policies.md — Template
Summary
Hard constraints an agent must obey. Written as checkable data, not prose, so the agent can validate its output against it. One policies file per channel/agent, or a shared base + channel overrides. See policies.md.
# Policies — <channel/agent name>
version: 0.1
last_updated: 2026-06-05
owner: <you>
## Hard limits (platform)
title_max_chars: 100
description_max_chars: 2200
hashtags_max: 5
slide_count_range: [6, 10]
## Banned words / claims (agent must reject output containing these)
banned_words:
- cure
- guaranteed
- <add health-claim words — critical for peptides>
banned_claims:
- any medical/treatment claim
- any dosage instruction
## Required
must_include:
- link-in-bio CTA
- source citation for any factual claim
disclaimer_required: true # e.g. "educational purposes only"
## Voice / strategy
approved_buzzwords: [ ... ]
hook_style: "<brainrot/curiosity-gap style notes>"
tone: "<from brand guideline>"
reading_level: simple
## Validation checklist (definition of done)
- [ ] title ≤ title_max_chars
- [ ] no banned_words / banned_claims present
- [ ] disclaimer included
- [ ] on-brand voice
- [ ] saved to correct path + tracker updatedKeep it machine-checkable
Numbers as numbers, banned items as a list, rules as a checklist. The agent should run this checklist before marking a task done — not just read it.
Related: skills.template · guardrail.template · Part A1 - Agent Operating Files