AI Documentation¶
This directory contains AI-agent-specific operating instructions for Devonshire Digital and its
ventures, built on top of the company governance in docs/governance/.
The root AGENTS.md is the top-level contract; start there, then use
context-loading-order.md to decide what else to read for your task.
Documents¶
| Document | Description |
|---|---|
| repository-context.md | Orientation for an AI agent landing in this repository: what it is, the venture list, where things live. |
| source-of-truth-rules.md | Short, agent-actionable restatement of the authority/precedence order for resolving conflicting instructions. |
| context-loading-order.md | Task-specific reading sequences for governance, venture, Shelfery, Digital Products, and commercialization work. |
| agent-operating-principles.md | General operating principles: read before writing, plan before large changes, cite sources, surface contradictions, ask when blocked, validate before done. |
| agent-role-definitions.md | Defines the operating roles an agent may take on (Repository Maintainer, Product Manager, Software Engineer, etc.) with purpose, inputs, outputs, and approval boundaries for each. |
| task-classification-and-routing.md | The risk model: low/moderate/high risk tasks and what process each requires. |
| planning-and-work-logs.md | How to create and maintain work plans in work/active/ for nontrivial tasks. |
| editing-discipline.md | Targeted edits only; no unrelated rewrites, renaming, or reformatting outside task scope. |
| human-review-and-approval.md | Which work requires human sign-off, how review is requested and recorded, and that AI cannot self-approve high-risk output. |
| evidence-and-citation-requirements.md | Agent-actionable evidence/citation rules and the list of prohibited fabrications. |
| validation-requirements.md | What to check (scripts/check.py, links, front matter, placeholders, duplicate IDs, secrets) before declaring work complete. |
| change-and-handoff-protocol.md | How to hand off work: what changed, what was validated, what remains open, where the work log lives. |
| escalation-and-ambiguity.md | When to stop and ask a human vs. proceed with a documented assumption. |
| prohibited-behaviors.md | Explicit list of behaviors that are never acceptable for an AI agent in this repository. |
| context-efficiency.md | Guidance on not re-reading the entire repository; use targeted reads and the context-loading-order sequences. |
| prompt-patterns.md | Pointer to reusable prompts in prompts/, venture-level prompt directories, and .claude/commands/. |
Related¶
- AGENTS.md — the root, tool-agnostic AI agent contract.
- docs/governance/README.md — the governance rules these AI documents build on.
- docs/company/README.md — company mission, principles, and brand.