← Back to docs

Reference

Skill catalog

The 19 portable Oh My QEMU skills, grouped by the work they support.

Every skill is independently installable and includes the same compact audit, workspace, build, and Git-exclude rules. Skills also ship one-level resources: references/ for detailed procedures, assets/ for copy-ready templates, and scripts/ for portable checks. Bash helpers are bash 3.2 function libraries; Python helpers use only the standard library. Install only the skills a project needs.

Choosing a skill

Pick the skill that matches the object being changed or the evidence being produced:

  • new register contract: qemu-register-extraction;
  • device behavior: qemu-peripheral-modeling;
  • board or SoC topology: qemu-board-modeling;
  • instruction decode or host code generation: qemu-tcg-frontend or qemu-tcg-backend;
  • build, image, or boot: qemu-build, qemu-kernel-build, qemu-uboot-build, qemu-image, qemu-boot-run, qemu-linux-boot;
  • evidence and regression: qemu-qtest and qemu-model-verification;
  • debugging or documentation: qemu-debug or qemu-rst-documentation;
  • coordination, provenance, round management, patch-series, and handoff: qemu-workflow, with qemu-agent-feedback for approved workflow feedback;
  • end-to-end modeling: start with qemu-modeler, which routes the domain skills and runs the implement -> build/test -> debug -> review loop; qemu-model-review is the independent reviewer it invokes each round.

A first modeling task normally runs in this order: qemu-register-extraction -> qemu-peripheral-modeling -> qemu-build -> qemu-qtest -> qemu-model-verification, then qemu-rst-documentation if documentation is requested. The orchestrator runs this as a reviewed loop.

If the agent does not list a needed skill, record the gap in the task audit and reinstall instead of improvising the skill’s rules.

Orchestration and review

  • qemu-modeler: top-level entry for a complete modeling task; phase protocol, round loop, task-state and review-request templates, and loop-state.sh.
  • qemu-model-review: independent review of one round; checklist, report template, and deterministic review-checks.sh; returns BLOCKER/MAJOR/MINOR/NOTE findings and a PASS/FAIL/INCONCLUSIVE verdict.

Coordination and feedback

  • qemu-workflow: planning, evidence, review, handoff, and a self-contained hard-gate library (scripts/gates.sh) for workspace, round, evidence, Git, build, marker, and handoff checks.
  • qemu-agent-feedback: sanitize one reusable workflow improvement and file it only after explicit approval; ships a redaction checker and an issue template and does not depend on repository scripts.

Modeling and TCG

  • qemu-register-extraction: extract a cited register and behavior contract plus a Register Semantics IR (RS-IR) text (Trigger/Condition/Transition rules, DEF-style vocabulary, parser/validator with canonical dump, translation worklist, and unimplemented placeholders).
  • qemu-peripheral-modeling: research and verify local peripheral or accelerator models, including the RS-IR to RegisterInfo/manual-MMIO translation, trace/log/error mapping, and field lessons.
  • qemu-board-modeling: work with boards, SoCs, memory maps, boot paths, and IRQ topology, including the unimplemented-device placeholder lifecycle and board field lessons.
  • qemu-tcg-frontend: inspect guest instruction decode and translation behavior.
  • qemu-tcg-backend: inspect host backend operations, constraints, and emission.

Build, image, and boot

  • qemu-build: configure and build QEMU in named builds/build-<target>/ directories.
  • qemu-kernel-build: build Linux kernel artifacts for QEMU testing.
  • qemu-uboot-build: build U-Boot, SPL/TPL, FIT, and firmware-chain artifacts.
  • qemu-image: inspect or package raw, qcow2, disk, flash, pflash, and firmware images.
  • qemu-boot-run: construct reproducible QEMU runs and classify their results.
  • qemu-linux-boot: boot Linux by direct kernel loading or through firmware.

Verification and support

  • qemu-qtest: design, run, and debug qtests.
  • qemu-debug: use host debuggers, the guest gdbstub, traces, or replay.
  • qemu-model-verification: select evidence and report PASS, FAIL, or INCONCLUSIVE.
  • qemu-rst-documentation: work with QEMU reStructuredText and documentation builds; ships a board/machine page template, requirements reference, and RST gate script.

The modeling, TCG, and documentation skills are for local research and verification. They do not authorize preparing or sending generated patches to the QEMU mailing-list recipients selected through MAINTAINERS. Local Git operations and pull requests do not by themselves constitute QEMU upstream contributions.