IEEE Agent Skills Library
Twenty-six executable agent skills covering the IEEE and ISO standards that govern serious software delivery — composable into any project repo via a single git submodule.

Sector
AI & Orchestration
Scale
26 skills, 11 specialised agents, ~20 covered standards
Role
Architect, sole author
Timeframe
2024 – current
Challenge
Most "AI for software engineering" tooling is prompt-engineering folk wisdom dressed up as methodology. There is no shared, executable representation of the IEEE and ISO standards that govern how serious software gets built. Every team reinvents requirements quality checks, V&V planning, and configuration management — usually badly.
Worse: agents asked to follow these standards generally don't, because the standards aren't in their prompt budget and they have never been "trained" on them in any deployable sense. The result is AI work that passes the "looks plausible" test and fails the "would survive an audit" test.
Approach
A modular Claude Code skill library — one skill per IEEE or ISO standard — that any project repo can consume via a single git submodule add. Each skill is structured as Claude Code expects: a SKILL.md frontmatter defining when to trigger, reference materials (worked examples, templates, checklists), and an inter-agent handoff contract so the output of one skill is the structured input of the next.
Skills are composable: a project pulls in only the standards it needs, version-pinned per project to avoid copy-paste drift. The canonical library is genericised — no client identifiers, no project-specific worked examples in the trunk.
The eleven specialised agents wrap the skills into role-specific personas: a requirements agent that owns IEEE 29148 work, a V&V agent that owns IEEE 1012 and 1028, a configuration management agent that owns IEEE 828, and so on. Each agent's contract with its peers is explicit, so a requirements-agent's output is consumable by a design-agent, whose output is consumable by a V&V agent.
Key Decisions
- 1Skills as data, not as prompts — prompt-as-code, not prompt-as-folk-wisdom
- 2One skill per standard — composability over monolithic agents
- 3Inter-agent handoff contracts — output format for skill A is the input format for skill B
- 4Submodule consumption — version-pinned per project, no copy-paste drift
- 5Canonical library is genericised — worked examples, where included, do not contain client identifiers
Outcome
Used internally on every Brumbiesoft project. Available to client engagements where IEEE compliance is a hard requirement. The library is the underlying engine of the FluxPM project-management platform's IV&V discipline.
Stack
Technologies
Markdown · YAML frontmatter (Claude Code skill format) · Git submodules · Inter-skill handoff contracts (structured Markdown sections)
Standards & Methodologies
ISO/IEC/IEEE 12207 · ISO/IEC/IEEE 15288 · ISO/IEC/IEEE 29148 · ISO/IEC/IEEE 42010 · IEEE 1016 · IEEE 1012 · IEEE 1028 · IEEE 730 · IEEE 828 · IEEE 1228 · IEEE 1633 · ISO/IEC/IEEE 16085 · ISO/IEC/IEEE 16326 · ISO/IEC/IEEE 15939 · ISO/IEC 25010 · ISO/IEC 25040 · ISO/IEC/IEEE 29119 (Parts 1–5) · WCAG 2.1
Related Projects
Need to bring AI agents into a regulated software-delivery process without losing your audit position?
Book a discovery call