Software

Software Engineering Research

Activate the /skill:engineering-research method and run it with this domain payload. Do not restate the research loop here.

Evidence landscape

• Primary sources: official documentation, language/library/framework

specs and RFCs, source repositories, issue trackers, release notes, vendor docs, published benchmarks with a reproducible methodology.

• Standards bodies: IETF RFCs, W3C, ISO/IEC, IEEE, NIST — cite the exact

document + section when one governs.

• Prior art: public codebases (read the actual code before describing

it), architectural decision records, well-cited engineering blogs, peer-reviewed systems literature. For papers and academic prior art, use the /skill:scholarly-research discovery layers (OpenAlex, Semantic Scholar, arXiv).

• Accept with caveats: established trade press and secondary summaries.

Deprioritize undated posts, listicles, and social media without primary links.

Verification criteria

• Claims about code: read the code or the official doc before describing

behavior. A repo exists ≠ a feature exists — verify in the source.

• Version specificity: claims must name the version, commit, or release

they were checked against. Unversioned claims are inferred.

• Benchmarks: a number must trace to a reproducible methodology — harness,

hardware, dataset, date. "Verified faster" is not a claim.

• Security: any security claim must name the exact CVE, advisory, or

standard (CWE, OWASP) and the version range it applies to.

• Paper-vs-code audits: compare claimed methods, defaults, metrics, and

data handling against the actual code; call out missing code, mismatches, ambiguous defaults, and reproduction risks.

Deliverable shape

Follow the method's artifact contract. In the final artifact, end with a Sources section of URLs (repo, doc, RFC, advisory), each annotated with what was checked (e.g. "read §2.3 of the spec" or "verified in commit abc1234"), and mark every claim verified, inferred, or blocked.

Invocation Flags

Accept --deep and --quick flags and pass them through to the engineering-research method. See that skill for flag semantics.

Gap Detection

If research hits an evidence dead-end — no standard, paper, codebase, or dataset addresses the question after exhausting the discovery layers — suggest running /gap-analysis software to formally validate and document the gap. Do not invoke gap-analysis automatically; offer it as a next step and wait for the user to confirm.

Scope and Boundaries

• This skill produces research — it does NOT produce final designs or implementation guidance.
• Research-only, not for final engineering sign-off. Licensed engineers must review and approve any design based on this research.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: adeerkhan
• Source: adeerkhan/vitruvius
• License: MIT

Install and usage instructions live in the source repository linked above.