AGENT SKILL

makefile-generator-cpp

Author or repair GNU Make builds for C++ executables, preserving compiler/linker flags, header dependencies and real project test contracts. Use the optional scaffold for a flat src directory; retain an existing build system unless the task requires changing it.

License 许可
Apache-2.0
最近更新
2026-11-15
Stars
★ 1
Forks
0

01 前言 Front Matter

name
makefile-generator-cpp
license
Apache-2.0
description
Author or repair GNU Make builds for C++ executables, preserving compiler/linker flags, header dependencies and real project test contracts. Use the optional scaffold for a flat src directory; retain an existing build system unless the task requires changing it.

02 资源文件 9

03 SKILL.md 正文

C++ Makefile builds

Apply the repository's mandatory agent tools and controlling instructions before relevant work. Original sources are retained in provenance/. This maintained overlay is unbenchmarked for agent efficacy.

Inspect the project's build commands, source graph and supported toolchain first. Use its actual C++ compiler and standard. Discover tools rather than assuming make, Bash or a compiler is installed. In this collection on Windows x64, ./dev.ps1 python supplies an isolated Python runtime for generation. Compiler qualification uses a separately admitted environment; do not install global packages or replace an existing build system to make a template applicable.

  • Preserve compiler options through compilation and linking. CXXFLAGS can select language/runtime behavior that also requires link support. Keep LDFLAGS before object inputs and LDLIBS after them; use the C++ driver to link C++ objects. Respect project, environment and command-line overrides. For flag, configuration or mixed-language changes, check the variable and configuration contracts.
  • Track included headers with compiler-generated dependency files. Build directories must exist before their recipes need them. Give distinct source files distinct object paths, including equal stems with different extensions.
  • Use the project's real test command. An arbitrary executable's --test option or an echo-only target does not establish that tests ran. Add install, packaging or deployment targets only when the project needs and defines them. Qualify installation in an owned staging directory; uninstall must remove the installed names while preserving unrelated files.
  • Before editing existing rules, inspect included Makefiles, expanded target names and multi-target rules. A line-based search cannot establish that a target is absent. Preserve the existing recipe and failure status; do not append generic targets with the original standard-target helper. Check the real command's behavior, including a failure control where practical.
  • Verify a clean build, an unchanged rebuild, a changed transitive header and a failing compiler/linker option where those boundaries changed. Preserve independent work under parallel make; do not combine cleanup and a build as unordered parallel goals. Rebuild affected outputs after changing flags or toolchain settings; this scaffold does not track flag values automatically.
  • For generated sources, recursive sub-builds or performance changes, apply build reliability. Verify regeneration and failure propagation before measuring speed; an unchanged build is not a clean-build comparison.

For a new, flat src directory of .cpp, .cc and .cxx files with names using letters, digits, dots, underscores or hyphens, an optional scaffold is available:

python <installed-skill-directory>/generate.py --name demo --output Makefile

The destination parent must exist. Existing files are never overwritten; review the Makefile before running it. --check compares an unmodified scaffold without writing. The default C++20 flags are overridable. GNU Make, a GCC-compatible C++ driver and POSIX recipe utilities are required for this scaffold. A project root containing spaces is supported; whitespace in source filenames, nested source layouts, MSVC and shared libraries require project-specific rules.

Generation works independently of a compiler. Run package controls with python -B -m unittest discover -s <installed-skill-directory> -p test_generator.py. Build-dependent controls require GNU Make, GCC C++ and OpenMP on Linux; missing tools are explicit skips unless TASK_REQUIRE_CPP=1 requests qualification. These component controls do not establish model improvement or full upstream package readiness. The original security guide and other language templates are outside this overlay's qualified scope.

04 同仓库更多 Skills

strmt7
strmt7/ai_skills_collection_benchmarked

makefile-generator-project-wrappers

Repair Python, Java and generic GNU Make wrappers that report false success, miss changed inputs or clean dependency environments. Preserve the actual project build system and test runner; use when adapting the collection's original Makefile generator.

★ 1
strmt7
strmt7/ai_skills_collection_benchmarked

makefile-generator-go

Author or repair GNU Make wrappers for Go modules, retaining Go's dependency discovery and build cache. Use when wrapper timestamps hide embedded-resource changes or build failures; preserve the project's existing build system and real test commands.

★ 1
strmt7
strmt7/ai_skills_collection_benchmarked

strategic-compact-resume

Checkpoint and resume long agent work when context pressure or a handoff could lose constraints, verified progress, or ownership of active jobs.

★ 1
strmt7
strmt7/ai_skills_collection_benchmarked

eval-harness-integrity

Design and verify matched coding-agent evaluations when comparing skills, prompts, models, or harnesses using executable acceptance checks and native result evidence.

★ 1
strmt7
strmt7/ai_skills_collection_benchmarked

eval-harness-evidence

Design and verify matched coding-agent evaluations when comparing skills, prompts, models, or harnesses using executable acceptance checks and native result evidence.

★ 1
strmt7
strmt7/ai_skills_collection_benchmarked

e2e-testing-current

Build or repair Playwright end-to-end tests, page objects, readiness checks and CI diagnostics. Use for browser journeys and flaky test investigation; preserve the application's real contracts and existing runner.

★ 1