# Nils CLI Deliver High Value Refactors > Find and implement high-value test/stability/shared-foundation refactors across crates, then deliver via deliver-feature-pr. Use when this capability is needed. - Skill: `tomevault-io/nils-cli-deliver-high-value-refactors` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/nils-cli-deliver-high-value-refactors` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/nils-cli-deliver-high-value-refactors/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/nils-cli-deliver-high-value-refactors --- # Nils CLI Deliver High Value Refactors ## Contract Prereqs: - Run inside the `nils-cli` git work tree. - Rust toolchain available on `PATH` (`cargo`, `rustfmt`, `clippy`). - `git`, `gh`, `semantic-commit`, and `git-scope` available when delivering the PR end-to-end. - Use this skill together with: - `$deliver-feature-pr` as the canonical delivery policy. - `$create-feature-pr` and `$close-feature-pr` through `$deliver-feature-pr`. Inputs: - Optional scope hints: - target crate(s) to prioritize - constraints (time, risk tolerance, out-of-scope areas) - Optional quality priorities: - `coverage-first`, `stability-first`, `shared-extraction-first` Outputs: - One of two outcomes: - `Implement`: at least one high-value refactor is implemented with tests and validation evidence, then delivered via `$deliver-feature-pr`. - `No Action`: no high-value target found; return concrete recommendations and potential issue list. - Reporting split (strict): - `Implement`: use `$deliver-feature-pr` delivery contract end-to-end (open PR, wait CI green, close PR). - `No Action`: use `.agents/skills/nils-cli-deliver-high-value-refactors/references/NO_ACTION_RESPONSE_TEMPLATE.md`. Exit codes: - `0`: completed workflow (implemented changes or no-action report) - `1`: command/runtime failure while executing workflow - `2`: usage/scope ambiguity that blocks safe execution Failure modes: - No candidate passes the value gate (avoid refactor-for-refactor). - Candidate requires behavior changes that break parity expectations. - Shared extraction crosses crate boundaries with unclear ownership or high regression risk. - Unable to run required validation commands in the current environment. ## Scripts (only entrypoints) - `.agents/skills/nils-cli-deliver-high-value-refactors/scripts/render-refactor-response-template.sh` (`No Action` response only) - `$AGENT_HOME/skills/workflows/pr/feature/deliver-feature-pr/scripts/deliver-feature-pr.sh` (`Implement` delivery only) ## Workflow 1. Build candidate inventory (all crates, evidence-first) - Review each crate for: - missing tests around observable behavior, edge cases, and error paths - flaky or brittle logic (implicit assumptions, weak error handling, unstable output contracts) - duplicated domain-neutral helpers that could move into shared foundations crates: - `crates/nils-common` - `crates/nils-term` - `crates/nils-test-support` - Capture each candidate with concrete evidence (file path + why it matters). 2. Apply the value gate (must pass before any refactor) - A candidate is implementable only if it satisfies at least one: - improves correctness/stability for user-visible behavior - adds meaningful coverage for uncovered critical paths - removes duplicated logic used by 2+ crates via shared foundations extraction - Reject candidates that are style-only, cosmetic-only, or low-impact churn. 3. Decide branch - If one or more candidates pass: - choose smallest high-value slice - implement with behavior parity preserved - add/expand tests first or alongside code changes - If none pass: - do not refactor - produce a no-action recommendations report using the no-action template 4. Implementation rules (when branch is `Implement`) - Prefer characterization tests before moving logic. - Keep crate-local adapters for user-facing messages/exit-code policy when extracting shared helpers. - Extract only domain-neutral primitives into shared foundations crates. - Avoid bundling unrelated cleanup in the same change set. 5. Validation - Run targeted tests for touched crates first. - If scope is broad or cross-crate, run: - `./.agents/skills/nils-cli-verify-required-checks/scripts/nils-cli-verify-required-checks.sh` - Report exact commands and pass/fail status. 6. Delivery (required for implemented changes) - Run branch-intent preflight: - `deliver-feature-pr.sh preflight --base main` - Use `$create-feature-pr` to create branch/commit/open PR from confirmed base branch. - Wait for checks to become fully green: - `deliver-feature-pr.sh wait-ci --pr ` - If checks fail, fix on the same feature branch, push, and re-run `wait-ci` until green. - Close after CI is green: - `deliver-feature-pr.sh close --pr ` - The `Implement` branch is not complete until `$deliver-feature-pr` workflow finishes successfully. 7. Response contract (always required) - `Implement` path: report `$deliver-feature-pr` artifacts: - PR URL - CI status summary - merge commit SHA - final branch state - `No Action` path: use the no-action template with concrete recommendation list and potential issues. - Render helpers: - `./.agents/skills/nils-cli-deliver-high-value-refactors/scripts/render-refactor-response-template.sh --mode no-action` --- > Converted and distributed by [TomeVault](https://tomevault.io/claim/graysurf) — claim your Tome and manage your conversions.