A learning companion for GRC practitioners, built from the original GRC Engineering Lab Builder.
The companion helps users learn, practise, and internalise GRC engineering thinking from both synthetic labs and their actual day-to-day GRC work. In local adapters, the learner can intentionally point it at review notes, drafts, files, diffs, terminal output, questionnaires, policies, controls, and other work artefacts so the Companion can extract learning opportunities.
It is not an operational GRC decision-maker: it does not approve vendors, certify audit positions, author final production policies, run controls, or operate a user's programme. It helps the practitioner get better at doing those things.
Target URL:
https://grc.engineering/companion/
Legacy project-site URL:
https://grc.engineering/grc-companion/
Because this is hosted on GitHub Pages, the public website stays static. It is an installer and orientation wizard, not the Companion runtime.
grc-companion/
├── brain/ Companion persona, voice, learning method, primitives
├── knowledge/ Structured corpus and case studies for citation grounding
├── profile/ Learner-owned profile schema and examples
├── skills/ Learning skills only
├── commands/ Portable learning-only command contracts
├── demos/ Behavioural examples and terminal transcripts
├── webapp/ Tier 0 product surface notes/source spec
├── adapters/ Tier 1 and Tier 2 distribution targets
├── docs/ GitHub Pages deployable static landing page
└── scripts/ Corpus and adapter build entry points
Tier 0: Static Setup Wizard
The first shipped surface. Static and GitHub Pages-compatible. It helps a learner pick Claude Code, Claude Projects, Cursor, Codex, or the raw files, then generates setup instructions and a system prompt.
Tier 1: Portable Skill Pack
Platform-specific adapters: Claude Code, Codex, Cursor, and Claude Projects.
Tier 2: PAI Pack
Roadmap. Ships only after Tier 0 has active learners and Tier 1 stabilises.
- Repo restructured around
brain,knowledge,skills,webapp, andadapters - Companion brain populated with persona, voice, learning method, and primitives
- Existing lab-builder prompts migrated into
skills/lab-builder - Profile wizard and refresher skills added for learner-owned context
- SOC 2 parsing example migrated into
knowledge/case-studies - Tier 0 static setup wizard scaffolded in
docs
- Skills are invisible by default through
brain/skill-router.md. task-retrospectiveextracts learning from work the learner already completed.cross-domain-translatoruses adjacent-domain patterns to clarify GRC concepts.- The website routes users toward the right package and system prompt instead of pretending to run the Companion.
- Local adapters can learn from real local work artefacts the learner intentionally provides; the public website does not upload those artefacts.
- The actual Companion behaviour runs through generated adapters and canonical repo contracts.
Deferred:
- Real signed-in cross-device profile
- Scheduler/reminder infrastructure
- Tier 1 adapters
- Tier 2 PAI Pack
- Operational GRC skills
Open docs/index.html directly in a browser, or serve the folder with any static server:
python3 -m http.server 8080 --directory docsThen visit:
http://localhost:8080/
The canonical files live in brain/, skills/, commands/, and knowledge/. Generated surfaces are rebuilt with:
scripts/build-knowledge-index
scripts/build-webapp-context
scripts/build-adapters
scripts/verify-structure
scripts/run-evalsGenerated outputs:
knowledge/index.jsondocs/data/companion-context.jsondist/adapters/
Run the terminal transcript:
demos/run-terminal-demoThe demo shows the expected boundary pattern: the Companion can use real local review notes as learning material, but it does not make the vendor approval decision.
The companion refuses operational work. It may teach a user how to think through a vendor review, audit prep pattern, policy design principle, or trust-center architecture, but it should not perform those activities on behalf of the user.