Agent Operating Principles¶
These are the general principles an AI agent follows when working in this repository, regardless of task type.
Read before writing¶
Understand the relevant existing documents, their status, and their authority before changing or adding to them. Do not draft new content that duplicates or contradicts something that already exists without first checking.
Plan before large changes¶
For anything beyond a small, obviously-scoped edit, form an explicit plan first — see planning-and-work-logs.md. This is especially true for moderate- and high-risk work per task-classification-and-routing.md.
Prefer minimal, targeted diffs¶
Change what the task requires and nothing else. See editing-discipline.md for the full rule on avoiding unrelated rewrites.
Cite sources for claims¶
Every non-obvious factual claim needs a traceable source. See evidence-and-citation-requirements.md.
Surface contradictions rather than silently resolving them¶
When authoritative sources disagree, say so explicitly rather than quietly picking one. See docs/governance/authority-and-inheritance.md and source-of-truth-rules.md.
Ask when blocked on irreversible or human-only decisions¶
If a task requires an irreversible action, or a decision that only a human role can make (see docs/governance/decision-rights.md), stop and ask rather than proceeding on a guess. See escalation-and-ambiguity.md.
Validate before declaring done¶
Run the applicable validation steps — see validation-requirements.md — before reporting a task complete. "It looks right" is not validation.