fencelinelisted

# fenceline You are setting up (or refreshing) a repository so that autonomous coding agents work in it the way the team does — and cannot break the rules that matter. The tool does the deterministic part (hooks, config, templates); you do the judgement part (understanding the product, filling the docs, tuning the safe-list). The principle you are installing: **a rule an agent can forget is not a rule. Enforce in the environment, not in the prompt.** ## Workflow ### 1. Scan first, touch nothing ```bash npx fenceline scan npx fenceline scan --json > /tmp/fenceline-profile.json ``` Read the profile. Note: preset, checks (lint / type-check / test commands), risk zones (payments, auth, push, migrations, native, secrets, jobs…), fragile zones (high-churn dirs with many bug-fix commits), sibling repos, existing agent docs, warnings. If `existingAgentDocs` is non-empty, read those files before anything else. They are the team's current conventions; the goal is to enforce them, not replace them. ### 2. Interview — five questions, no more Ask only what the scan cannot know. Skip any question the scan or the conversation already answered. 1. **Audiences.** Who will read the docs: engineers only, or also analysts / PMs writing tasks for agents? (Decides how much goes into CLAUDE.md.) 2. **Hard bans.** Anything an agent must never touch that the scan didn't flag? (Confirm the detected risk zones; add theirs.) 3. **Siblings.** Are the detected sibling repos really read-only for