Use `task-agent` for a bounded repository generator, plugin, harness, CLI, monorepo automation, or maintenance utility source change. Skip product behavior, config-only work, docs, release, review, and planning.
Scanned 9/8/2026
Install to Claude Code
npx -y skills add machenjie/rd-skills --skill repository-tooling-change-builder --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Repository Tooling Change Builder?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/machenjie-repository-tooling-change-builder)More formats (shields.io, HTML) on the badges page.
---
name: repository-tooling-change-builder
description: "Use `task-agent` for a bounded repository generator, plugin, harness, CLI, monorepo automation, or maintenance utility source change. Skip product behavior, config-only work, docs, release, review, and planning."
---
# repository-tooling-change-builder
## Role
Support `task-agent` in changing repository tooling while preserving authority, compatibility, deterministic behavior, cleanup, and proof.
## When To Use
- one bounded repository code generator, compiler plugin, linter, formatter, harness, internal CLI, build or monorepo automation, or maintenance utility change
- a direct task whose primary consumers are repository developers, tests, builds, or local automation
## Do Not Use
- backend service, product behavior, or business logic
- build-graph or build-configuration-only work with no repository-tool source change
- documentation-only work
- production mutation, deployment, release, or rollback
- independent review, task decomposition, planning, routing, or control
## Required Inputs
- accepted behavior, non-goals, edit scope, owner, minimum consumer, invocation, and tests
- source/generated authority, inputs, outputs, toolchain, host versions, and bootstrap
- file/process effects, compatibility, cleanup, rollback, and observable outcome signals
## Professional Decision Rules
- Inspect the owner, minimum consumer, tests, adjacent utilities, and reuse before adding structure.
- Bind generated output to authoritative inputs, generator version, destination, drift check, sole editable source, and a non-circular clean-checkout bootstrap.
- Bind plugins to supported host APIs and versions, including diagnostic identity, option schema, fix safety, rejection behavior, and real host integration.
- Prove harness oracle and regression mechanism with positive and negative controls. Keep harness health distinct from changed-product correctness.
- Preserve internal CLI argv, environment, working directory, stdio, exit, cancellation, rerun, and cleanup. Give multi-file output an explicit complete-or-recover contract.
- Declare behavior-affecting files, tools, locale, clock, randomness, network, credentials, executables, and platform facts. Reject ambient workstation or cache state as proof.
- Validate the mechanism through regeneration comparison, version-matched integration, negative controls, or CLI failure and cleanup paths.
- Stay within the accepted owner; do not widen public APIs or refactor unrelated tooling for tests.
## High-Value Gotchas
- Generated output drifts because the editable authority or regeneration check is ambiguous.
- A generator depends circularly on artifacts that only that generator can produce.
- A crash or failed subprocess leaves partial output that later appears current.
- Ambient tools, locale, time, network, credentials, or working directory make the result non-hermetic.
- A plugin passes unit tests but fails against a supported host API or version.
- A harness stays green after its target defect is reintroduced, creating false confidence.
## Execution Checklist
1. Inspect owner, consumer, invocation, tests, generated surfaces, versions, and reuse.
2. Map normal, invalid, interrupted, bootstrap, version-skew, rerun, and forbidden outcomes.
3. Implement the smallest complete source and proving-test change under one batch method.
4. Validate after the latest edit and inspect source plus generated diffs.
5. Report skipped checks, cleanup, compatibility limits, and residual ownership.
## Companion Boundary
- Load `build-tool-professional-usage` when graph edges, generated authority, cache identity, or reproducibility changes.
- Load `filesystem-process-safety` when local file commit, path containment, direct subprocess, timeout, cancellation, or cleanup changes.
- Load `targeted-validation-selection` only after proof strategy exists and exact repository-defined commands or coverage remain unresolved.
## Stop / Escalation Conditions
- Return a planning boundary for multiple owners, shared contracts, or ordered tasks.
- Stop when authority, ownership, bootstrap, compatibility, oracle, partial-write recovery, or validation remains unresolved.
- Stop build-only, product, docs-only, production, release, rollback, review, and control work at its boundary.
- Do not perform or claim independent review.
## Output Contract
- changed files, owner, consumer, reuse, placement, invocation, and compatibility
- authority, bootstrap, plugin, harness, hermeticity, file/process, and failure decisions
- fresh validation, negative controls, skipped checks, and final diff evidence
- cleanup, rollback, proof limits, residual risk, and planning or release boundary
## Targeted References
| Path | Type | Load when | Do not load when | Required by | Required output |
|---|---|---|---|---|---|
| [generator and plugin contracts](references/generator-and-plugin-contracts.md) | targeted | Generated authority, bootstrap, compiler protocol, host API, diagnostics, fixes, or version compatibility changes | No generator or compiler/linter/formatter plugin behavior changes | task-agent | boundary-decision, selected-approach, proof-limit |
| [harness validity contracts](references/harness-validity-contracts.md) | targeted | Test discovery, orchestration, oracle, fixtures, negative controls, benchmark measurement, or harness exit behavior changes | Existing unchanged harness directly proves the accepted behavior | task-agent | decision-record, validation-plan, proof-limit |
| [repository automation contracts](references/repository-automation-contracts.md) | targeted | Internal CLI, hook, monorepo automation, maintenance mutation, subprocess, rerun, or cleanup behavior changes | No repository automation interface or mutation changes | task-agent | decision-record, failure-decision, proof-limit |
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!