workflow-policy-layering
Separate workflow instructions from safety, authority, escalation, and validation rules. Use when an assistant, agent, or skill pack is getting muddled because process steps, permissions, forbidden actions, and review gates are mixed together.
Install
npx skills add https://github.com/markoblogo/abvx-agent-skills/tree/main/skills/workflow-policy-layering
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install markoblogo-abvx-agent-skills@llmmart
git clone https://github.com/markoblogo/abvx-agent-skills.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole markoblogo/abvx-agent-skills collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
Workflow Policy Layering
Separate operating workflow from policy and authority.
Goal
Turn a messy assistant spec into clear layers:
- workflow;
- authority and boundaries;
- escalation;
- validation.
Workflow
- Extract the workflow
- write the actual operating sequence without policy noise;
- keep it in business or task language.
- Extract authority separately
- list what the assistant may do directly;
- list what requires confirmation;
- list what is forbidden.
- Extract escalation paths
- define what should be handed to a human, another role, or another system;
- make the next step concrete instead of vague refusal.
- Extract validation
- define what should be checked before a response, mutation, recommendation, or delivery is considered complete;
- keep this as a short checklist or gate list.
- Rewrite the spec
- keep workflow in the main skill body;
- keep detailed policy and validation in
references/or clearly named sections; - ensure the assistant does not confuse “can explain” with “can execute”.
- Run a conflict check
- look for steps that assume authority the policy layer does not grant;
- look for policy text that silently changes the workflow;
- fix contradictions before release.
Output Shape
Return:
- workflow layer;
- authority layer;
- escalation layer;
- validation layer;
- open contradictions, if any.
Design Rules
- Do not hide forbidden actions inside workflow prose.
- Do not let validation sprawl into generic QA philosophy.
- Keep escalation concrete and actionable.
- Treat drafts, summaries, and recommendations as different from real execution authority.
Files (abvx-agent-skills)
-
agents
-
openai.yaml 257 B
Loading…
-
-
SKILL.md 2 KB
--- name: workflow-policy-layering description: Separate workflow instructions from safety, authority, escalation, and validation rules. Use when an assistant, agent, or skill pack is getting muddled because process steps, permissions, forbidden actions, and review gates are mixed together. license: MIT metadata: abvx_status: experimental abvx_origin: original --- # Workflow Policy Layering Separate operating workflow from policy and authority. ## Goal Turn a messy assistant spec into clear layers: - workflow; - authority and boundaries; - escalation; - validation. ## Workflow 1. **Extract the workflow** - write the actual operating sequence without policy noise; - keep it in business or task language. 2. **Extract authority separately** - list what the assistant may do directly; - list what requires confirmation; - list what is forbidden. 3. **Extract escalation paths** - define what should be handed to a human, another role, or another system; - make the next step concrete instead of vague refusal. 4. **Extract validation** - define what should be checked before a response, mutation, recommendation, or delivery is considered complete; - keep this as a short checklist or gate list. 5. **Rewrite the spec** - keep workflow in the main skill body; - keep detailed policy and validation in `references/` or clearly named sections; - ensure the assistant does not confuse “can explain” with “can execute”. 6. **Run a conflict check** - look for steps that assume authority the policy layer does not grant; - look for policy text that silently changes the workflow; - fix contradictions before release. ## Output Shape Return: - workflow layer; - authority layer; - escalation layer; - validation layer; - open contradictions, if any. ## Design Rules - Do not hide forbidden actions inside workflow prose. - Do not let validation sprawl into generic QA philosophy. - Keep escalation concrete and actionable. - Treat drafts, summaries, and recommendations as different from real execution authority. -
SKILL_CARD.md 1.4 KB
Loading…
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.