible in-scope findings, run the relevant acceptance check and configured lint/format checks, and inspect the diff. Before reporting a missing declared dependency as blocked verification, apply [Project Dependency Preparation](../../shared/native-skill-contract.md#project-dependency-preparation) and rerun the affected check; this recovery does not require switching to the detailed workflow. After acceptance and configured checks pass, no critical findings remain, and shared confidence policy passes, run the [Local Git approval grant](../../shared/native-skill-contract.md#local-git-approval-grant) check and apply its completion-commit rules; otherwise record `commit skipped`. A user instruction for no commit, leave unstaged or to wait takes precedence. 4. Report the outcome, verification, and material limits concisely. Do not create unnecessary standalone execution-plan artifacts, specialist packs, confidence worksheets, five-entry gate bundles, or final-handoff/result artifacts solely for this path. Reuse still-valid checks for unchanged source and environment; never claim independent review of a later change the reviewer did not assess.

This path takes precedence over the detailed workflow and role examples below. Select the detailed workflow for medium/large or unexpectedly nontrivial work under Proportional Execution, an explicitly requested structured report/review, or specialist coverage beyond the bounded challenge. Actual runtime permissions, protected-state decisions, and project-required checks still apply. Missing portable parallel controls select parent-serial work for that route; they do not prohibit an authorized native bounded delegation route or require another permission request for already-authorized local edits. Missing independence stays an explicit limit, never a fabricated review.

Execution Route Selection

Before preparing or dispatching any child, resolve invocation `--execution=<mode>` → `CODEX_RIG_EXECUTION` → `auto` and apply the loaded Proportional Execution selector. Honor `serial`; `parallel-read` chooses only the strict portable section; reject unsupported `parallel-write`. Absent, explicit, or environment `auto` chooses suitable observed native bounded work or parent-seria